Mitja Felicijan's posts

Using Address Sanitizer with clang

Thu, 15 Jan 2026 16:13:13 +0200

What -fsanitize=address does

Enables AddressSanitizer (ASan): a compile and link time instrumentation plus runtime library that detects memory errors.
Detects: out-of-bounds accesses (heap, stack, globals), use-after-free, some use-after-return, double/invalid free, and (on some platforms) leaks.
How it works: instrumented memory accesses are checked against a shadow memory; violations produce an error report with a stack trace and abort the program.
Trade-offs: ~2x runtime slowdown (varies), higher memory use, large virtual address space reservation on 64-bit, and requires linking the ASan runtime (not suitable for production builds).
Usage: compile and link with -fsanitize=address (and typically -g and -O0). Runtime behavior can be tuned via ASAN_OPTIONS and symbolization via llvm-symbolizer.

More about ASan on https://clang.llvm.org/docs/AddressSanitizer.html.

An example how to use it

#include <stdlib.h>
#include <stdio.h>
#include <string.h>

int main(void) {
    char *p = malloc(10);
    if (!p) return 1;

    // Out-of-bounds write (heap buffer overflow).
    strcpy(p, "This string is way too long for the buffer");

    // Use-after-free (unreachable if program aborts on previous error).
    free(p);
    p[0] = 'x';

    return 0;
}

Now let's compile with proper flags with clang -O0 -g -fsanitize=address -o main main.c.

If you run the binary it should trigger ASan. You can also specify what to show with ASAN_OPTIONS=detect_leaks=1:verbosity=1:symbolize=1 ./main and you should see something like this.

MemToShadow(shadow): 0x00008fff7000 0x000091ff6dff 0x004091ff6e00 0x02008fff6fff
redzone=16
max_redzone=2048
quarantine_size_mb=256M
thread_local_quarantine_size_kb=1024K
malloc_context_size=30
SHADOW_SCALE: 3
SHADOW_GRANULARITY: 8
SHADOW_OFFSET: 0x00007fff8000
==28825==Installed the sigaction for signal 11
==28825==Installed the sigaction for signal 7
==28825==Installed the sigaction for signal 8
==28825==T0: FakeStack created: 0x7be4d10f7000 -- 0x7be4d1c00000 stack_size_log: 20; mmapped 11300K, noreserve=0
==28825==T0: stack [0x7ffcf551c000,0x7ffcf5d1c000) size 0x800000; local=0x7ffcf5d1a664
==28825==AddressSanitizer Init done
=================================================================
==28825==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x7c04d25e001a at pc 0x5653c583826a bp 0x7ffcf5d1a5f0 sp 0x7ffcf5d19da8
WRITE of size 43 at 0x7c04d25e001a thread T0
    #0 0x5653c5838269 in strcpy (/home/m/Junk/fsanitize/main+0xb7269) (BuildId: 69a0723cc8e27d59eb584f6cc902f6f12915111a)
    #1 0x5653c5894e13 in main /home/m/Junk/fsanitize/main.c:10:5
    #2 0x7fe4d328bbfb in __libc_start_call_main csu/../sysdeps/nptl/libc_start_call_main.h:58:16
    #3 0x7fe4d328bcb4 in __libc_start_main@GLIBC_2.2.5 csu/../csu/libc-start.c:360:3
    #4 0x5653c57ad360 in _start /builddir/glibc-2.41/csu/../sysdeps/x86_64/start.S:115

0x7c04d25e001a is located 0 bytes after 10-byte region [0x7c04d25e0010,0x7c04d25e001a)
allocated by thread T0 here:
    #0 0x5653c5851ca4 in malloc (/home/m/Junk/fsanitize/main+0xd0ca4) (BuildId: 69a0723cc8e27d59eb584f6cc902f6f12915111a)
    #1 0x5653c5894de8 in main /home/m/Junk/fsanitize/main.c:6:15
    #2 0x7fe4d328bbfb in __libc_start_call_main csu/../sysdeps/nptl/libc_start_call_main.h:58:16

SUMMARY: AddressSanitizer: heap-buffer-overflow (/home/m/Junk/fsanitize/main+0xb7269) (BuildId: 69a0723cc8e27d59eb584f6cc902f6f12915111a) in strcpy
Shadow bytes around the buggy address:
  0x7c04d25dfd80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x7c04d25dfe00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x7c04d25dfe80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x7c04d25dff00: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  0x7c04d25dff80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
=>0x7c04d25e0000: fa fa 00[02]fa fa fa fa fa fa fa fa fa fa fa fa
  0x7c04d25e0080: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x7c04d25e0100: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x7c04d25e0180: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x7c04d25e0200: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
  0x7c04d25e0280: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa
Shadow byte legend (one shadow byte represents 8 application bytes):
  Addressable:           00
  Partially addressable: 01 02 03 04 05 06 07
  Heap left redzone:       fa
  Freed heap region:       fd
  Stack left redzone:      f1
  Stack mid redzone:       f2
  Stack right redzone:     f3
  Stack after return:      f5
  Stack use after scope:   f8
  Global redzone:          f9
  Global init order:       f6
  Poisoned by user:        f7
  Container overflow:      fc
  Array cookie:            ac
  Intra object redzone:    bb
  ASan internal:           fe
  Left alloca redzone:     ca
  Right alloca redzone:    cb
==28825==ABORTING

Change default applications on Linux from terminal

Wed, 14 Jan 2026 16:13:13 +0200

Changing default applications is done with command xdg-mime. This is a command line tool for querying information about file type handling and adding descriptions for new file types. Make sure you have this program installed.

Application will use .desktop files to associate applications with specific types.

Location and structure of `.desktop` files

ls /usr/share/applications
ls ~/.local/share/applications

You can add your own .desktop files to ~/.local/share/applications.

An example of .desktop file for Brave browser located in ~/.local/share/applications/brave.desktop.

[Desktop Entry]
Exec=/home/m/Applications/brave
Type=Application
Categories=Applications
Name=Brave Browser

Query current associations

You can query specific types with xdg-mime.

xdg-mime query default text/plain
xdg-mime query default text/html
xdg-mime query default x-scheme-handler/http
xdg-mime query default x-scheme-handler/https
xdg-mime query default inode/directory

Or you can look at the files containing this data.

less ~/.config/mimeapps.list
less /usr/share/applications/mimeapps.list

Set default application

# Set Brave as default browser.
xdg-mime default brave.desktop x-scheme-handler/http
xdg-mime default brave.desktop x-scheme-handler/https

# Set Thunar as default file explorer.
xdg-mime default thunar.desktop inode/directory

# Set Mousepad as default txt editor.
xdg-mime default mousepad.desktop text/plain

Interfacing with C

This simple example lists all registered types. But you can see where this can go. This could be turned into ncurses CLI application that is used for setting and changing default applications.

#include <gio/gio.h>

int main(void) {
    GList *types = g_content_types_get_registered();

    for (GList *l=types; l!=NULL; l=l->next) {
        g_print("%s\n", (char *)l->data);
    }

    g_list_free_full(types, g_free);
    return 0;
}

Compile with clang -o main main.c $(pkg-config --cflags --libs gio-2.0).

Reading material

Integrated GDB Debugging in Vim with TermDebug

Fri, 09 Jan 2026 16:13:13 +0200

Motivation

Over time I have tried making my ~/.vimrc more universal and capable of serving all the different projects, and I always fail. Configuration just balloons, and sometimes I need different tools for different projects, that fight against each other.

Based on that, I realized that per project .vimrc would be much better, and would better suit my development style. This way I can keep my main ~/.vimrc short and to the point and have project specific configuration separate.

Important: GDB is amazing but sometimes you do require something with better UI and better ergonomics. If you need a graphical debugger that uses GDB as a backend give https://github.com/nakst/gf a try. If you use gf2 than this post does not apply.

Main goals and requirements

Easy to launch make and/or run the application I'm working on.
If the compilation fails, I want Quickfix List to be populated with errors.
Start a debugger and break on main.
Start a debugger and break on currently highlighted line in Vim.

A quick demonstration

Main `~/.vimrc`

My ~/.vimrc is very minimal. I only use these four plugins:

https://github.com/tpope/vim-commentary - comment stuff out
https://github.com/ctrlpvim/ctrlp.vim - fuzzy file, buffer finder
https://github.com/dense-analysis/ale - LSP integration
https://github.com/mitjafelicijan/sniper.vim - buffer bookmark manager

At the top of my configuration file I have the following.

set nocompatible exrc secure

exrc - Allows Vim to read local configuration files.
secure - Restricts what local vimrc/exrc files are allowed to do.

exrc is the important one. This allows us to have .vimrc in the directory of our project. And this will then only apply to that specific project while not polluting the main ~/.vimrc file.

Project `~/project/foo/.vimrc`

The project for testing showcasing this will be a simple C project using GNU Make and Clang.

Project structure

~/Projects/cproject
    main.c
    Makefile
    .vimrc

# Makefile
cprogram: main.c
    clang -g -o cprogram main.c

// main.c

#include <stdio.h>
#include <stdlib.h>

typedef struct {
	int q;
	int w;
} Bar;

int main(void) {
	const char *myenv = getenv("MYENV");

	int a = 100;
	int b = 123;
	int c = a + b;

	Bar bar = { .q = 565, .w = 949 };

	printf("> MYENV: %s\n", myenv);
	printf("> c: %d\n", c);
	printf("> bar.q: %d\n", bar.q);

	for (int i=0; i<10; i++) {
		printf("> loop %d\n", i);
	}

	return 0;
}

" .vimrc
let g:_executable = 'cprogram'
let g:_arguments = ''
let g:_envs = { 'MYENV': 'howdy' }
let g:_make = 'make -B'

set makeprg=make
set errorformat=%f:%l:%c:\ %m
packadd termdebug

let g:termdebug_config = {}
let g:termdebug_config['variables_window'] = v:true

nnoremap <leader>x :call LocalRun()<CR>
nnoremap <leader>c :call LocalMake()<CR>
nnoremap <leader>m :call LocalDebugMain()<CR>
nnoremap <leader>l :call LocalDebugLine()<CR>

function! LocalRun() abort
	let envs = join( map(items(g:_envs), { _, kv -> kv[0] . '=' . kv[1] }), ' ')
	execute printf("term env %s ./%s %s", envs, g:_executable, g:_arguments)
endfunction

function! LocalDebugMain() abort
	execute printf('Termdebug %s %s', g:_executable, g:_arguments)

	for [k, v] in items(g:_envs)
		call TermDebugSendCommand(printf('set env %s %s', k, v))
	endfor

	call TermDebugSendCommand('break main')
	call TermDebugSendCommand('run')
endfunction

function! LocalDebugLine() abort
        let cmd = printf("break %s:%d", expand('%'), line('.'))
        execute printf('Termdebug %s %s', g:_executable, g:_arguments)

        for [k, v] in items(g:_envs)
                call TermDebugSendCommand(printf('set env %s %s', k, v))
        endfor

        call TermDebugSendCommand(cmd)
        call TermDebugSendCommand('run')
endfunction

function! LocalMake() abort
	for [k, v] in items(g:_envs)
		execute printf("let $%s = %s", k, string(v))
	endfor

	silent make

	" Filter non valid errors out of quicklist.
	let qfl = getqflist()
	let filtered = filter(copy(qfl), {_, entry -> entry.valid == 1})
	call setqflist(filtered, 'r')

	redraw!
	execute len(filtered) > 0 ? 'copen' : 'cclose'
endfunction

Note: If any errors are found during the compilation mode the Quickfix list will be populated and opened. You could use CtrlP Quickfix with :CtrlPQuickfix instead of :copen for this as well, I just want to stick to what Vim supports natively.

Lets check these keybindings.

Leader+x - runs the binary in the terminal above and providing environment variables and argument
Leader+c - runs make and puts error in Quickfix List then opens with :copen
Leader+m - launches DebugTerm/DBG debugger with all variables and breaks on main
Leader+l - launches DebugTerm/DBG debugger with all variables and breaks on current line

This setup can get even more elaborate. Depending on your needs. This example works really well for projects using C, but anything goes here.

Why use `:term` and `:TermDebug`

It's very easy to yank and paste from internal terminal buffers even if you are not using tmux as multiplexer. Just makes the whole thing much easier. I do however use tmux as well but for compile/debug loop this proved to be a much better experience.

:TermDebug is also a no brainier. The integration of GDB directly in Vim makes adding new breakpoints with :Break just seamless. This goes for all other commands as well. You can read more about other commands with :h Termdebug or on https://vimhelp.org/terminal.txt.html.

Isn't this setup a bit elaborate?

Yes and no. I don't change these files much at all. When I start a new C project I just copy an existing one from a different project to my new one and call it a day. And any significantly complex project will require certain per project specific configurations anyway.

This also makes it shareable with other people without dumping your whole setup on them. There are drawback like with anything else, but I have found this approach to be the lesser of two evils kinda think. Even Vim users can ignore these files by not enabling local configuration files.

Like any other tool, it's on you to make your development workflow as frictionless as possible and no tool is perfect.

Minimal desktop setup with cwm and conky

Sat, 19 Jul 2025 16:13:13 +0200

Configuration for the setup in the picture is located within commit d29a085.

Important files to check:

Other files are not important.

Change Alt key to Win key under Xfce4

Mon, 30 Jun 2025 16:13:13 +0200

To change the modifier key from Alt to the Win (Super/Meta) key for moving and resizing windows in Xfce:

Open Window Manager Tweaks:
- Go to Settings → Window Manager Tweaks (or run xfwm4-tweaks-settings from a terminal).
Go to the “Accessibility” Tab:
- Look for the option "Key used to grab and move windows".
Change the Modifier Key:
- By default, this is set to Alt.
- Change it to Super (or Meta, depending on your keyboard layout - this is the Windows key on most keyboards).

Lua performance in different environments

Fri, 11 Apr 2025 16:13:13 +0200

A quick look into difference in performance when running Lua normally vs Luajit vs Lua embedded in C program.

You can also check code on GitHub @mitjafelicijan/probe/c-luajit.

Do not take this data as a benchmark, it's just a naive quick checkup.

Naive and intentionally slow implementation of Fibonacci sequence

-- Naive recursive implementation to increase the time of computation.
function fibonacci(n)
    if n == 0 then
        return 0
    elseif n == 1 then
        return 1
    else
        return fibonacci(n - 1) + fibonacci(n - 2)
    end
end

local n = 40
local result = fibonacci(n)
print("The " .. n .. "th Fibonacci number is: " .. result)

Embedded Lua primer

#include <lua.h>
#include <lualib.h>
#include <lauxlib.h>

int main(void) {
    lua_State *L = luaL_newstate();
    luaL_openlibs(L);

    if (luaL_loadfile(L, "fibonacci.luac") || lua_pcall(L, 0, 0, 0)) {
        fprintf(stderr, "Error: %s\n", lua_tostring(L, -1));
        return 1;
    }

    lua_close(L);
    return 0;
}

Quick results

Lua	Luajit	Lua embedded in C
6.006s	1.065s	1.065s

Test script

#!/usr/bin/env bash

ITERATIONS=120

# Just Lua interpreter
for i in $(seq 1 $ITERATIONS); do
    echo "> lua run #$i/$ITERATIONS"
    /usr/bin/time -f "%e,%U,%S" lua fibonacci.lua > /dev/null 2>> out.lua.txt
done

# Using Luajit
for i in $(seq 1 $ITERATIONS); do
    echo "> luajit run #$i/$ITERATIONS"
    /usr/bin/time -f "%e,%U,%S" luajit fibonacci.lua > /dev/null 2>> out.luajit.txt
done

# With C and Luajit
for i in $(seq 1 $ITERATIONS); do
    echo "> cluajit run #$i/$ITERATIONS"
    /usr/bin/time -f "%e,%U,%S" ./fibonacci > /dev/null 2>> out.cluajit.txt
done

Embedding game assets within your executable binary

Tue, 08 Apr 2025 16:13:13 +0200

Why?

Normally, assets live outside the binary, but there is a valid reason to embed them. Specially if you do not want to deal with reading and parsing external files. This makes development and distribution much easier.

In this example, I'm using raylib and C, but this can also be done with SDL or other libraries.

Code for these notes is available as an embedding-binary-data.tar.xz tarball, but beware that this only includes the Linux build of raylib so please change to appropriate operating system.

You can also check code on GitHub @mitjafelicijan/probe/c-embedding-data.

Project structure

We are going to keep it clean and simple here. I am using pre-build version of raylib just to simplify compilation. All the code is located in the main.c file, so refer to that.

├── data                            Contains assets
│   ├── armor.png                   Image example
│   └── dejavusans-mono.ttf         Font example
├── libs
│   └── raylib-5.5_linux_amd64      Includes prebuild raylib
├── main.c                          Main file
└── Makefile                        Build instruction

Main game loop

We first need to setup main game loop and get the game running.

#include <stdio.h>
#include "raylib.h"

int main(void) {
    SetConfigFlags(FLAG_WINDOW_RESIZABLE | FLAG_VSYNC_HINT | FLAG_WINDOW_HIGHDPI);
    InitWindow(900, 400, "Embedding assets");
    SetTargetFPS(60);

    while (!WindowShouldClose()) {
        BeginDrawing();
        ClearBackground(BLACK);
        EndDrawing();
    }

    CloseWindow();
    return 0;
}

And then prepare basic Makefile to compile this.

RAYLIB_VER  := 5.5_linux_amd64
CC          ?= cc
CFLAGS      := -Wall -Wextra -Wunused -Wno-unused-parameter -Wswitch-enum -Wpedantic -Wno-bitwise-instead-of-logical
LDFLAGS     := -I./libs -I./libs/raylib-$(RAYLIB_VER)/include ./libs/raylib-$(RAYLIB_VER)/lib/libraylib.a -lm

game:
	$(CC) $(CFLAGS) -o game main.c $(LDFLAGS)

Now we can compile with make game and get the following.

Converting assets to C header files

Here we use xxd which is used to make a hexdump or do the reverse. This tool contains an interesting option -i which allows us to output the file in C include file style. And this is exactly what we need.

Modified Makefile now looks like.

RAYLIB_VER  := 5.5_linux_amd64
CC          ?= cc
CFLAGS      := -Wall -Wextra -Wunused -Wno-unused-parameter -Wswitch-enum -Wpedantic -Wno-bitwise-instead-of-logical
LDFLAGS     := -I./libs -I./libs/raylib-$(RAYLIB_VER)/include ./libs/raylib-$(RAYLIB_VER)/lib/libraylib.a -lm

game: embed
	$(CC) $(CFLAGS) -o game main.c $(LDFLAGS)

embed:
	xxd -i data/armor.png > data/armor.h
	xxd -i data/dejavusans-mono.ttf > data/dejavusans-mono.h

This converted binary data files into C header style files which contain the array of bytes and the size of the array of bytes. This will be useful later with raylib code.

If we execute make embed we will create C header files. But running make game will also call embed as well, so no need to call it separately.

An example of such a file (in our case armor.png) looks like this.

// data/armor.h
unsigned char data_armor_png[] = {
    0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00, 0x00, 0x0d,
    0x49, 0x48, 0x44, 0x52, 0x00, 0x00, 0x01, 0x40, 0x00, 0x00, 0x01, 0x40,
    0x08, 0x06, 0x00, 0x00, 0x00, 0xcd, 0x90, 0xa5, 0xaa, 0x00, 0x00, 0x00,
    0x01, 0x73, 0x52, 0x47, 0x42, 0x00, 0xae, 0xce, 0x1c, 0xe9, 0x00, 0x00,
    0x00, 0x06, 0x62, 0x4b, 0x47, 0x44, 0x00, 0xff, 0x00, 0xff, 0x00, 0xff,
    0xa0, 0xbd, 0xa7, 0x93, 0x00, 0x00, 0x00, 0x09, 0x70, 0x48, 0x59, 0x73,
    0x00, 0x00, 0x17, 0x12, 0x00, 0x00, 0x17, 0x12, 0x01, 0x67, 0x9f, 0xd2,
    0x52, 0x00, 0x00, 0x00, 0x07, 0x74, 0x49, 0x4d, 0x45, 0x07, 0xdc, 0x0c,
    0x1b, 0x07, 0x30, 0x17, 0xb2, 0x1a, 0xee, 0xda, 0x00, 0x00, 0x20, 0x00,
    ...
    0x81, 0x98, 0xa6, 0xb9, 0x2d, 0x37, 0xac, 0x6d, 0x57, 0xb0, 0xed, 0xea,
    0x86, 0xac, 0xa1, 0xa6, 0x6b, 0xf8, 0x54, 0x1f, 0x8e, 0xe3, 0x90, 0xcb,
    0xe6, 0x36, 0x7d, 0x4e, 0x6b, 0xab, 0xcb, 0x78, 0xbd, 0x5e, 0x6c, 0xc7,
    0x01, 0x8f, 0xa7, 0x5e, 0x46, 0x22, 0x37, 0x17, 0x76, 0x13, 0x4d, 0x6c,
    0xab, 0x0c, 0xb0, 0x89, 0x26, 0x9a, 0x68, 0xe2, 0x45, 0xa3, 0x99, 0x2b,
    0x34, 0xd1, 0x44, 0x13, 0xff, 0xb7, 0xf8, 0x27, 0x9b, 0x96, 0x5c, 0x0d,
    0x8b, 0xbb, 0x21, 0xa6, 0x00, 0x00, 0x00, 0x00, 0x49, 0x45, 0x4e, 0x44,
    0xae, 0x42, 0x60, 0x82
};
unsigned int data_armor_png_len = 118324;

Embedding and compiling

Now it's time to include this files in our main.c code and use them. raylib's API fits perfectly with this style of converting binary files.

#include <stdio.h>
#include "raylib.h"

#include "data/armor.h"
#include "data/dejavusans-mono.h"

#define FONT_SIZE 24

int main(void) {
    SetConfigFlags(FLAG_WINDOW_RESIZABLE | FLAG_VSYNC_HINT | FLAG_WINDOW_HIGHDPI);
    InitWindow(900, 400, "Embedding assets");
    SetTargetFPS(60);

    // Load font from memory.
    Font font = LoadFontFromMemory(".ttf", data_dejavusans_mono_ttf, data_dejavusans_mono_ttf_len, FONT_SIZE, NULL, 0);
    SetTextureFilter(font.texture, TEXTURE_FILTER_TRILINEAR); // Font antialising.

    // Load image from memory and create texture from it.
    Image armor = LoadImageFromMemory(".png", data_armor_png, data_armor_png_len);
    Texture2D armor_texture = LoadTextureFromImage(armor);
    UnloadImage(armor);

    while (!WindowShouldClose()) {
        BeginDrawing();
        ClearBackground(BLACK);

        // Draw the armor texture.
        DrawTexture(armor_texture, 20, 20, WHITE);

        // Draw some text on the screen.
        DrawTextEx(font, "Hello embedded assets.", (Vector2){ 400, 20 }, FONT_SIZE, 0, WHITE);
        DrawTextEx(font, "This is example how we can use embedded fonts.", (Vector2){ 400, 50 }, FONT_SIZE - 4, 0, WHITE);

        EndDrawing();
    }

    UnloadTexture(armor_texture);
    CloseWindow();
    return 0;
}

This makes embedding assets quite straightforward. All we need to do is include header files and then provide byte arrays to the appropriate functions. Like I said raylib has a bunch of FromMemory functions that make this quite easy.

This produces a single binary game that includes both assets, so there is no need to also copy over these assets. Mind you, these binaries can potentially get quite big and if that is the problem, you could always compile this into a so library and include that. This way you could create data packs for audio, graphics, etc. and ship that alongside your game binary.

Run make -B game and run the game.

Honorable mention: C23-embed-directive

#embed is a preprocessor directive to include (binary) resources in the build, where a resource is defined as a source of data accessible from the translation environment. This has been introduces with the C23 standard so it's quite new.

Speaking plainly, #embed allows inclusion of binary data in a program executable image, as arrays of unsigned char or other types, without the need for an external script run from a Makefile.

I do like this approach, but some compilers might not support this feature, and this is why I will be sticking with manual approach for now.
One additional drawback is that every time you compile your game, all the assets need to be re-read and converted. I have not tested this heavily, but this could potentially significantly increase build times when you have a lot of assets.
Not using Make could be detrimental to incremental builds. But this would also need to be tested. I do not claim that this is a real problem. Test this yourself.

Read more about Binary resource inclusion.

Below is a quick example how to use this new directive.

#include <stdint.h>
#include <stdio.h>

const uint8_t image_data[] = {
    #embed "image.png"
};

int main(void) {
    printf("Image size: %d bytes\n", sizeof(image_data));
    return 0;
}

Credits

Currated list of Vim ALE linters

Mon, 17 Mar 2025 16:13:13 +0200

Reading material

Linters

Language	Install command
C	`sudo dnf install clang clang-tidy clang-tools-extra clangd clang-analyzer`
Go	`go install golang.org/x/tools/gopls@latest`
Python	`pipx install pyright`
JavaScript	`sudo npm i -g @biomejs/biome`

Male and female body proportion reference images

Wed, 26 Feb 2025 16:13:13 +0200

Use option key as meta in Alacritty under macOS

Thu, 19 Sep 2024 16:13:13 +0200

To use option key under macOS as Meta key use the option option_as_alt and set it to "Both".

This is the example for the newer Toml file format.

# ~/.alacritty.toml
[window]
option_as_alt = "Both"

And this is the older YAML format.

# ~/.alacritty.yml
window:
  option_as_alt: Both

Check more at https://alacritty.org/config-alacritty.html#s20.

Write and read structs to/from files in C

Sat, 22 Jun 2024 16:13:13 +0200

First let's define a shared header file for the struct definition.

// struct.h
typedef struct {
    char name[50];
    int health;
    float damage;
} Character;

Now lets write it to a character.dat file.

// write.c
#include <stdio.h>
#include <string.h>

#include "struct.h"

int main(void) {
    printf("Write struct\n");

    Character ch;

    strcpy(ch.name, "John Doe");
    ch.health = 30;
    ch.damage = 5.9;

    FILE* file = fopen("character.dat", "wb");
    if (file == NULL) {
        perror("Error opening file");
        return 1;
    }

    fwrite(&ch, sizeof(Character), 1, file);
    fclose(file);

    return 0;
}

If we check the contents of the character.dat file it should look like this.

$ xxd character.dat
00000000: 4a6f 686e 2044 6f65 0000 0000 0000 0000  John Doe........
00000010: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000020: 0000 0000 0000 0000 0000 0000 0000 0000  ................
00000030: 0000 0000 1e00 0000 cdcc bc40            ...........@

Reading and serializing back to a struct.

// read.c
#include <stdio.h>

#include "struct.h"

int main(void) {
    printf("Read struct\n");

    Character ch;

    FILE* file = fopen("character.dat", "rb");
    if (file == NULL) {
        perror("Error opening file");
        return 1;
    }

    fread(&ch, sizeof(Character), 1, file);
    fclose(file);

    printf("Name: %s\n", ch.name);
    printf("Health: %d\n", ch.health);
    printf("Damage: %.1f\n", ch.damage);

    return 0;
}

Embedding resources into binary with C

Wed, 19 Jun 2024 16:13:13 +0200

Binary resource inclusion preprocessor has been put into the C23 standard but has not yet been implemented by the compilers.

Until then a workaround with xxd is possible without spending time on rolling out your own.

xxd has an option to export to C header file which makes this much easier. This works for all files be that text files or binary ones such as images, etc.

Convert text.txt into a C header file with xxd -i test.txt > test.h. This creates the following file and uses the filename for variable names.

// test.h
unsigned char test_txt[] = {
  0x54, 0x68, 0x65, 0x20, 0x66, 0x69, 0x72, 0x73, 0x74, 0x20, 0x72, 0x75,
  0x6c, 0x65, 0x20, 0x69, 0x73, 0x20, 0x74, 0x79, 0x70, 0x69, 0x63, 0x61,
  0x6c, 0x20, 0x6f, 0x66, 0x20, 0x71, 0x75, 0x61, 0x6e, 0x74, 0x75, 0x6d,
  ...
  };
unsigned int test_txt_len = 547;

Then use it in C in this manner.

// main.c
#include <stdio.h>
#include "test.h"

int main(void) {
  printf("Testing embedding of files into binary.\n");

  for (unsigned int i = 0; i < test_txt_len; i++) {
    printf("%02x ", test_txt[i]);
  }
  printf("\n\n");

  for (unsigned int i = 0; i < test_txt_len; i++) {
    printf("%c", test_txt[i]);
  }
  printf("\n\n");

  return 0;
}

Sending signals to C programs

Mon, 17 Jun 2024 16:13:13 +0200

For simple and easy IPC to the C program we can use signals. https://www.man7.org/linux/man-pages/man7/signal.7.html

Only two user defined signals are available (SIGUSR1 and SIGUSR2) but you can hijack other ones even though this is really not advisable.

// signal.c
#include <stdio.h>
#include <stdlib.h>
#include <signal.h>
#include <unistd.h>

void handle_signal(int signal) {
    printf("Signal received: %d\n", signal);
}

int main() {
    signal(SIGUSR1, handle_signal); // is equal to 10
    signal(SIGUSR2, handle_signal); // is equal to 12

    while(1) {
        sleep(1);
    }

    return 0;
}

Compile with gcc -o signal signal.c and run the program with ./signal and then from another terminal send the signal to the program with kill -10 $(pidof signal) which should print out Signal received: 10.

Signal	x86/ARM	Alpha/	MIPS	PARISC	Notes
SIGHUP	1	1	1	1
SIGINT	2	2	2	2
SIGQUIT	3	3	3	3
SIGILL	4	4	4	4
SIGTRAP	5	5	5	5
SIGABRT	6	6	6	6
SIGIOT	6	6	6	6
SIGBUS	7	10	10	10
SIGEMT	-	7	7	-
SIGFPE	8	8	8	8
SIGKILL	9	9	9	9
SIGUSR1	10	30	16	16
SIGSEGV	11	11	11	11
SIGUSR2	12	31	17	17
SIGPIPE	13	13	13	13
SIGALRM	14	14	14	14
SIGTERM	15	15	15	15
SIGSTKFLT	16	-	-	7
SIGCHLD	17	20	18	18
SIGCLD	-	-	18	-
SIGCONT	18	19	25	26
SIGSTOP	19	17	23	24
SIGTSTP	20	18	24	25
SIGTTIN	21	21	26	27
SIGTTOU	22	22	27	28
SIGURG	23	16	21	29
SIGXCPU	24	24	30	12
SIGXFSZ	25	25	31	30
SIGVTALRM	26	26	28	20
SIGPROF	27	27	29	21
SIGWINCH	28	28	20	23
SIGIO	29	23	22	22
SIGPOLL					Same as SIGIO
SIGPWR	30	29/-	19	19
SIGINFO	-	29/-	-	-
SIGLOST	-	-/29	-	-
SIGSYS	31	12	12	31
SIGUNUSED	31	-	-	31

Calling assembly functions from C

Mon, 17 Jun 2024 16:12:13 +0200

This is using the portable GNU assembler and TinyCC compiler but GCC or Clang can be used as well.

First lets define a simple function in assembly.

# sum.s
.intel_syntax noprefix

.global sum

.text

sum:
    add rdi, rsi
    mov rax, rdi
    ret

Lets compile this with GNU assembler as sum.s -o sum.o.

Now we need a C program that calls this function.

// main.c
#include <stdio.h>

// We need to define the signature of the function.
int sum(int a, int b);

int main() {
    for(int i=0; i<10; ++i) {
        printf("SUM of 5+%d is %d\n", i, sum(5, i));
    }
    return 0;
}

Now lets compile and link into final program with tcc main.c sum.o -o main.

Use this table to see what each register is being used for.

Register	Conventional use	Low 32-bits	Low 16-bits	Low 8-bits
%rax	Return value, caller-saved	%eax	%ax	%al
%rdi	1st argument, caller-saved	%edi	%di	%dil
%rsi	2nd argument, caller-saved	%esi	%si	%sil
%rdx	3rd argument, caller-saved	%edx	%dx	%dl
%rcx	4th argument, caller-saved	%ecx	%cx	%cl
%r8	5th argument, caller-saved	%r8d	%r8w	%r8b
%r9	6th argument, caller-saved	%r9d	%r9w	%r9b
%r10	Scratch/temporary, caller-saved	%r10d	%r10w	%r10b
%r11	Scratch/temporary, caller-saved	%r11d	%r11w	%r11b
%rsp	Stack pointer, callee-saved	%esp	%sp	%spl
%rbx	Local variable, callee-saved	%ebx	%bx	%bl
%rbp	Local variable, callee-saved	%ebp	%bp	%bpl
%r12	Local variable, callee-saved	%r12d	%r12w	%r12b
%r13	Local variable, callee-saved	%r13d	%r13w	%r13b
%r14	Local variable, callee-saved	%r14d	%r14w	%r14b
%r15	Local variable, callee-saved	%r15d	%r15w	%r15b
%rip	Instruction pointer
%rflags	Status/condition code bits

You can read more about x86-64 Machine-Level Programming.

The abysmal state of GNU/Linux and a case against shared object libraries

Sun, 10 Mar 2024 21:41:52 +0100

Personal critique

This is in part difficult to write since I have been daily driving GNU/Linux for 20 years now, but I think it is necessary to be honest about this. How come GNU/Linux is worse than it was 10 years ago? This may very well be a subjective opinion, GNU/Linux being better in the past, or maybe I am looking at the situation with rose-tinted glasses.

A full disclaimer, this weekend my system that is pretty vanilla just decided to die after an update. And this was not a full version upgrade. Just a normal update that I do many times per week.

Sure, we now have PipeWire and Wayland. We enjoy many modern advances and yet, the practical use for me is worse than it was 10 years ago. Now all of a sudden, I can't rely on the system to be stable like it was. I don't remember the system bricking after an update, or the system becoming laggy after 10 days of uptime. This may be the issue with Fedora, though.

Over the years, I have daily driven many distributions. From Gentoo, Arch, Fedora to Ubuntu. My best memories were always with Debian. Just pure Debian always proved to be the most stable system. I never had issue or system breaking after an update. I can't say the same for Fedora.

From the get-go, I had issues. I have an Nvidia card and even booting presented issues on occasion. This never happened on other distributions, though they had their problems. Updating the system was basically an exercise in gambling. How come an operating system that boasts with the stability is so unstable? And this was not isolated to my main machine. This also happened on my X220 ThinkPad with Fedora on.

Shared dependencies were a mistake! There, I said it! I understand that disk space was limited back then and this was a legitimate constraint. But this has given me more grief than any other thing. I am all in for AppImages or something like that. I don't care if these images are 10x bigger. Disk space now is plenty, and they solve the issue with "libFlac.8.so is missing" and I have version 12 installed. Which comes with unnecessary symlinking, downloading of older versions and hoping that this will resolve the issue.

Now, the biggest apologist of GNU/Linux will never admit this and even saying something is wrong with this is considered a mortal sin. I, however, am not concerned with cultist behaviors. This is bullshit! Things should be better than 10 years ago, not worse. And I don't care how much lipstick you put on this pig. After more than 20 years of using Linux as my main system, I think I have earned a badge that gives me the right to say the truth.

Regardless of all this, I am still a massive fan. I still think GNU/Linux is probably the most unobtrusive operating system, bar none. But the complexity has gotten the best of it. It's bloated and too complicated at this point. Understandably, you can't have a modern operating system that competes with alternatives without sacrificing simplicity. But I still think that there is another way.

One of the best aspects of GNU/Linux must be outstanding package manager support. Nevertheless, they are essentially solving a problem that should have been solved and done with years ago. The number of gymnastics that happen in the background for you to install a software is just mind-boggling. The dependency graphs are insane. And Snaps and Flatpaks tried to solve some of these things, but until a distribution comes out that is completely devoid of shared dependencies, we will still live in this purgatory.

Let's compare these two distributions when it comes to packages sizes and shared object libraries and see how they fair.

Debian

root@debian: cat /etc/os-release

PRETTY_NAME="Debian GNU/Linux 12 (bookworm)"
NAME="Debian GNU/Linux"
VERSION_ID="12"
VERSION="12 (bookworm)"
VERSION_CODENAME=bookworm
ID=debian
HOME_URL="https://www.debian.org/"
SUPPORT_URL="https://www.debian.org/support"
BUG_REPORT_URL="https://bugs.debian.org/"

Top 20 packages by size

dpkg-query -W --showformat='${Installed-Size}\t${Package}\n' | sort -nr | head -n 20 | awk '{size=$1;unit="KB"; if (size >= 1024) {size=size/1024; unit="MB"}; if (size >= 1024) {size=size/1024; unit="GB"}; printf "%.2f %s\t%s\n", size, unit, $2}'

Size	Package
651.87 MB	libwine
389.26 MB	linux-image-6.1.0-18-amd64
345.76 MB	brave-browser
323.74 MB	google-chrome-stable
265.31 MB	llvm-14-dev
225.76 MB	firefox-esr
141.77 MB	fluid-soundfont-gm
114.67 MB	libreoffice-core
113.41 MB	containerd.io
112.77 MB	libnvidia-rtcore
111.92 MB	libllvm15
106.56 MB	ibus-data
104.92 MB	libllvm14
99.20 MB	docker-ce
77.39 MB	docker-buildx-plugin
72.00 MB	libwebkit2gtk-4.1-0
71.93 MB	libwebkit2gtk-4.0-37
70.51 MB	libwebkitgtk-6.0-4
69.25 MB	nvidia-kernel-dkms
66.64 MB	gcc-12

This is more or less a fresh system that is being used daily for work.

Number of packages installed

dpkg -l | grep '^ii' | wc -l

2217

Number of shared object libraries on system (*.so)

Note: Some of them could be symlinks to each other so take this with a grain of salt.

find /lib /lib64 /usr/lib /usr/lib64 -follow -type f -name "*.so.*" | wc -l

5156

Shared objects sorted by size

find /lib /lib64 /usr/lib /usr/lib64 -type f -name "*.so.*" -exec du -h {} + | sort -rh | head -n 20

Size	Package
113M	/usr/lib/x86_64-linux-gnu/libnvidia-rtcore.so.525.147.05
112M	/usr/lib/x86_64-linux-gnu/libLLVM-15.so.1
105M	/usr/lib/x86_64-linux-gnu/libLLVM-14.so.1
70M	/usr/lib/x86_64-linux-gnu/libwebkit2gtk-4.1.so.0.12.8
70M	/usr/lib/x86_64-linux-gnu/libwebkit2gtk-4.0.so.37.67.8
69M	/usr/lib/x86_64-linux-gnu/libwebkitgtk-6.0.so.4.4.8
57M	/usr/lib/llvm-14/lib/libclang-cpp.so.14
45M	/usr/lib/x86_64-linux-gnu/libnode.so.108
43M	/usr/lib/x86_64-linux-gnu/libnvidia-glcore.so.525.147.05
41M	/usr/lib/x86_64-linux-gnu/libnvidia-eglcore.so.525.147.05
31M	/usr/lib/x86_64-linux-gnu/libmozjs-102.so.102.15.1
30M	/usr/lib/x86_64-linux-gnu/libicudata.so.72.1
30M	/usr/lib/x86_64-linux-gnu/libclang-14.so.14.0.6
29M	/usr/lib/x86_64-linux-gnu/nvidia/current/libcuda.so.525.147.05
28M	/usr/lib/x86_64-linux-gnu/libjavascriptcoregtk-6.0.so.1.1.14
28M	/usr/lib/x86_64-linux-gnu/libjavascriptcoregtk-4.1.so.0.4.14
28M	/usr/lib/x86_64-linux-gnu/libjavascriptcoregtk-4.0.so.18.23.14
24M	/usr/lib/x86_64-linux-gnu/libnvidia-glvkspirv.so.525.147.05
23M	/usr/lib/x86_64-linux-gnu/libz3.so.4
22M	/usr/lib/x86_64-linux-gnu/libgs.so.10.00

Fedora

root@fedora: cat /etc/os-release

NAME="Fedora Linux"
VERSION="36 (Workstation Edition)"
ID=fedora
VERSION_ID=36
VERSION_CODENAME=""
PLATFORM_ID="platform:f36"
PRETTY_NAME="Fedora Linux 36 (Workstation Edition)"
ANSI_COLOR="0;38;2;60;110;180"
LOGO=fedora-logo-icon
CPE_NAME="cpe:/o:fedoraproject:fedora:36"
HOME_URL="https://fedoraproject.org/"
DOCUMENTATION_URL="https://docs.fedoraproject.org/en-US/fedora/f36/system-administrators-guide/"
SUPPORT_URL="https://ask.fedoraproject.org/"
BUG_REPORT_URL="https://bugzilla.redhat.com/"
REDHAT_BUGZILLA_PRODUCT="Fedora"
REDHAT_BUGZILLA_PRODUCT_VERSION=36
REDHAT_SUPPORT_PRODUCT="Fedora"
REDHAT_SUPPORT_PRODUCT_VERSION=36
PRIVACY_POLICY_URL="https://fedoraproject.org/wiki/Legal:PrivacyPolicy"
SUPPORT_END=2023-05-16
VARIANT="Workstation Edition"
VARIANT_ID=workstation

Top 20 packages by size

rpm -qa --queryformat '%{SIZE}\t%{NAME}\n' | sort -rn | head -n 20 | awk '{size=$1;unit="B"; if (size >= 1024) {size=size/1024; unit="KB"}; if (size >= 1024) {size=size/1024; unit="MB"}; if (size >= 1024) {size=size/1024; unit="GB"}; printf "%.2f %s\t%s\n", size, unit, $2}'

Size	Package
572.91 MB	google-cloud-sdk
559.76 MB	wine-core
530.12 MB	wine-core
383.76 MB	code
350.98 MB	golang-bin
318.04 MB	brave-browser
302.78 MB	google-chrome-stable
282.24 MB	libreoffice-core
273.43 MB	firefox
253.00 MB	ocaml
223.60 MB	proj-data-us
219.46 MB	zoom
217.81 MB	glibc-all-langpacks
199.31 MB	qemu-user
196.79 MB	edk2-aarch64
194.77 MB	fluid-soundfont-lite-patches
194.25 MB	java-17-openjdk-headless
176.75 MB	java-11-openjdk-headless
156.80 MB	pandoc
154.82 MB	qt5-qtwebengine

What is interesting that the most of these packages are from a a system that are daily in use.

Number of packages installed

dnf list installed | tail -n +3 | wc -l

3484

Number of shared object libraries on system (*.so)

Note: Some of them could be symlinks to each other so take this with a grain of salt.

find /lib /lib64 /usr/lib /usr/lib64 -follow -type f -name "*.so.*" | wc -l

8894

Shared objects sorted by size

find /lib /lib64 /usr/lib /usr/lib64 -type f -name "*.so.*" -exec du -h {} + | sort -rh | head -n 20

Size	Package
128M	/usr/lib64/libQt5WebEngineCore.so.5.15.10
55M	/usr/lib64/llvm13/lib/libclang-cpp.so.13
53M	/usr/lib64/libclang-cpp.so.14
50M	/usr/lib64/libwebkit2gtk-4.0.so.37.57.6
49M	/usr/lib64/libnode.so.93
45M	/usr/lib64/libOpenImageDenoise.so.1.4.3
33M	/usr/lib64/llvm13/lib/libclang.so.13.0.1
30M	/usr/lib64/libclang.so.14.0.5
29M	/usr/lib64/libopenvdb.so.9.0.0
28M	/usr/lib/libicudata.so.69.1
28M	/usr/lib64/libmozjs-78.so.0.0.0
28M	/usr/lib64/libicudata.so.69.1
26M	/usr/lib64/libusd_usd_ms.so.0
26M	/usr/lib64/libgdal.so.30.0.3
22M	/usr/lib64/libjavascriptcoregtk-4.0.so.18.21.6
22M	/usr/lib64/libgs.so.9.56
22M	/usr/lib64/google-cloud-sdk/platform/bundledpythonunix/lib/libpython3.9.so.1.0
17M	/usr/lib64/libgtkd-3.so.0.9.0
17M	/usr/lib64/libgeocoding.so.8.12
14M	/usr/lib64/libmfxhw64.so.1.35

Quick observation

/usr/lib/x86_64-linux-gnu/libjavascriptcoregtk-6.0.so.1.1.14
/usr/lib/x86_64-linux-gnu/libjavascriptcoregtk-4.1.so.0.4.14
/usr/lib/x86_64-linux-gnu/libjavascriptcoregtk-4.0.so.18.23.14

These three packages are probably used by three different applications, and nothing else needs them. I could be mistaken, but in any case they should just be packaged alongside the application that requires them and be done with it.

So, now what?

The main point is that, yes these packages can be big, but if we are honest, what would a couple of additional megabytes that would include shared object libraries actually do? Probably nothing! And it would make maintenance an entirely different game.

There is also this big elephant in the room, the users. They aren't concerned about package dependencies. They don't care if an application is 20 megabytes bigger. Nobody cares! But they certainly do care about borked systems and non-working dependencies and hunting for solutions for why libFlac was not found even though they have it installed (allthough slightly different version).

Operating systems should abstract these complexities away from the user. And I am not saying that the Linux kernel is at fault. I mean the operating system as a whole.

This is why I am a massive proponent of AppImages. Flatpaks and Snaps are OK, but they still have their own package management and the whole dependency problem because they try to optimize for space, and they sacrifice the simplicity.

It would be an interesting exercise to make a prototype distribution that does not rely on shared objects, but has everything packed in AppImages. Probably a foolish endeavor, but maybe worth looking into. I sense this kind of distribution would be highly unusable. Interesting how far we have gotten that this kind of distribution is almost not possible anymore.

The year of the GNU/Linux desktop? I have strong doubts. We are in a worse state than we were. Not only that, but the sheer amount of choice is paralyzing at this point. This is very relatable to The Paradox of Choice. The more options we have, the worse it gets. Wayland competing with X. So many window managers, you just get lost. So many choices. I have no idea if this is even salvageable, or something new must be made.

A quick shoutout to AppImageHub. If possible, I continually try to find applications there and take care of updating them myself. I don't need hand holding or a constant up-to date system. I just want my system to be stable and when some application has gotten some significant new features I can download that myself. It's about the choice and not being forced into this churn that requires constant updating and keeping up with things. At this point, using GNU/Linux is more like a second job, and I was so stoked when this was not a case anymore in the past. This is why I feel like the last 10 years were a regression disguised as progress.

Some interesting talks and videos

Using custom software with Github Actions to deploy a site

Sun, 10 Mar 2024 15:30:11 +0100

By default, GitHub uses Jekyll for their site generator which is fine, but it has some issues and the complexity is not really worth it for me.

You could argue that Jekyll is simple, which it is to some degree, but it can become complicated quite quickly if you start adding your own spin on things.

A while ago I wrote a simple static site generator called "jbmafp - Just Build Me A Fucking Page" which was a protest against Hugo. Hugo is fine but again, if you try doing something that conflicts with the dogma they prescribe you are in trouble.

I also moved this blog from self-hosted virtual machine to just GitHub Pages. I didn't want to bother myself managing that server anymore. And this presented a slight problem because I didn't want to use the default _docs folder GitHub wants you to use, and I also didn't want to upload public folder that gets generated to GitHub.

Thankfully, there is a way to use custom software to generate your site like jbmafp.

To achieve this you need to create a file .github/workflows/deploy.yaml in the root of your repository.

name: Build and Deploy to Pages

on:
  push:
    branches: ["master"]
  workflow_dispatch:

permissions:
  contents: read
  pages: write
  id-token: write

concurrency:
  group: "pages"
  cancel-in-progress: false

jobs:
  deploy:
    environment:
      name: github-pages
      url: ${{ steps.deployment.outputs.page_url }}
    runs-on: ubuntu-latest
    steps:
      - name: Checkout
        uses: actions/checkout@v3
      - name: Setup Pages
        uses: actions/configure-pages@v3
      - name: Run a multi-line script
        run: |
          wget https://github.com/mitjafelicijan/jbmafp/releases/download/v0.1/jbmafp.zip
          unzip jbmafp.zip
          chmod +x jbmafp
          ./jbmafp -b          
      - name: Upload artifact
        uses: actions/upload-pages-artifact@v2
        with:
          path: './public'
      - name: Deploy to GitHub Pages
        id: deployment
        uses: actions/deploy-pages@v2

This is an example for jbmafp. When I execute ./jbmafp -b the program creates public folder and puts all content there. And the directive with: path: './public' tells actions/deploy-pages@v2 to look for public folder instead of docs folder.

This can be used with anything, actually. Hugo, Gatsby, Astro, you name it.

Converting Valgrind callgrinds to SVG format

Wed, 28 Feb 2024 03:23:00 +0100

Lately I have been doing a lot of systems programming and profiling is the name of the game when it comes to developing good software.

I found Valgrind indispensable for profiling and getting callgraphs.

Most of the time there are better alternatives that SVG to drill intro the callgraphs but if you need to put put a callgraph on a webpage or maybe send it to somebody that does not have all of the necessary software to view these things then SVG is the perfect format for this.

Theses are couple of amazing applications to view callgraphs that get exported by Valgrind:

Kcachegrind - Linux
WinCacheGrind - Windows
Profiling Viewer - macOS

This is how Kcachekrind looks with a callgraph loaded in. Not only that, with Kcachegrind you can also explore assembly produced by the compiler and get much more insight into the profile of your application.

After this point you will need couple of things installed on your system. I will show how you do this on Fedora 39.

# Install valgrind which will do the actual profiling.
sudo dnf install valgrind

# Install Kcachegrind for local visualizing.
sudo dnf install kcachegrind

# Install gprof2dot for conversion from .dot to .svg.
pip install gprof2dot

Let's make a simple C program and test out profiling and the whole shebang.

#include <stdio.h>

int main() {
  printf("Oh, hi Mark\n");
  return 0;
}

Then let's compile, convert to dot and then SVG file.

clang hi.c -o c-hi
valgrind --tool=callgrind --dump-instr=yes --collect-jumps=yes ./c-hi
gprof2dot --format=callgrind --output=out.c.dot --colormap=print callgrind.out.546168
cat out.c.dot | dot -Tsvg > out.c.svg

This gives us an SVG file like this.

And this also works on other binaries.

Lets give Zig a go.

const std = @import("std");

pub fn main() !void {
  std.debug.print("Oh, hi Mark!\n", .{});
}

Now repeat the whole compile, convert cycle.

zig build-exe hi.zig --name zig-hi
valgrind --tool=callgrind --dump-instr=yes --collect-jumps=yes ./zig-hi
gprof2dot --format=callgrind --output=out.zig.dot --colormap=print callgrind.out.546168
cat out.zig.dot | dot -Tsvg > out.zig.svg

Now, to be fair Kcachegrind is much nicer for local exploration and digging deep into the callgraphs, but the SVG format can still provide valid information for documentation and things of that nature.

Uninstall Ollama from a Linux box

Fri, 23 Feb 2024 06:35:11 +0200

I have had some issues with Ollama not being up-to-date. If Ollama is installed with a curl command, it adds a systemd service.

sudo systemctl stop ollama
sudo systemctl disable ollama
sudo rm /etc/systemd/system/ollama.service
sudo systemctl daemon-reload

sudo rm /usr/local/bin/ollama

sudo userdel ollama
sudo groupdel ollama

rm -r ~/.ollama
sudo rm -rf /usr/share/ollama

That is about it.

Sane default for Dungeon Crawl Stone Soup Online edition

Wed, 21 Feb 2024 06:35:11 +0200

I mostly play Dungeon Crawl Stone Soup online on Ohio, USA: cbro.berotato.org server and when you start playing you can select the version you want to play. Each instance also has rc file that can customize the way the game behave.

This is my sane defaults config. It zooms in the game without needing to zoom in the browser and it also adds a bit of delays in exploring and it stops at fight.

autofight_stop = 80
explore_auto_rest = true
explore_delay = 20
default_manual_training = true
travel_open_doors = false

tile_cell_pixels = 64
tile_font_crt_size = 32
tile_font_stat_size = 32
tile_font_msg_size = 32
tile_font_tip_size = 32
tile_font_lbl_size = 32
tile_map_pixels = 0
tile_filter_scaling = false

All the possible options are documented in the Dungeon Crawl Stone Soup Options Guide file.

Extract lines from a file with sed

Thu, 15 Feb 2024 10:04:28 +0200

Easy way to extract line ranges (from line 200 to line 210) with sed.

sed -n '200,210p' data/Homo_sapiens.GRCh38.dna.chromosome.18.fa

# then pipe it to a new file with

sed -n '200,210p' data/Homo_sapiens.GRCh38.dna.chromosome.18.fa > new.fa

head or tail could be used to extract from begining of the end of the file.

Personal sane Vim defaults

Tue, 07 Nov 2023 01:04:28 +0200

I have found many "sane" default configs on the net and this is my favorite personal list. This is how my .vimrc file looks like.

Updated version is available on GitHub.

" General sane defaults.
syntax enable
colorscheme sorbet
nnoremap q: <nop>
set nocompatible
set relativenumber
set nohlsearch
set smartcase
set ignorecase
set incsearch
set autoindent
set nowrap
set nobackup
set noswapfile
set autoread
set wildmenu
set encoding=utf8
set backspace=2
set scrolloff=4
set spelllang=en_us

" Status Line enhancements.
set laststatus=2
set statusline=%f%m%=%y\ %{strlen(&fenc)?&fenc:'none'}\ %l:%c\ %L\ %P
hi StatusLine cterm=NONE ctermbg=black ctermfg=brown
hi StatusLineNC cterm=NONE ctermbg=black ctermfg=darkgray

" Commenting blocks of code.
augroup commenting_blocks_of_code
  autocmd!
  autocmd FileType c,cpp,go,scala   let b:comment_leader = '// '
  autocmd FileType sh,ruby,python   let b:comment_leader = '# '
  autocmd FileType conf,fstab       let b:comment_leader = '# '
  autocmd FileType lua              let b:comment_leader = '-- '
  autocmd FileType vim              let b:comment_leader = '" '
augroup END
noremap <silent> ,cc :<C-B>silent <C-E>s/^/<C-R>=escape(b:comment_leader,'\/')<CR>/<CR>:nohlsearch<CR>
noremap <silent> ,cu :<C-B>silent <C-E>s/^\V<C-R>=escape(b:comment_leader,'\/')<CR>//e<CR>:nohlsearch<CR>

" Language specific indentation.
filetype plugin indent on
autocmd Filetype make,go,c,cpp setlocal noexpandtab tabstop=4 shiftwidth=4
autocmd Filetype html,js,css setlocal expandtab tabstop=2 shiftwidth=2

I keep it pretty vanilla so this is about everything I have in the file.

Add lazy loading of images in Jekyll posts

Sun, 05 Nov 2023 09:04:28 +0200

Normally you define images with ![]() in markdown files. But jekyll also provides a way to adding custom attributes to tags with {:attr="value"}.

If you have lots of posts this command will append `` to all images in all your markdown files.

![image-title](/path/to/your/image.jpg)

will become

![image-title](/path/to/your/image.jpg)

Shell line bellow. Go into the folder where your posts are (probably _posts).

find . -type f -name "*.md" -exec sed -i -E 's/(\!\[.*\]\((.*?)\))$/\1/' {} \;

Under the hood this adds loading="lazy" to HTML img nodes.

That is about it.

Using ffmpeg to combine videos side by side

Sat, 04 Nov 2023 09:04:28 +0200

I had a 4 webm videos (each 492x451) that I wanted to combine to be played side by side and I tried iMovie and Kdenlive and failed to do it in an easy way. I needed this for Github readme file so it also needed to be a GIF.

The following is the ffmpeg version of it.

ffmpeg -y \
  -i 01.webm \
  -i 02.webm \
  -i 03.webm \
  -i 04.webm \
  -filter_complex "\
  	[0:v] trim=duration=8, setpts=PTS-STARTPTS, scale=492x451, fps=6 [a0]; \
  	[1:v] trim=duration=8, setpts=PTS-STARTPTS, scale=492x451, fps=6 [a1]; \
  	[2:v] trim=duration=8, setpts=PTS-STARTPTS, scale=492x451, fps=6 [a2]; \
  	[3:v] trim=duration=8, setpts=PTS-STARTPTS, scale=492x451, fps=6 [a3]; \
  	[a0][a1][a2][a3] xstack=inputs=4:layout=0_0|w0_0|w0+w1_0|w0+w1+w2_0, scale=1000:-1 [v]" \
  -map "[v]" \
  -crf 23 \
  -preset veryfast \
  trigraphs.gif

This will produce trigraphs.gif that is also scaled to max 1000px in width (refer to scale=1000:-1).
The important part for 4x1 stack is xstack=inputs=4:layout=0_0|w0_0|w0+w1_0|w0+w1+w2_0.
This will also cap frame rate to 6 (refer to fps=6) since that is enough and this makes playback of GIFs smoother in a browser.

Compile drawterm on Fedora 38

Mon, 25 Sep 2023 09:04:28 +0200

First install two dependencies:

sudo dnf install libX11-devel libXt-devel

Clone the repo and compile it:

git clone git://git.9front.org/plan9front/drawterm
cd drawterm
CONF=unix make

That should produce drawterm binary.

AWS EB PyYAML fix

Mon, 18 Sep 2023 07:27:29 +0200

Recent update of my system completely borked EB CLI on my machine.

I tried installing it with pip install awsebcli --upgrade --user and it failed.

The error was the following.

Collecting PyYAML<6.1,>=5.3.1 (from awsebcli)
  Using cached PyYAML-5.4.1.tar.gz (175 kB)
  Installing build dependencies ... done
  Getting requirements to build wheel ... error
  error: subprocess-exited-with-error
  
  × Getting requirements to build wheel did not run successfully.
  │ exit code: 1
  ╰─> [68 lines of output]

To fix this issue with PyYAML you must install PyYAML separately.

Do the following and try installing eb again after.

echo 'Cython < 3.0' > /tmp/constraint.txt
PIP_CONSTRAINT=/tmp/constraint.txt pip install 'PyYAML==5.4.1'

Make B/W SVG charts with matplotlib

Tue, 01 Aug 2023 17:04:10 +0200

Install pip requirements.

pip install matplotlib
pip install pandas

Example of data being used.

Epoch,Connect (NLB),Processing (NLB),Waiting (NLB),Total (NLB),Connect (ALB),Processing (ALB),Waiting (ALB),Total (ALB)
1,57.7,315.7,309.4,321.6,9,104.4,98.3,105.7
2,121.9,114.4,100.3,176.9,5.8,99.1,97.1,101.1
3,5.3,229.4,231.2,231.4,14.2,83,69.4,87.9
4,4.2,134.5,112.2,135.3,5.3,132.4,105.5,134.1
5,5.8,247.4,246.8,248.1,6,74.3,70.2,75.5
6,9.9,122.9,100.6,122.7,7.5,241.1,79.3,242.3
7,6.1,170.2,106.4,170.5,7.2,382.4,375.1,383.8
8,6.6,194.3,201.4,195.5,7.1,130.9,104.8,132.6
9,6.4,146.1,122.3,147.7,9.4,95.6,74,96.4

In the code you can use df as dataframes and use the headers like df["Epoch"]. This is how you get a column data with pandas.

The Python code responsible for generating a chart:

import csv
import sys

import matplotlib.pyplot as plt
import pandas as pd

# Read the data
df = pd.read_csv("data.csv")

# Settings
plt.title("Connect median NLB vs ALB")
plt.tight_layout(pad=2)
fig = plt.gcf()
fig.set_size_inches(10, 4)

# Plotting
plt.plot(df["Epoch"], df["Connect (ALB)"], label = "ALB", color="black", linestyle="-")
plt.plot(df["Epoch"], df["Connect (NLB)"], label = "NLB", color="black", linestyle="--")

# Adding x and y axis labels
plt.xlabel("Epoch", fontstyle="italic")
plt.ylabel("Median value (ms)", fontstyle="italic")

# Legend
legend = plt.legend()
legend.get_frame().set_linewidth(0)

# Export as SVG
plt.savefig("plot.svg", format="svg")

The image above is SVG and you can zoom in and out and check that the image is vector.

Set color temperature of displays on i3

Fri, 14 Jul 2023 09:19:31 +0200

I have been using Gnome's night shift for a while now and I have been missing this feature under i3wm. This can be done with redshift.

On Debian install with sudo apt install redshift
And then manually set it with redshift -O 3000
Reset the current settings with redshift -x

Fix screen tearing on Debian 12 Xorg and i3

Mon, 10 Jul 2023 04:21:48 +0200

I have been experiencing some issues with Intel® Integrated HD Graphics 3000 under Debian 12 with Xorg and i3. Using picom compositor didn't help. To fix this issue create new file /etc/X11/xorg.conf.d/20-intel.conf as root and put the following in the file.

Section "Device"
  Identifier "Intel Graphics"
  Driver "intel"
  Option "TearFree" "true"
EndSection

Reboot the system and that should be it.

Online radio streaming with MPV from terminal

Mon, 10 Jul 2023 03:34:45 +0200

Recently I have been using my Thinkpad x220 more and there are some constraints I have faced with it. CPU is not as powerful as on my main machine and I really want to listen to some music while using the machine. Browsers really are bloat.

Check out this site https://streamurl.link/ and copy the stream url and then do mpv streamlink.

60's IBM Computers Commercial

Thu, 29 Jun 2023 22:13:45 +0200

Likely aired during an hour-long program during the 1960s, long commercials such as this typically aired during hour-long programs. They would not have aired during a half-hour program.

10/GUI 10 Finger Multitouch User Interface

Thu, 29 Jun 2023 14:51:39 +0200

Message from 10/GUI team (page 10gui.com does not exist anymore):

Over a quarter-century ago, Xerox introduced the modern graphical user interface paradigm we today take for granted.

That it has endured is a testament to the genius of its design. But the industry is now at a crossroads: New technologies promise higher-bandwidth interaction, but have yet to find a truly viable implementation.

10/GUI aims to bridge this gap by rethinking the desktop to leverage technology in an intuitive and powerful way.

Alacritty open links with modifier

Sun, 25 Jun 2023 17:17:16 +0200

Alacritty by default makes all links in the terminal output clickable and this gets annoying rather quickly. I liked the default behavior of Gnome terminal where you needed to hold Control key and then you could click and open links.

To achieve this in Alacritty you need to provide a hint in the configuration file. Config file is located at ~/.config/alacritty/alacritty.yml.

hints:
  enabled:
  - regex: "(mailto:|gemini:|gopher:|https:|http:|news:|file:|git:|ssh:|ftp:)\
            [^\u0000-\u001F\u007F-\u009F<>\"\\s{-}\\^⟨⟩`]+"
    command: xdg-open
    post_processing: true
    mouse:
      enabled: true
      mods: Control

The following should work under any Linux system. For macOS, you will need to change command: xdg-open to something else.

Now the links will be visible and clickable only when Control key is being pressed.

Source: https://github.com/alacritty/alacritty/issues/5246

Development environments with Nix

Sun, 25 Jun 2023 16:38:10 +0200

Nix is amazing for making reproducible cross OS development environment.

First you need to install Nix package manager.

Create a file shell.nix in your project folder.
In the section that has python3 etc add programs you want to use. These can be CLI or GUI applications. It doesn't matter to Nix.

{ pkgs ? import <nixpkgs> {} }:
  pkgs.mkShell {
    nativeBuildInputs = with pkgs.buildPackages; [
	  python3
	  tinycc
	];
}

And then run it nix-shell. By default it will look for shell.nix file. If you want to specify a different file use nix-shell file.nix. That is about it.

When the shell is spawned it could happen that your PS1 prompt will be overwritten and your prompt will look differently. In that case you need to either do NIX_SHELL_PRESERVE_PROMPT=1 nix shell or add NIX_SHELL_PRESERVE_PROMPT variable to your bashrc or zshrc file and set it to 1.

I also have a modified PS1 prompt for Bash that I use and it also catches the usage of Nix shell.

NIX_SHELL_PRESERVE_PROMPT=1

parse_git_branch() {
	git branch 2> /dev/null | sed -e '/^[^*]/d' -e 's/* \(.*\)/ (\1)/'
}

is_inside_nix_shell() {
	nix_shell_name="$(basename "$IN_NIX_SHELL" 2>/dev/null)"
	if [[ -n "$nix_shell_name" ]]; then
		echo " \e[0;36m(nix-shell)\e[0m"
	fi
}

export PS1="[\033[38;5;9m\]\u@\h\[$(tput sgr0)\]]$(is_inside_nix_shell)\[\033[33m\]\$(parse_git_branch)\[\033[00m\] \w\[$(tput sgr0)\] \n$ "

And this is what it looks like when you are in a Nix shell. Otherwise that part of prompt is omitted

More resources:

Making cgit look nicer

Sat, 24 Jun 2023 13:33:58 +0200

For personal use I have a private Git server set up and I use GitHub just as a mirror. By default the cgit theme looks a bit dated so I made the flowing theme.

/etc/cgitrc

css=/cgit.css
logo=/startrek.gif
favicon=/favicon.png
source-filter=/usr/lib/cgit/filters/syntax-highlighting-edited.sh
about-filter=/usr/lib/cgit/filters/about-formatting.sh

local-time=1
snapshots=tar.gz
repository-sort=age
cache-size=1000
branch-sort=age
summary-log=200
max-atom-items=50
max-repo-count=100

enable-index-owner=0
enable-follow-links=1
enable-log-filecount=1
enable-log-linecount=1

root-title=Place for code, experiments and other bullshit!
root-desc=
clone-url=git@git.mitjafelicijan.com:/home/git/$CGIT_REPO_URL

mimetype.gif=image/gif
mimetype.html=text/html
mimetype.jpg=image/jpeg
mimetype.jpeg=image/jpeg
mimetype.pdf=application/pdf
mimetype.png=image/png
mimetype.svg=image/svg+xml

readme=:README.md
readme=:readme.md

# Must be at the end!
virtual-root=/
scan-path=/home/git/

For syntax-highlighting-edited.sh follow instructions on https://wiki.archlinux.org/title/Cgit.

/usr/share/cgit/cgit.css

* {
        font-size: 11pt;
}

body {
        font-family: monospace;
        background: white;
        padding: 1em;
}

th, td {
        text-align: left;
}

/* HEADER */

#header {
        margin-bottom: 1em;
}

#header .logo img {
        display: block;
        height: 3em;
        margin-right: 10px;
}

#header .sub.right {
        display: none;
}

/* FOOTER */

.footer {
        margin-top: 2em;
        font-style: italic;
}

.footer, .footer a {
        color: gray;
}

/* TABS */

.tabs a {
        margin-bottom: 2em;
        display: inline-block;
        margin-right: 1em;
}

.tabs td a:only-child {
        display: none;
}

/* HIDING ELEMENTS */

.cgit-panel, .form {
        display: none;
}

/* LISTS */

.list td, .list th {
        padding-right: 2em;
}

.list .nohover a {
        color: black;
}

.list .button {
        padding-right: 0.5em;
}

/* COMMIT */

.commit-subject {
        padding: 1em 0;
}

.decoration a {
        padding-left: 0.5em;
}

.commit-info th {
        padding-right: 1em;
}

.commit-subject {
        padding: 2em 0;
}

table.diff div.head {
        padding-top: 2em;
}

table.diffstat td {
        padding-right: 1em;
}

/* CONTENT */

.linenumbers {
        padding-right: 0.5em;
}

.linenumbers a {
        color: gray;
}

.pager {
        display: flex;
        list-style-type: none;
        padding: 0;
        gap: 0.5em;
}

/* DIFF COLORS */

table.diff {
        width: 100%;
}

table.diff td {
        white-space: pre;
}

table.diff td div.head {
        font-weight: bold;
        margin-top: 1em;
        color: black;
}

table.diff td div.hunk {
        color: #009;
}

table.diff td div.add {
        color: green;
}

table.diff td div.del {
        color: red;
}

Simple presentations with Markdown

Wed, 21 Jun 2023 08:54:48 +0200

A simple way to make presentations without using desktop apps or using online services is https://github.com/remarkjs/remark.

First create index.html and be sure you make changes to config variable.

<!DOCTYPE html>
<html>

<head>
    <title></title>
    <meta charset="utf-8">
    <style>
        body {
            font-family: 'SF Pro Display';
        }

        .remark-code,
        .remark-inline-code {
            font-family: 'SF Mono';
            font-size: medium;
            background-color: gainsboro;
            border-radius: 5px;
            padding: 0 5px;
        }
    </style>
</head>

<body>
    <textarea id="source"></textarea>
    <script src="https://remarkjs.com/downloads/remark-latest.min.js"></script>
    <script>
		const config = {
			title: 'My presentation',
			file: 'presentation.md',
		};
	
		document.title = config.title;
        remark.create({ sourceUrl: config.file });
    </script>
</body>

</html>

Now the markdown file presentation.md with presenetation. --- is used to separate slides. Other stuff is just pure markdown.

class: center, middle

# Main title of the presentation

---

# Fist slide

Eveniet mollitia nemo architecto rerum aut iure iste. Sit nihil nobis libero iusto fugit nam laudantium ut. Dignissimos corrupti laudantium nisi.

- Lorem ipsum dolor sit amet, consectetur adipiscing elit.
- Integer aliquet mauris a felis fringilla, ut congue massa finibus.

---

# Slide two

- Lorem ipsum dolor sit amet, consectetur adipiscing elit.
- Vestibulum eget leo ac dolor venenatis pulvinar.

Bulk thumbnails

Sun, 04 Jun 2023 20:46:56 +0200

Make bulk thumbnails of JPGs with ImageMagick.

#!/bin/bash

directory="./images/"
dimensions="360x360"

for file in "$directory"*.jpg; do
  convert "$file" -resize $dimensions "$file" "${file%.*}-thumbnail.jpg"
done

Edsger W. Dijkstra Manuscripts ebook

Thu, 01 Jun 2023 22:47:56 +0200

I love reading the original manuscripts of Edsger W. Dijkstra. They are available online at the University of Texas at Austin website, but I also found MOBI version. I converted it into ePub as well.

Downloads:

Sources and credits:

Extending dte editor

Wed, 31 May 2023 08:12:45 +0200

dte is an interesting editor I started using lately more and more. Since it is using execvp() it can be easily extended. I needed comment/uncomment feature so I created a small utility that does this for me. Code lives on repository dte extensions but this utilities can be used for whatever you want. Make sure you have version 1.11 or above.

Next one will be invoking formatter based on the type of a file.

My config that works for me.

set show-line-numbers true;
set tab-width 4;
set case-sensitive-search false;

# Special aliases
alias m_comment 'exec -s -i line -o buffer -e errmsg ~/.dte/bin/comment'
alias m_format 'save; exec go fmt .; reload'
alias m_duplicate 'copy;paste';

# Useful aliases.
alias m_force_close 'quit -f';
alias m_reload 'close; open $FILE'

# Key bindings.
bind M-s save;
bind M-q m_force_close;
bind M-z refresh;
bind C-down blkdown;
bind C-up blkup;
bind C-_ m_comment;
bind M-. m_format;
bind C-d m_duplicate;

# Syntax highlighting.
hi preproc magenta;
hi keyword red;
hi linenumber blue;
hi comment cyan;

Grep to Less that maintain colors

Mon, 29 May 2023 21:27:07 +0200

I often use grep to search for todo's in my code and other people's code and then pipe them in less and I missed having colors that grep outputs in less.

Grep's --color=always use markers to highlight the matching strings.
Less's -R option outputs "raw" control characters.

You could use alias grep='grep --color=always' and alias less='less -R' or create todo function in your .bashrc that accepts first argument as search string.

# This is where the magic happens.
grep --color=always -rni "TODO:" | less -R

Easy measure time took in a bash script

Sun, 28 May 2023 17:53:20 +0200

In Bash, the $SECONDS variable is a special variable that automatically keeps track of the number of seconds since the current shell or script started executing. It starts counting from the moment the script begins running.

#!/bin/bash

# Reset the timer to zero.
SECONDS=0

# Do something.
sleep 5

# Print the time elapsed.
echo "Time taken: $SECONDS seconds"

Make DCSS playable on 4k displays

Sat, 27 May 2023 19:35:11 +0200

Dungeon Crawl Stone Soup has a a very small font by default. On a 4k display, it is barely readable. This is how I made it playable.

Make a file ~/.crawlrc with the following content:

# Adjust the sizes to your liking.

tile_font_crt_size = 32
tile_font_stat_size = 32
tile_font_msg_size = 32
tile_font_tip_size = 32
tile_font_lbl_size = 32
tile_sidebar_pixels = 64

To zoom in and out in viewport, press Ctrl+ and Ctrl- respectively.

All the possible options are documented in the Dungeon Crawl Stone Soup Options Guide file.

Drawing Pixels in Plan9

Sat, 27 May 2023 17:41:33 +0200

I have started exploring Plan9's graphics capabilities. This is a hello world alternative for drawing that draws a yellow square on a blue background.

More information:

draw.h header file contains all the drawing functions
draw man page has a bit more digestable descriptions of the draw functions
graphics man page has a bit more digestable descriptions of the graphics functions
all man pages can be a valuable resource for learning about the system

// main.c
#include <u.h>
#include <libc.h>
#include <draw.h>
#include <cursor.h>

void
main()
{
  ulong co;
  Image *im, *bg;
  co = 0x0000FFFF;

  if (initdraw(nil, nil, argv0) < 0)
  {
    sysfatal("%s: %r", argv0);
  }

  im = allocimage(display, Rect(0, 0, 300, 300), RGB24, 0, DYellow);
  bg = allocimage(display, Rect(0, 0, 1, 1), RGB24, 1, co);

  if (im == nil || bg == nil)
  {
    sysfatal("not enough memory");
  }

  draw(screen, screen->r, bg, nil, ZP);
  draw(screen, screen->r, im, nil, Pt(-40, -40));

  flushimage(display, Refnone);

  // Wait 10 seconds before exiting.
  sleep(10000);

  exits(nil);
}

And then compile with mk (mkfile below):

# mkfile
</$objtype/mkfile

RC=/rc/bin
BIN=/$objtype/bin
MAN=/sys/man

main:
	$CC $CFLAGS main.c
	$LD $LDFLAGS -o main main.$O

And run with ./main. To exit the program, press Delete key (strange but this is the alternative for Ctrl+C).

This is very cool indeed!

Cronjobs on Github with Github Actions

Sat, 27 May 2023 00:35:36 +0200

In the root of your repository create a folder .github/workflows and in that folder create a file a file cron.yaml. This file can be named whatever you wish. But it has to be a yaml file.

File below (.github/workflows/cron.yaml) describes an action that will trigger every six hours and it will curl example.com.

However. Be sure that you have enough credits. Free account is not that generous with the minutes they give you for free. Check more about GitHub Actions usage on their website https://docs.github.com/en/actions.

# .github/workflows/cron.yaml
name: Do a curl every 6 hours
on:
  schedule:
    - cron: '0 */6 * * *'
jobs:
  cron:
    runs-on: ubuntu-latest
    steps:
      - name: Call some url
        run: curl 'https://example.com'

Dungeon Crawl Stone Soup - New player guide

Thu, 25 May 2023 22:00:00 +0200

An amazing game deserves an amazing guide. All this material can be find in some form on another on craw's official repository.

DCSS Quickstart - Very short introduction to the game
DCSS Manual - Extensive manual about the game

Movement and Exploration

You can move around with the numpad (try numlock on and off), vi-keys, or clicking with the mouse. Arrow keys work, though you can't move diagonally with them. Pressing Shift and a direction will move until you see/hit something.
Pressing > will take you down a staircase, and < to go up a staircase.
You can open doors by walking into them, and close them with C.
You can autoexplore by pressing o.
You can re-view recent messages with Ctrl-p.

Monsters and Combat

You can pick up items with , or g.
Wield weapons with w. Weapons have different stats.
- (You may also engage in Unarmed Combat, though it isn't very effective when untrained).
Attack monsters in melee by walking in their direction (or with Ctrl-direction).
You can wait with . or s, passing your turn - such as to get monsters into a corridor with you.
You can rest with 5, waiting until you are fully healed, or something noteworthy happens.
Either mouseover and rightclick, or use x then v on the monster to examine monsters. Monsters with a red border are 'dangerous' relative to your current XP level (XL).
Quiver (often ranged) actions for further use with Q.
You can fire ranged weapons manually with f, or auto-target your quiver with p or Shift-Tab. Throwing weapons can be thrown immediately, while launchers (like bows) need to be wielded first.

Items and Inventory

View your inventory by pressing i. Most item related commands can also be done with this menu.
You can wear amour with W; amour gives AC, while heavier body armour reduces EV.
Autoexplore will automatically pick up useful items, such as potions and scrolls, if you aren't in danger.
You can read scrolls with r and drink ("quaff") potions with q.
Equipment items may have brands, with special properties. Branded equipment is blue when unidentified.
Equipment items may be artifacts, often with unique properties, and are unmodifiable. They are written in white.
You can evoke wands with V.
You can put on jewelry with P, and remove it with R.
Gold is used in shops, which can be interacted with by either > or <.

Magic and Spellcasting

Once you find a spellbook, you can memorize spells with M.
You need to be the same XL as the spell's spell level in order to learn it, in addition to training magical skill (to lower failure rate).
Cast spells by pressing z, then the letter assigned to the spell. You may also Quiver a spell and then use it like a ranged weapon (with Shift-Tab).
You can view your memorized spells by pressing I (capital-i) or z.
Like HP, you can recover MP by resting (with 5).
Many spells can be positioned more effectively, or combined with other spells, in order to get (more effective) use out of them.
Heavier body amour and shields hamper spellcasting.

Gods and Divine Abilities

You may look at a god's overview by praying at their altar (with > or <). After praying, you can worship the god by pressing Enter afterwards.
Gods all have unique features about them. Trog, the god of the tutorial, is also the god of rage and bloodshed, and so despises spellcasting.
Gods like and dislike different things. Most gods either like killing things (like Trog) or exploring new areas (like Elyvilon), rewarding you piety (divine favor) for doing so.
You should learn to use and even rely on divine abilities often, as they are usually very strong. Trog's Berserk gives you 1.5x health, 1.5x speed (to all valid actions), and a big damage boost. Note that Berserk prevents most actions other than move and melee attack, and runs out very quickly if you aren't attacking. And after berserk ends, you are slowed down and can't berserk again for a short time.
In addition, the vast majority of abilities consume piety in the process. Regardless, this ability is very cheap, and the benefits are incredible, so don't hold back!
Pressing ^ will let you view your current god, abilities, and piety.

Sane defaults for tmux with more visible statusbar

Thu, 25 May 2023 12:00:00 +0200

# Remap prefix from 'C-b' to 'M-a'.
unbind C-b
set-option -g prefix M-a
bind-key M-a send-prefix

# Split panes using | and -.
bind | split-window -h
bind - split-window -v
unbind '"'
unbind %

# Start counting windows with 1.
set-option -g allow-rename on
set -g base-index 1
setw -g pane-base-index 1

# Statusbar: purple bg and white fg.
set -g status-bg '#480b8e'
set -g status-fg '#ffffff'

# Active window: black bg and white fg. 
set -g window-status-current-format "#[fg=#ffffff]#[bg=#111111]#[fg=#ffffff]#[bg=#111111] #I:#W #[fg=#ffffff]#[bg=#111111]"

# Disable mouse mode (tmux 2.1 and above).
set -g mouse off

Display xterm color palette

Thu, 25 May 2023 12:00:00 +0200

bash xterm-palette.sh - will show you number of max colors available
bash xterm-palette.sh -v - will create a list of all colors with codes

#!/usr/bin/env bash
# xterm-palette.sh

trap 'tput sgr0' exit		# Clean up even if user hits ^C

function setfg () {
    printf '\e[38;5;%dm' $1
}

function setbg () {
    printf '\e[48;5;%dm' $1
}

function showcolors() {
    # Given an integer, display that many colors
    for ((i=0; i<$1; i++))
    do
	printf '%4d ' $i
	setbg $i
	tput el
	tput sgr0
	echo
    done
    tput sgr0 el
}

# First, test if terminal supports OSC 4 at all.
printf '\e]4;%d;?\a' 0
read -d $'\a' -s -t 0.1 </dev/tty
if [ -z "$REPLY" ]
then
    # OSC 4 not supported, so we'll fall back to terminfo
    max=$(tput colors)
else
    # OSC 4 is supported, so use it for a binary search
    min=0
    max=256
    while [[ $((min+1)) -lt $max ]]
    do
	i=$(( (min+max)/2 ))
	printf '\e]4;%d;?\a' $i
	read -d $'\a' -s -t 0.1 </dev/tty
	if [ -z "$REPLY" ]
	then
            max=$i
	else
	    min=$i
	fi
    done
fi


# If -v is given, show all the colors
case ${1-none} in
    none)
	echo $max
	;;
    -v)
	showcolors $max
	;;
    *)
	if [[ "$1" -gt 0 ]]; then
	    showcolors $1
	else
	    echo $max
	fi
	;;
esac | less --raw-control-chars --QUIT-AT-EOF --no-init

My brand new Plan9/9front desktop

Wed, 24 May 2023 12:00:00 +0200

I have been experimenting with Plan9/9front for a week now. Noice! This is how my desktop looks like.

Parse RSS feeds with Lua

Tue, 23 May 2023 12:00:00 +0200

Example of parsing RSS feeds with Lua. Before running the script install:

feedparser with luarocks install feedparser
luasocket with luarocks install luasocket

local http = require("socket.http")
local feedparser = require("feedparser")

local feed_url = "https://mitjafelicijan.com/index.xml"

local response, status, _ = http.request(feed_url)
if status == 200 then
  local parsed = feedparser.parse(response)

  -- Print out feed details.
  print("> Title   ", parsed.feed.title)
  print("> Author  ", parsed.feed.author)
  print("> ID      ", parsed.feed.id)
  print("> Entries ", #parsed.entries)

  for _, item in ipairs(parsed.entries) do
    print("GUID    ", item.guid)
    print("Title   ", item.title)
    print("Link    ", item.link)
    print("Summary ", item.summary)
  end
else
  print("! Request failed. Status:", status)
end

Extend Lua with custom C functions using Clang

Tue, 23 May 2023 12:00:00 +0200

Here is a boilerplate for extending Lua with custom C functions. This requires Clang and Lua 5.1 to be installed. GCC can be used instead of Clang, but the Makefile will need to be modified.

nativefunc.c

#include <lua.h>
#include <lauxlib.h>

static int l_mult50(lua_State *L) {
  double number = luaL_checknumber(L, 1);
  lua_pushnumber(L, number * 50);
  return 1;
}

int luaopen_nativefunc(lua_State *L) {
  static const struct luaL_Reg nativeFuncLib[] = { {"mult50", l_mult50}, {NULL, NULL} };

  luaL_register(L, "nativelib", nativeFuncLib);
  return 1;
}

main.lua

require "nativefunc"
print(nativelib.mult50(50))

Makefile

CC       = clang
CFLAGS   =
INCLUDES = `pkg-config lua5.1 --cflags-only-I`

all:
  $(CC) -shared -o nativefunc.so -fPIC nativefunc.c $(CFLAGS) $(INCLUDES)

clean:
  rm *.so

Execute not blocking async shell command in C#

Mon, 22 May 2023 12:00:00 +0200

Execute a shell command in async in C# while not blocking the UI thread.

private async Task executeCopyCommand()
{
  await Task.Run(() =>
  {
    var processStartInfo = new ProcessStartInfo("cmd", "/c dir")
    {
      RedirectStandardOutput = true,
      UseShellExecute = false,
      CreateNoWindow = true
    };

    var process = new Process
    {
      StartInfo = processStartInfo
    };

    process.Start();
    process.WaitForExit();
  });
}

Make sure that async is present in the function definition and await is used in the method that calls executeCopyCommand().

private async void button_Click(object sender, EventArgs e)
{
  await executeCopyCommand();
}

Change permissions of matching files recursively

Tue, 16 May 2023 12:00:00 +0200

Replace *.xml with your pattern. This will remove executable bit from all files matching the pattern. Change + to - to add executable bit.

find . -type f -name "*.xml" -exec chmod -x {} +

Previews how man page written in Troff will look like

Mon, 15 May 2023 12:00:00 +0200

Troff is used to write man pages and it is difficult to read it so this will preview how it will look like when it is rendered.

# On Linux system.
groff -man -Tascii filename

# On Plan9 system.
man 1 filename

Convert all MKV files into other formats

Sun, 14 May 2023 12:00:00 +0200

You will need ffmpeg installed on your system. This will convert all MKV files into WebM format.

# Convert all MKV files into WebM format.
find ./ -name '*.mkv' -exec bash -c 'ffmpeg -i "$0" -vcodec libvpx -acodec libvorbis -cpu-used 5 -threads 8 "${0%%.mp4}.webm"' {} \;

# Convert all MKV files into MP4 format.
find ./ -name '*.mkv' -exec bash -c 'ffmpeg -i "$0" c:a copy -c:v copy -cpu-used 5 -threads 8 "${0%%.mp4}.mp4"' {} \;

Download list of YouTube files

Sat, 13 May 2023 12:00:00 +0200

If you need to download a list of YouTube videos and don't want to download the actual YouTube list (which yt-dlp supports), you can use the following method.

// Used to get list of raw URL's from YouTube's video tab'.
// Copy them into videos.txt.
document.querySelectorAll('#contents a.ytd-thumbnail.style-scope.ytd-thumbnail').forEach(el => console.log(el.href))

Download and install https://github.com/yt-dlp/yt-dlp.

# This will download all videos in videos.txt.
yt-dlp --batch-file videos.txt -N `nproc` -f webm

Install Plan9port on Linux

Fri, 12 May 2023 12:00:00 +0200

Install Plan9port on Linux. This applies to Plan9port. This is a port of many Plan 9 programs to Unix-like operating systems. Useful for programs like 9term and rc.

sudo apt-get install gcc libx11-dev libxt-dev libxext-dev libfontconfig1-dev
git clone https://github.com/9fans/plan9port $HOME/plan9
cd $HOME/plan9/plan9port
./INSTALL -r $HOME/plan9

Fix bootloader not being written in Plan9

Thu, 11 May 2023 12:00:00 +0200

If the bootloader is not being written to a disk when installing 9front on real harware try clearing first sector of the disk with the following command.

dd if=/dev/zero of=/dev/sdX bs=512 count=1

# If command above doesn't work try this one, wait couple of seconds and
# press delete key to stop the command.
cat </dev/zero >/dev/sd*/data

Take a screenshot in Plan9

Wed, 10 May 2023 12:00:00 +0200

Take a screenshot in Plan9. This applies to Plan9 and 9front. This will take a screenshot of the screen and output it to /dev/screen. You can then use topng to convert it to a png image.

# Instant screenshot.
cat /dev/screen | topng > screen.png

# Delayed screenshot (5 seconds).
sleep 5; cat /dev/screen | topng > screen.png

#cat-v on weechat configuration

Tue, 09 May 2023 12:00:00 +0200

Set up weechat to connect to #cat-v on oftc. This applies to weechat but should be similar for other irc clients.

# Install weechat and launch it and execute the following commands.

/server add oftc irc.oftc.net -tls
/set irc.server.oftc.autoconnect on
/set irc.server.oftc.autojoin "#cat-v"
/set irc.server.oftc.nicks "nick1,nick2,nick3"

Write ISO to USB Key

Mon, 08 May 2023 12:00:00 +0200

Write ISO to USB key. Nothing fancy here.

sudo dd if=iso_file.iso of=/dev/sdX bs=4M status=progress conv=fdatasync

Mount Plan9 over network

Sun, 07 May 2023 12:00:00 +0200

First install libfuse with sudo apt install libfuse-dev.
Then clone https://github.com/ftrvxmtrx/9pfs and compile it with make.
Copy 9pfs to your path.

# On Plan9 side
ip/ipconfig # enables network
aux/listen1 -tv tcp!*!9999 /bin/exportfs -r tmp # export tmp folder

# On Linux side
9pfs 172.18.0.1 -p 9999 local_folder # mount
umount local_folder # unmount

Push to multiple origins at once in Git

Sat, 06 May 2023 12:00:00 +0200

Sometimes you want to push to multiple origins at once. This is useful if you have a mirror of your repository on another server. You can do this by adding multiple push urls to your git config. This is a shorthand for command above.

git config --global alias.pushall '!sh -c "git remote | xargs -L1 git push --all"'

Run 9front in Qemu

Fri, 05 May 2023 12:00:00 +0200

Run 9front in Qemu. This applies to Plan9 and 9front.

Download from here http://9front.org/iso/.

# Create a qcow2 image.
qemu-img create -f qcow2 $HOME/VM/9front.qcow2.img 30G

# Run the VM.
qemu-system-x86_64 -cpu host -enable-kvm -m 1024 \
    -net nic,model=virtio,macaddr=52:54:00:00:EE:03 -net user \
    -device virtio-scsi-pci,id=scsi \
    -drive if=none,id=vd0,file=$HOME/VM/9front.qcow2.img \
    -device scsi-hd,drive=vd0 \
    -drive if=none,id=vd1,file=$HOME/VM/ISO/9front.386.iso \
    -device scsi-cd,drive=vd1,bootindex=0

Cache busting in Hugo

Mon, 01 May 2023 12:00:00 +0200

\{\{ $cachebuster := delimit (shuffle (split (md5 "6fab11c6669976d759d2992eff1dd5be") "" )) "" \}\}

<link rel="stylesheet" href="/style.css?v=\{\{ $cachebuster \}\}">

This 6fab11c6669976d759d2992eff1dd5be can be random string you generate use. You can use whatever you want.

Trying to build a New kind of terminal emulator for the modern age

Thu, 26 Jan 2023 12:00:00 +0200

Over the past few weeks, I have been really thinking about terminal emulators, how we interact with computers, the separation of text-based programs and GUI ones. To be perfectly honest, I got pissed off one evening when I was cleaning up files on my computer. Normally, I go into console and do ncdu and check where the junk is. Then I start deleting stuff. Without any discrimination, usually. But when it comes to screenshots, I have learned that it's good to keep them somewhere near if I need to refer to something that I was doing. I am an avid screenshot taker. So at that point I checked Pictures folder and also did a basic search find . -type f -name "*.jpg" for all the JPEG files in my home directory and immediately got pissed off. Why can’t I see thumbnails in my terminal? I know why, but why in the year of 2022 this is still a problem. I am used to traversing my disk via terminal. I am faster, and I am more comfortable this way. But when it comes to visualization, I then need to revert to GUI applications and again find the same file to see it. I know that programs like feh and sxiv are available, but I would just like to see the preview. Like Jupyter notebook or something similar. Just having it inline. Part of a result.

It also didn’t help that I was spending some time with the Plan 9 Operating system. More specifically 9FRONT. The way that ACME editor handles text editing is just wonderful. Different and fresh somehow, even though it’s super old.

So, I went on a lookout for an interesting way of visualizing results of some query. I found these applications to be outstanding examples of how not to be a captive of a predetermined way of doing things.

My idea is not as out there as ACME is, but it is a spin on the terminal emulators. I like the modes that Vi/Vim provides you with. I like the way the Emacs does its own M-x M-c. Furthermore, I really like how Mathematica and Jupyter present the data in a free flowing form. And I love how Temple OS is basically a C interpreter on some level.

Note: This is part 1 of the journey. Nowhere finished yet. I am just tinkering with this at the moment. This whole thing can easily spectacularly fail.

So I started. I knew that I wanted to have the couple of modes, but I didn’t like the repetition of keystrokes, so the only option was to have some sort of toggle and indicate to the user that they are in a special mode. Like Vi does for Normal and Visual mode.

These modes would for the first version be:

Preview mode (toggle with Ctrl + P)
- When this mode would be enabled, the ls command would try to find images from the results and display thumbnails from them in the terminal itself. No ASCII art. Proper images. In a grid!
Detach mode (toggle with Ctrl + D)
- When this mode would be enabled, every command would open a new window and execute that command in it. This would be useful for starting htop in a separate window.

The reason for having these modes togglable is to not ask for previews every time. You enable a mode and until you disable it, it behaves that way. Purely out of ergonomic reasons.

I would like to treat every terminal I open as a session mentally. When I start using the terminal, I start digging deeper into the issue I am trying to resolve. And while I am doing this, I would like to open detached windows etc. A lot of these things can be done easily with something like i3, but also that pull you out of the context of what you were doing. I would like to orchestrate everything from one single point.

In planning for this project, I knew that I would need to use a language like C and a library such as SDL2 in order to achieve the desired results. I had considered other options, but ultimately determined that SDL2 was the best fit based on its capabilities and reputation in the programming community.

At first, I thought the idea of a hardware accelerated terminal was a bit of a joke. It seemed like such a niche and unnecessary feature, especially given the fact that terminal emulators have been around for decades and have always relied on software rendering. But to be fair, Alacritty is doing the same thing. Well, they are doing a remarkable job at it.

So, I embarked on a journey. Everything has to start somewhere. For me, it started with creating a window! It has to start somewhere. 🙂

// Oh, Hi Mark!
// Create the window, obviously.
SDL_Window *window = SDL_CreateWindow(
  WINDOW_TITLE, SDL_WINDOWPOS_UNDEFINED, SDL_WINDOWPOS_UNDEFINED,
  WINDOW_WIDTH, WINDOW_HEIGHT,
  SDL_WINDOW_RESIZABLE | SDL_WINDOW_OPENGL | SDL_WINDOW_SHOWN);

I continued like this to get some text displayed on the screen.

I noted that TTF_RenderText_Solid rendered text really poorly. There were no antialiasing at all. In my wisdom, I never checked the documentation. Well, that was a fail. To uneducated like me: TTF_RenderText_Solid renders Latin1 text at fast quality to a new 8-bit surface. So, that's why the texts looked like shit. No wonder.

Remarks on TTF_RenderText_Solid: This function will allocate a new 8-bit, palettized surface. The surface's 0 pixel will be the colorkey, giving a transparent background. The 1 pixel will be set to the text color.

After I replaced it with TTF_RenderText_LCD which renders Latin1 text at LCD subpixel quality to a new ARGB surface, the text started looking good. Really make sure you read the documentation. It’s actually good. As a side note, you can find all the documentation regarding SDL2 on their Wiki.

After that was done, I started working on displaying other things like Preview and Detach modes. This wasn’t really that hard. In SDL2 you can check all the available events with while (SDL_PollEvent(&event) > 0) and have a bunch of switch statements to determine which key is currently being pressed. More about keys, SDLKey and mroe about pooling the events on SDL_PollEvent.

while (SDL_PollEvent(&event) > 0)
{
  switch (event.type)
  {
  case SDL_QUIT:
    running = false;
    break;

  case SDL_TEXTINPUT:
    if (!meta_key_pressed)
    {
      strncat(input_prompt_text, event.text.text, 1);
      update_input_prompt = true;
    }
    break;
	}
}

After that was somewhat working correctly, I started creating a struct that would hold all the commands and results and I call them Cells. Yes, I stole that naming idea from Jupyter.

typedef struct
{
  char *command;
  char *result;
  SDL_Surface *surface;
  SDL_Texture *texture;
  SDL_Rect rect;
} Cell;

I am at a place now where I am starting to implement scrolling. This will for sure be fun to code. Memory management in C is super easy. 😂

I have also added a simple INI file like configuration support. It is done in an STB style of header and maps to specific options supported by the terminal. It is not universal, and the code below demonstrates how I will use it in the future.

#ifndef CONFIG_H
#define CONFIG_H

/*
# This is a comment

# This is the first configuration option
dettach=value11111

# This is the second configuration option
preview=value22222

# This is the third configuration option
debug=value33333
*/

// Define a struct to hold the configuration options
typedef struct
{
    char dettach[256];
    char preview[256];
    char debug[256];
} Config;

// Read the configuration file and return the options as a struct
extern Config read_config_file(const char *filename)
{
  // Create a struct to hold the configuration options
  Config config = {0};

  // Open the configuration file
  FILE *file = fopen(filename, "r");

  // Read each line from the file
  char line[256];
  while (fgets(line, sizeof(line), file))
  {
    // Check if this line is a comment or empty
    if (line[0] == '#' || line[0] == '\n')
      continue;

    // Parse the line to get the option and value
    char option[128], value[128];
    if (sscanf(line, "%[^=]=%s", option, value) != 2)
      continue;

    // Set the value of the appropriate option in the config struct
    if (strcmp(option, "dettach") == 0)
    {
      strncpy(config.option1, value, sizeof(config.option1));
    }
    else if (strcmp(option, "preview") == 0)
    {
      strncpy(config.option2, value, sizeof(config.option2));
    }
    else if (strcmp(option, "debug") == 0)
    {
      strncpy(config.option3, value, sizeof(config.option3));
    }
  }

  // Close the configuration file
  fclose(file);

  // Return the configuration options
  return config;
}

#endif

This is as far as I managed to get for now. I have a daily job and this prohibits me to work on these things full time. But I should probably get back and finish this. At least have a simple version working out, so I can start testing it on my machines. Fingers crossed. 🕵️‍♂️

Microsoundtrack — That sound that machine makes when struggling

Sun, 16 Oct 2022 12:00:00 +0200

A couple of months ago, I got an idea about micro soundtracks. In this concept, you are the observer, director, and audience in this tiny movies.

What you do is to attempt to imagine what would be happening around you based on a title of the song and let the song help you fill the void in your story.

I made these songs is Logic Pro X. Every year or so I do this kind of thing and make a couple of songs similar to this. But this is the first time I am posting about it.

You can listen to the whole set on Youtube or scroll down the page and there are embedded players for each song.

A bunch of inter-dimensional people with loud clocks

A group of inter-dimensional people are going up and down the elevator with you while having loud clocks around their necks. Each clock ticks on a different frequency. A lot of other sounds are getting drawn into your dimension, resulting in a strange merging of dimensions.

That sound that machine makes when struggling by Mitja Felicijan

Two black holes conversing about the weather

You are a traveler in a spaceship flying very close to two colliding black holes having a discussion about the weather while tearing each other apart. During all this your ship is getting pulled into the event horizon of both black holes, putting a lot of strain on your spaceship.

That sound that machine makes when struggling by Mitja Felicijan

A planet where every organism is a plant

You land on a planet where every living organism is a plant and among those plants some of them are highly intelligent, and you were asked to make first contact with the native species. Your visit takes place in a giant cave where you are meeting these plants, and they are talking to you.

That sound that machine makes when struggling by Mitja Felicijan

Bio implants having a fit and reprogramming your brain

In a distant future where everybody has bio implants, you have just received your first one, which happens to be a brain implant. Something goes wrong, and your implant is starting to misbehave, and you are experiencing brain malfunctions. You are on the streets at night a couple of hours after your procedure. You can feel your sanity breaking down.

That sound that machine makes when struggling by Mitja Felicijan

Cow animation

I also made this little cow animation. Go into full screen to see the effects in more details.

Aerial photography of algae spotted on river Sava

Sat, 13 Aug 2022 12:00:00 +0200

This is a bit of a different post than I usually write, but quite interesting one to me. River Sava has plenty of hydropower plants located down the stream. This makes regulating the strength of a current easier than normally. Because of lower stream strength and high temperatures, algae has formed on the river. This is the first time I've seen something like this in my whole life.

Below are some photographs taken from a DJI drone capturing the event.

I will try to get more photos of this in the future days and if something intriguing shows up will post it again on the blog.

What would DNA sound if synthesized to an audio file

Tue, 05 Jul 2022 12:00:00 +0200

Introduction

Lately, I have been thinking a lot about the nature of life, what are the foundation blocks of life and things like that. It's remarkable how complex and on the other hand simple the creation is when you look at it. The miracle of life keeps us grounded when our imagination goes wild. If the DNA are the blocks of life, you could consider them to be an API nature provided us to better understand all of this chaos masquerading as order.

I have been reading a lot about superintelligence and our somehow misguided path to create general artificial intelligence. What would the building blocks or our creation look like? Is the compression really the ultimate storage of information? Will our creation also ponder this questions when creating new worlds for themselves, or will we just disappear into the vastness of possibilities? It is a little offensive that we are playing God whilst being completely ignorant of our own reality. Who knows! Like many other breakthroughs, this one will also come at a cost not known to us when it finally happens.

To keep things a bit lighter, I decided to convert some popular DNA sequences into an audio files for us to listen to. I am not the first one, nor I will be the last one to do this. But it is an interesting exercise in better understanding the relationship between art and science. Maybe listening to DNA instead of parsing it will find a way into better understanding, or at least enjoying the creation and cryptic nature of life.

DNA encoding and primer example

I have been exploring DNA in the past in my post from about 3 years ago in Encoding binary data into DNA sequence where I have been converting all sorts of data into DNA sequences.

This will be a similar exercise but instead of converting to DNA, I will be generating tones from Nucleotides.

Nucleotides	Note	Frequency
A (Adenine)	A	440 Hz
C (Cytosine)	C	783.99 Hz
G (Guanine)	G	523.25 Hz
T (Thymine)	D	587.33 Hz

Since we do not have T in equal-tempered scale, I choose D to represent T note.

You can check Frequencies for equal-tempered scale, A4 = 440 Hz. For this tuning, we also choose Speed of Sound = 345 m/s = 1130 ft/s = 770 miles/hr.

Now that we have this out of the way, we can also brush up on the DNA sequencing a bit. This is a famous quote I also used for the encoding tests, and it goes like this.

How wonderful that we have met with a paradox. Now we have some hope of making progress. ― Niels Bohr

>SEQ1
GACAGCTTGTGTACAAGTGTGCTTGCTCGCGAGCGGGTACGCGCGTGGGCTAACAAGTGA
GCCAGCAGGTGAACAAGTGTGCGGACAAGCCAGCAGGTGCGCGGACAAGCTGGCGGGTGA
ACAAGTGTGCCGGTGAGCCAACAAGCAGACAAGTAAGCAGGTACGCAGGCGAGCTTGTCA
ACTCACAAGATCGCTTGTGTACAAGTGTGCGGACAAGCCAGCAGGTGCGCGGACAAGTAT
GCTTGCTGGCGGACAAGCCAGCTTGTAAGCGGACAAGCTTGCGCACAAGCTGGCAGGCCT
GCCGGCTCGCGTACAAATTCACAAGTAAGTACGCTTGCGTGTACGCGGGTATGTATACTC
AACCTCACCAAACGGGACAAGATCGCCGGCGGGCTAGTATACAAGAACGCTTGCCAGTAC
AACC

This is what we gonna work with to get things rolling forward, when creating parser and waveform generator.

Parsing DNA data

This step is rather simple one. All we need to do is parse input DNA sequence in FASTA format well known in Bioinformatics to extract single Nucleotides that will be converted into separate tones based on equal-tempered scale explained above.

nucleotide_tone_map = {
  'A': 440,
  'C': 523.25,
  'G': 783.99,
  'T': 587.33,  # converted to D
}

def split(word):
  return [char for char in word]

def generate_from_dna_sequence(sequence):
  for nucleotide in split(sequence):
    print(nucleotide, nucleotide_tone_map[nucleotide])

Generating sine wave

Because we are essentially creating a long stream of notes we will be appending sine notes to a global array we will later use for creating a WAV file out of it.

import math

def append_sinewave(freq=440.0, duration_milliseconds=500, volume=1.0):
  global audio

  num_samples = duration_milliseconds * (sample_rate / 1000.0)

  for x in range(int(num_samples)):
    audio.append(volume * math.sin(2 * math.pi * freq * (x / sample_rate)))

  return

The sine wave generated here is the standard beep. If you want something more aggressive, you could try a square or saw tooth waveform.

Generating a WAV file from accumulated sine waves

import wave
import struct

def save_wav(file_name):
  wav_file = wave.open(file_name, 'w')
  nchannels = 1
  sampwidth = 2

  nframes = len(audio)
  comptype = 'NONE'
  compname = 'not compressed'
  wav_file.setparams((nchannels, sampwidth, sample_rate, nframes, comptype, compname))

  for sample in audio:
    wav_file.writeframes(struct.pack('h', int(sample * 32767.0)))

  wav_file.close()

44100 is the industry standard sample rate - CD quality. If you need to save on file size, you can adjust it downwards. The standard for low quality is, 8000 or 8kHz.

WAV files here are using short, 16 bit, signed integers for the sample size. So, we multiply the floating-point data we have by 32767, the maximum value for a short integer.

It is theoretically possible to use the floating point -1.0 to 1.0 data directly in a WAV file, but not obvious how to do that using the wave module in Python.

Generating Spectograms

I have tried two methods of doing this and both were just fine. I however opted out to use the SoX - Sound eXchange, the Swiss Army knife of audio manipulation one because it didn't require anything else.

sox output.wav -n spectrogram -o spectrogram.png

An example spectrogram of Ludwig van Beethoven Symphony No. 6 First movement.

The other option could also be in combination with gnuplot. This would require an intermediary step, however.

sox output.wav audio.dat
tail -n+3 audio.dat > audio_only.dat
gnuplot audio.gpi

And input file audio.gpi that would be passed to gnuplot looks something like this.

# set output format and size
set term png size 1000,280

# set output file
set output "audio.png"

# set y range
set yr [-1:1]

# we want just the data
unset key
unset tics
unset border
set lmargin 0
set rmargin 0
set tmargin 0
set bmargin 0

# draw rectangle to change background color
set obj 1 rectangle behind from screen 0,0 to screen 1,1
set obj 1 fillstyle solid 1.0 fillcolor rgbcolor "#ffffff"

# draw data with foreground color
plot "audio_only.dat" with lines lt rgb 'red'

Pre-generated sequences

What I did was take interesting parts from an animal's genome and feed it to a tone generator script. This then generated a WAV file and I converted those to MP3, so they can be played in a browser. The last step was creating a spectrogram based on a WAV file.

Niels Bohr quote

Mouse

This is part of a mouse genome Mus_musculus.GRCm39.dna.nonchromosomal. You can get genom data here.

Bison

This is part of a bison genome Bison_bison_bison.Bison_UMD1.0.cdna. You can get genom data here.

Taurus

This is part of a taurus genome Bos_taurus.ARS-UCD1.2.cdna. You can get genom data here.

Making a drummer out of a DNA sequence

To make things even more interesting, I decided to send this data via MIDI to my Elektron Model:Samples. This is a really cool piece of equipment that supports MIDI in via USB and 3.5 mm audio jack.

Elektron is connected to my MacBook via USB cable and audio out is patched to a Sony Bluetooth speaker I have that supports 3.5 mm audio in. Elektron doesn't have internal speakers.

For communicating with Elektron, I choose pygame Python module that has MIDI built in. With this, it was rather simple to send notes to the device. All I did was map MIDI notes to the actual Nucleotides.

Before all of this I also checked Audio MIDI Setup app under MacOS and checked MIDI Studio by pressing ⌘-2.

The whole script that parses and send notes to the Elektron looks like this.

import pygame.midi
import time

pygame.midi.init()

print(pygame.midi.get_default_output_id())
print(pygame.midi.get_device_info(0))

player = pygame.midi.Output(1)
player.set_instrument(2)

def send_note(note, velocity):
  global player
  player.note_on(note, velocity)
  time.sleep(0.3)
  player.note_off(note, velocity)


nucleotide_midi_map = {
  'A': 60,
  'C': 90,
  'G': 160,
  'T': 180,  # is D
}

with open("quote.fa") as f:
  sequence = f.read().replace('\n', '')

for nucleotide in [char for char in sequence]:
  print("Playing nucleotide {} with MIDI note {}".format(
      nucleotide, nucleotide_midi_map[nucleotide]))
  send_note(nucleotide_midi_map[nucleotide], 127)

del player
pygame.midi.quit()

All of this could be made much more interesting if I choose different instruments for different Nucleotides, or doing more funky stuff with Elektron. But for now, this should be enough. It is just a proof of concept. Something to play around with.

Going even further

As you probably notice, the end results are quite similar to each other. This is to be expected because we are operating only with 4 notes essentially. What could make this more interesting is using something like Supercollider to create more interesting sounds. By transposing notes or using effects based on repeated data in a sequence. Possibilities are endless.

It is really astonishing what can be achieved with a little bit of code and an idea. I could see this becoming an interesting background soundscape instrument if done properly. It could replace random note generator with something more intriguing, biological, natural.

I actually find the results fascinating. I took some time and listened to this music of nature. Even though it's quite the same, it's also quite different. The subtle differences on repeat kind of creates music on its own. Makes you wonder. It kind of puts Occam’s Razor in its place. Nature for sure loves to make things as energy efficient as possible.

Wireless Application Protocol and the mobile web before the web

Thu, 30 Dec 2021 12:00:00 +0200

A little stroll down the history lane

About two weeks ago, I watched this outstanding documentary on YouTube Springboard: the secret history of the first real smartphone about the history of smartphones and phones in general. It brought back so many memories. I never had an actual smartphone before the Android. The closest to smartphone was Sony Ericsson P1. A fantastic phone and I broke it in Prague after a party and that was one of those rare occasions where I was actually mad at myself. But nevertheless, after that phone, the next one was an Android one.

Before that, I only owned normal phones from Nokia and Siemens etc. Nothing special, actually. These are the phones we are talking about. Before 2007. Apple and Android phones didn't exist yet.

These phones were rocking:

No selfie cameras.
~2 inch displays.
~120 MHz beast CPU's.
144p main cameras.
But they had a headphone jack.

Let's take a look at these beauties.

WAP - Wireless Application Protocol

Not that one! We are talking about Wireless Application Protocol and not Cardi B's song 😃

WAP stands for Wireless Application Protocol. It is a protocol designed for micro-browsers, and it enables the access of internet in the mobile devices. It uses the mark-up language WML (Wireless Markup Language and not HTML), WML is defined as XML 1.0 application. Furthermore, it enables creating web applications for mobile devices. In 1998, WAP Forum was founded by Ericson, Motorola, Nokia and Unwired Planet whose aim was to standardize the various wireless technologies via protocols. (source)

WAP protocol was resulted by the joint efforts of the various members of WAP Forum. In 2002, WAP forum was merged with various other forums of the industry, resulting in the formation of Open Mobile Alliance (OMA). (source)

These were some wild times. Devices had tiny screens and data transmission rates were abominable. But they were capable of rendering WML (Wireless Markup Language). This was very similar to HTML, actually. It is a markup language, after all.

These pages could be served by Apache and could be generated by CGI scripts on the backend. The only difference was the limited markup language.

WML - Wireless Markup Language

Just like web browsers use HTML for content structure, older mobile device browsers use WML - if you need to support really old mobile phones using WML browsers, you will need to know about it. WML is XML-based (an XML vocabulary just like XHTML and MathML, but not HTML) and does not use the same metaphor as HTML. HTML is a single document with some metadata packed away in the head, and a body encapsulating the visible page. With WML, the metaphor does not envisage a page, but rather a deck of cards. A WML file might have several pages or cards contained within it. (source)

<?xml version="1.0"?>
<!DOCTYPE wml PUBLIC "-//WAPFORUM//DTD WML 1.1//EN" "http://www.wapforum.org/DTD/wml_1.1.xml">
<wml>
  <card id="home" title="Example Homepage">
    <p>Welcome to the Example homepage</p>
  </card>
</wml>

There is an amazing tutorial on Tutorialpoint about WML.

Converting Digg to WML

This task is completely useless and not really feasible nowadays, but I had to give it a try for old-time sake. Since the data is already there in a form of RSS feed, I could take this feed and parse it and create a WML version of the homepage.

We will need:

Python3 + Pip
ImageMagick
feedparser and mako templating

# for fedora 35
sudo dnf install ImageMagick python3-pip

# tempalting engine for python
pip install mako --user

# for parsing rss feeds
pip install feedparser --user

Project folder structure should look like the following.

12:43:53 m@khan wap → tree -L 1
.
├── generate.py
└── template.wml

After that, I created a small template for the homepage.

<?xml version="1.0"?>
<!DOCTYPE wml PUBLIC "-//WAPFORUM//DTD WML 1.2//EN" "http://www.wapforum.org/DTD/wml_1.2.xml">

<wml>

  <card title="Digg - What the Internet is talking about right now">

    % for item in entries:
      <p><img src="/images/${item.id}.jpg" width="175" height="95" alt="${item.title}" /></p>
      <p><small>${item.kicker}</small></p>
      <p><big><b>${item.title}</b></big></p>
      <p>${item.description}</p>
    % endfor

  </card>

</wml>

And the parser that parses RSS feed looks like this.

import os
import feedparser
from mako.template import Template

os.system('mkdir -p www/images')

template = Template(filename='template.wml')

feed = feedparser.parse('https://digg.com/rss/top.xml')

entries = feed.entries[:15]

for entry in entries:
  print('Processing image with id {}'.format(entry.id))
  os.system('wget -q -O www/images/{}.jpg "{}"'.format(entry.id, entry.links[1].href))
  os.system('convert www/images/{}.jpg -type Grayscale -resize 175x -depth 3 -quality 30 www/images/{}.jpg'.format(entry.id, entry.id))

html = template.render(entries = entries)

with open('www/index.wml', 'w+') as fp:
  fp.write(html)

This script will create a folder www and in the folder www/images for storing resized images.

Be sure you don't use SSL and use just normal HTTP for serving the content. These old phones will have problems with TLS 1.3 etc.

If you look at the python file, I convert all the images into tiny B&W images. They should be WBMP (Wireless BitMaP) but I choose JPEGs for this, and it seems to work properly.

Because I currently don't have a phone old enough to test it on, I used an emulator. And it was really hard to find one. I found WAP Proof on shareware junction, and it did the job well enough. I will try to find and actual device to test it on.

If you are using Nginx to serve the contents, add a directive to the hosts file that will automatically server index.wml file.

server {
  index index.wml index.html index.htm index.nginx-debian.html;
}

Conclusion

Well, this was pointless, but very fun! I hope you enjoyed it as much as I did. I will try to find an old phone to test it on. If you have any questions, feel free to ask in the comments.

Running Golang application as PID 1 with Linux kernel

Sat, 25 Dec 2021 12:00:00 +0200

Unikernels, kernels, and alike

I have been reading a lot about unikernernels lately and found them very intriguing. When you push away all the marketing speak and look at the idea, it makes a lot of sense.

A unikernel is a specialized, single address space machine image constructed by using library operating systems. (Wikipedia)

I really like the explanation from the article Unikernels: Rise of the Virtual Library Operating System. Really worth a read.

If we compare a normal operating system to a unikernel side by side, they would look something like this.

From this image, we can see how the complexity significantly decreases with the use of Unikernels. This comes with a price, of course. Unikernels are hard to get running and require a lot of work since you don't have an actual proper kernel running in the background providing network access and drivers etc.

So as a half step to make the stack simpler, I started looking into using Linux kernel as a base and going from there. I came across this Youtube video talking about Building the Simplest Possible Linux System by Rob Landley and apart from statically compiling the application to be run as PID1 there was really no other obstacles.

What is PID 1?

PID 1 is the first process that Linux kernel starts after the boot process. It also has a couple of unique properties that are unique to it.

When the process with PID 1 dies for any reason, all other processes are killed with KILL signal.
When any process having children dies for any reason, its children are re-parented to process with PID 1.
Many signals which have default action of Term do not have one for PID 1.
When the process with PID 1 dies for any reason, kernel panics, which result in system crash.

PID 1 is considered as an Init application which takes care of running other and handling services like:

sshd,
nginx,
pulseaudio,
etc.

If you are on a Linux machine, you can check what your process is with PID 1 by running the following.

$ cat /proc/1/status
Name:	systemd
Umask:	0000
State:	S (sleeping)
Tgid:	1
Ngid:	0
Pid:	1
PPid:	0
...

As we can see on my machine the process with id of 1 is systemd which is a software suite that provides an array of system components for Linux operating systems. If you look closely you can also see that the PPid (process id of the parent process) is 0 which additionally confirms that this process doesn't have a parent.

So why even run application as PID 1 instead of just using a container?

Containers are wonderful, but they come with a lot of baggage. And because they are in their nature layered, the images require quite a lot of space and also a lot of additional software to handle them. They are not as lightweight as they seem, and many popular images require 500 MB plus disk space.

The idea of running this as PID 1 would result in a significantly smaller footprint, as we will see later in the post.

You could run a simple init system inside Docker container described more in this article Docker and the PID 1 zombie reaping problem.

The master plan

Compile Linux kernel with the default definitions.
Prepare a Hello World application in Golang that is statically compiled.
Run it with QEMU and providing Golang application as init application / PID 1.

For the sake of simplicity we will not be cross-compiling any of it and just use the 64bit version.

Compiling Linux kernel

$ wget https://cdn.kernel.org/pub/linux/kernel/v5.x/linux-5.15.7.tar.xz
$ tar xf linux-5.15.7.tar.xz

$ cd linux-5.15.7

$ make clean

# read more about this https://stackoverflow.com/a/41886394
$ make defconfig

$ time make -j `nproc`

$ cd ..

At this point we have kernel image that is located in arch/x86_64/boot/bzImage. We will use this in QEMU later.

To make our lives a bit easier lets move the kernel image to another place. Lets create a folder bin/ in the root of our project with mkdir -p bin.

At this point we can copy bzImage to bin/ folder with cp linux-5.15.7/arch/x86_64/boot/bzImage bin/bzImage.

The folder structure of this experiment should look like this.

pid1/
  bin/
    bzImage
  linux-5.15.7/
  linux-5.15.7.tar.xz

Preparing PID 1 application in Golang

This step is relatively easy. The only thing we must have in mind that we will need to compile the binary as a static one.

Let's create init.go file in the root of the project.

package main

import (
  "fmt"
  "time"
)

func main() {
  for {
    fmt.Println("Hello from Golang")
    time.Sleep(1 * time.Second)
  }
}

If you notice, we have a forever loop in the main, with a simple sleep of 1 second to not overwhelm the CPU. This is because PID 1 should never complete and/or exit. That would result in a kernel panic. Which is BAD!

There are two ways of compiling Golang application. Statically and dynamically.

To statically compile the binary, use the following command.

$ go build -ldflags="-extldflags=-static" init.go

We can also check if the binary is statically compiled with:

$ file init
init: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), statically linked, Go BuildID=Ypu8Zw_4NBxm1Yxg2OYO/H5x721rQ9uTPiDVh-VqP/vZN7kXfGG1zhX_qdHMgH/9vBfmK81tFrygfOXDEOo, not stripped

$ ldd init
not a dynamic executable

At this point, we need to create initramfs (abbreviated from "initial RAM file system", is the successor of initrd. It is a cpio archive of the initial file system that gets loaded into memory during the Linux startup process).

$ echo init | cpio -o --format=newc > initramfs
$ mv initramfs bin/initramfs

The projects at this stage should look like this.

pid1/
  bin/
    bzImage
    initramfs
  linux-5.15.7/
  linux-5.15.7.tar.xz
  init.go

Running all of it with QEMU

QEMU is a free and open-source hypervisor. It emulates the machine's processor through dynamic binary translation and provides a set of different hardware and device models for the machine, enabling it to run a variety of guest operating systems.

$ qemu-system-x86_64 -serial stdio -kernel bin/bzImage -initrd bin/initramfs -append "console=ttyS0" -m 128

$ qemu-system-x86_64 -serial stdio -kernel bin/bzImage -initrd bin/initramfs -append "console=ttyS0" -m 128
[    0.000000] Linux version 5.15.7 (m@khan) (gcc (GCC) 11.2.1 20211203 (Red Hat 11.2.1-7), GNU ld version 2.37-10.fc35) #7 SMP Mon Dec 13 10:23:25 CET 2021
[    0.000000] Command line: console=ttyS0
[    0.000000] x86/fpu: x87 FPU will use FXSAVE
[    0.000000] signal: max sigframe size: 1440
[    0.000000] BIOS-provided physical RAM map:
[    0.000000] BIOS-e820: [mem 0x0000000000000000-0x000000000009fbff] usable
[    0.000000] BIOS-e820: [mem 0x000000000009fc00-0x000000000009ffff] reserved
[    0.000000] BIOS-e820: [mem 0x00000000000f0000-0x00000000000fffff] reserved
[    0.000000] BIOS-e820: [mem 0x0000000000100000-0x0000000007fdffff] usable
[    0.000000] BIOS-e820: [mem 0x0000000007fe0000-0x0000000007ffffff] reserved
[    0.000000] BIOS-e820: [mem 0x00000000fffc0000-0x00000000ffffffff] reserved
[    0.000000] NX (Execute Disable) protection: active
[    0.000000] SMBIOS 2.8 present.
[    0.000000] DMI: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-6.fc35 04/01/2014
[    0.000000] tsc: Fast TSC calibration failed
...
[    2.016106] ALSA device list:
[    2.016329]   No soundcards found.
[    2.053176] Freeing unused kernel image (initmem) memory: 1368K
[    2.056095] Write protecting the kernel read-only data: 20480k
[    2.058248] Freeing unused kernel image (text/rodata gap) memory: 2032K
[    2.058811] Freeing unused kernel image (rodata/data gap) memory: 500K
[    2.059164] Run /init as init process
Hello from Golang
[    2.386879] tsc: Refined TSC clocksource calibration: 3192.032 MHz
[    2.387114] clocksource: tsc: mask: 0xffffffffffffffff max_cycles: 0x2e02e31fa14, max_idle_ns: 440795264947 ns
[    2.387380] clocksource: Switched to clocksource tsc
[    2.587895] input: ImExPS/2 Generic Explorer Mouse as /devices/platform/i8042/serio1/input/input3
Hello from Golang
Hello from Golang
Hello from Golang

The whole log file here.

Size comparison

The cool thing about this approach is that the Linux kernel and the application together only take around 12 MB, which is impressive as hell. And we need to also know that the size of bzImage (Linux kernel) could be greatly decreased by going into make menuconfig and removing a ton of features from the kernel, making the size even smaller. I managed to get kernel size down to 2 MB and still working properly.

total 12M
-rw-r--r--. 1 m m 9.3M Dec 13 10:24 bzImage
-rw-r--r--. 1 m m 1.9M Dec 27 01:19 initramfs

Creating ISO image and running it with Gnome Boxes

First we need to create proper folder structure with mkdir -p iso/boot/grub.

Then we need to download the grub binary. You can read more about this program on https://github.com/littleosbook/littleosbook.

$ wget -O iso/boot/grub/stage2_eltorito https://github.com/littleosbook/littleosbook/raw/master/files/stage2_eltorito

$ tree iso/boot/
iso/boot/
├── bzImage
├── grub
│   ├── menu.lst
│   └── stage2_eltorito
└── initramfs

Let's copy files into proper folders.

$ cp stage2_eltorito iso/boot/grub/
$ cp bin/bzImage iso/boot/
$ cp bin/initramfs iso/boot/

Lets create a GRUB config file at nano iso/boot/grub/menu.lst with contents.

default=0
timeout=5

title GoAsPID1
kernel /boot/bzImage
initrd /boot/initramfs

Let's create iso file by using genisoimage:

genisoimage -R                              \
            -b boot/grub/stage2_eltorito    \
            -no-emul-boot                   \
            -boot-load-size 4               \
            -A os                           \
            -input-charset utf8             \
            -quiet                          \
            -boot-info-table                \
            -o GoAsPID1.iso                 \
            iso

This will produce GoAsPID1.iso which you can use with Virtualbox or Gnome Boxes.

Is running applications as PID 1 even worth it?

Well, the answer to this is not as simple as one would think. Sometimes it is and sometimes it's not. For embedded systems and very specialized applications it is worth for sure. But in normal uses, I don't think so. It was an interesting exercise in compiling kernels and looking at the guts of the Linux kernel, but sticking to containers for most of the things is a better option in my opinion.

An interesting experiment would be creating an image that supports networking and could be deployed to AWS as an EC2 instance and observing how it fares. But in that case, we would need to write some sort of supervisor that would run on a separate EC2 that would check if other EC2 instances are running properly. Remember that if your application fails, kernel panics and the whole machine is inoperable in this case.

Debian based riced up distribution for Developers and DevOps folks

Fri, 03 Dec 2021 12:00:00 +0200

Introduction

I have been using Ubuntu for quite a longtime now. I have used Debian in the past and Manjaro. Also had Arch for some time and even ran Gentoo way back.

What I learned from all this is that I prefer running a bit older versions and having them be stable than run bleeding edge rolling release. For that reason, I stuck with Ubuntu for a couple of years now. I am also at a point in my life where I just don't care what is cool or hip anymore. I just want a stable system that doesn't get in my way.

During all this, I noticed that these distributions were getting very bloated and a lot of software got included that I usually uninstall on fresh installation. Maybe this is my OCD speaking, but why do I have to give fresh installation min 1 GB of ram out of the box just to have a blank screen in front of me? I get it, there are many things included in the distro to make my life easier. I understand. But at this point I have a feeling that modern Linux distributions are becoming similar to Node.js project with node_modules. Just a crazy number of packages serving very little or no purpose, just supporting other software.

I felt I needed a fresh start. To start over with something minimal and clean. Something that would put a little more joy into using a computer again.

For the first version, I wanted to target the following machines I have at home that I want this thing to work on.

# My main stationary work machine
Resolution: 3840x1080 (Super Ultrawide Monitor 32:9)
CPU: Intel i7-8700 (12) @ 4.600GHz
GPU: AMD ATI Radeon RX 470/480/570/570X/580/580X/590
Memory: 32020MiB

# Thinkpad x220 for testing things and goofing around
Resolution: 1366x768
CPU: Intel i5-2520M (4) @ 3.200GHz
GPU: Intel 2nd Generation Core Processor Family
Memory: 15891MiB

How should I approach this?

I knew I wanted to use minimal Debian netinst for the base to give myself a head start. No reason to go through changing the installer and also testing all that behemoth of a thing. So, some sort of ricing was the only logical option to get this thing of the grounds somewhat quickly.

What is ricing anyway? The term “RICE” stands for Race Inspired Cosmetic Enhancement. A group of people (could be one, idk) decided to see if they could tweak their own distros like they/others did their cars. This gave rise to a community of Linux/Unix enthusiasts trying to make their distros look cooler and better than others... For more information, read this article What in the world is ricing!?.

I didn't want this to just be a set of config files for theming purpose. I wanted this to include a set of pre-installed tools and services that are being used all the time by a modern developer. Theming is just a tiny part of it. Fonts being applied across the distro and things like that.

First, I choose terminal installer and left it to load additional components. Avoid using graphical installer in this case.

After that I selected hostname and created a normal user and set password for that user and root user and choose guided mode for disk partitioning.

I left it run to install all the things required for the base system and opted out of scanning additional media for use by the package manager. Those will be downloaded from the internet during installation.

I opted out of the popularity contest, and now comes the important part. Uncheck all the boxes in Software selection and only leave 'standard system utilities'. I also left an SSH server, so I was able to log in to the machine from my main PC.

At this point, I installed GRUB bootloader on the disk where I installed the system.

That concluded the installation of base Debian and after restarting the computer I was prompted with the login screen.

Now that I had the base installation, it was time to choose what software do I want to include in this so-called distribution. I wanted out of the box developer experience, so I had plenty to choose.

Let's not waste time and go through the list.

Desktop environments

I have been using Gnome for my whole Linux life. From version 2 forward. It's been quite a ride. I hated version 3 when it came out and replaced version 2. But I got used to it. And now with version 40+ they also made couple of changes which I found both frustrating and presently surprised.

The amount of vertical space you loose because of the beefy title bars on windows is ridiculous. And then in case of Tilix you also have tabs, and you are 100px deep. Vertical space is one of the most important things for a developer. The more real estate you have, the more code you can have in a viewport.

But on the other hand, I still love how Gnome feels and looks. I gotta give them that. They really are trying to make Gnome feel unified and modern.

Regardless of all the nice things Gnome has, I was looking at the tiling window managers for some time, but never had the nerve to actually go with it. But now was the ideal time to give it a go. No guts, no glory kind of a thing.

One of the requirements for me was easy custom layouts because I use a really strange monitor with aspect ratio of 32:9. So relying on included layouts most of them have is a non-starter.

What I was doing in Gnome was having windows in a layout like the diagram below. This is my common practice. And if you look at it you can clearly see I was replicating tiling window manager setup in Gnome.

That made me look into a bunch of tiling window managers and then tested them out. Candidates I was looking at were:

You can also check article 13 Best Tiling Window Managers for Linux I was referencing while testing them out.

While all of them provided what I needed, I liked i3 the most. What particular caught my eye was the ease to use and tree based layouts which allows flexible layouts. I know others can be set up also to have custom layouts other than
spiral, dwindle etc. I think i3 is a good entry-level window manager for somebody like me.

Batteries included

The source for the whole thing is located on Github https://github.com/mitjafelicijan/dfd-rice.

Currenly included:

non-free (enables non-free packages in apt)
sudo (adds sudo and adds user to sudo group)
essentials (gcc, htop, zip, curl, etc...)
wifi (network manager nmtui)
desktop (i3, dmenu, fonts, configurations)
pulseaudio (pulseaudio with pavucontrol)
code-editors (vim, micro, vscode)
ohmybash (make bash pretty)
file-managers (mc)
git-ui (terminal git gui)
meld (diff tool)
profiling (kcachegrind, valgrind, strace, ltrace)
browsers (brave, firefox, chromium)
programming languages:
- python
- golang
- nodejs
- rust
- nim
- php
- ruby
docker (with docker-compose)
ansible

Install script also allows you to install only specific packages (example for: essentials ohmybash docker rust).

su - root \
    bash -c "$(wget -q https://raw.github.com/mitjafelicijan/dfd-rice/master/tools/install.sh -O -)" -- \
    essentials ohmybash docker rust

Currently, most of these recipes use what Debian and this is totally fine with me since I never use bleeding edge features of a package. But if something major would come to light, I will replace it with a possible compilation script or something similar.

This is some of the output from the installation script.

Let's take a look at some examples in the installation script.

Docker recipe

# docker
print_header "Installing Docker"
curl -fsSL https://download.docker.com/linux/debian/gpg | gpg --yes --dearmor -o /usr/share/keyrings/docker-archive-keyring.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/docker-archive-keyring.gpg] https://download.docker.com/linux/debian $(lsb_release -cs) stable" | tee /etc/apt/sources.list.d/docker.list > /dev/null
apt update
apt -y install docker-ce docker-ce-cli containerd.io docker-compose

systemctl start docker
systemctl enable docker
systemctl status docker --no-pager

/sbin/usermod -aG docker $USERNAME

Making bash pretty

I really like Oh My Zsh, but I don't like zsh shell. When I used it, I constantly needed to be aware of it and running bash scripts was a pain. So, I was really delighted when I found out that a version for bash existed called Oh My Bash. Let's take a look at the recipe for installing it.

# ohmybash
print_header "Enabling OhMyBash"
sudo -u $USERNAME sh -c "$(curl -fsSL https://raw.github.com/ohmybash/oh-my-bash/master/tools/install.sh)" &
T1=${!}
wait ${T1}

Because OhMyBash does exec bash at the end, this traps our script inside another shell and our script cannot continue. For that reason, I executed this in background. But that presents a new problem. Because this is executed in background, we lose track of progress naturally. And that strange trick with T1=${!} and wait ${T1} waits for the background process to finish before continuing to another task in bash script.

Check Multi-Threaded Processing in Bash Scripts for more details.

Conclusion

Take a look at https://github.com/mitjafelicijan/dfd-rice/blob/develop/tools/install.sh script to get familiar with it. This is just a first iteration and I will continue to update it because I need this in my life.

The current version boots in 4s to the login prompt, and after you log in, the desktop environment loads in 2s. So, its fast, very fast. And on clean boot, I measured ~230 MB of RAM usage.

And this is how it looks with two terminals side by side. I really like the simplicity and clean interface. I will polish the colors and stuff like that, but I really do like the results.

List of essential Linux commands for server management

Sun, 01 Aug 2021 12:00:00 +0200

Generate SSH key

ssh-keygen -t ed25519 -C "your_email@example.com"

# when no support for Ed25519 present
ssh-keygen -t rsa -b 4096 -C "your_email@example.com"

Note: By default SSH keys get stored to /home/<username>/.ssh/ folder.

Login to host via SSH

# connect to host as your local username
ssh host

# connect to host as user
ssh <user>@<host>

# connect to host using port
ssh -p <port> <user>@<host>

Execute command on a server through SSH

# execute one command
ssh root@100.100.100.100 "ls /root"

# execute many commands
ssh root@100.100.100.100 "cd /root;touch file.txt"

Displays currently logged in users in the system

Displays Linux system information

uname

Displays kernel release information

uname -r

Shows the system hostname

hostname

Shows system reboot history

last reboot

Displays information about the user

sudo apt install finger
finger <username>

Displays IP addresses and all the network interfaces

ip addr show

Downloads a file from an online source

wget https://example.com/example.tgz

Note: If URL contains ?, & enclose the URL in double quotes.

Compress a file with gzip

# will not keep the original file
gzip file.txt

# will keep the original file
gzip --keep file.txt

Interactive disk usage analyzer

sudo apt install ncdu

ncdu
ncdu <path/to/directory>

Install Node.js using the Node Version Manager

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.35.3/install.sh | bash
source ~/.bashrc

nvm install v13

Too long; didn't read

npm install -g tldr

tldr tar

Combine all Nginx access logs to one big log file

zcat -f /var/log/nginx/access.log* > /var/log/nginx/access-all.log

Set up Redis server

sudo apt install redis-server redis-tools

# check if server is running
sudo service redis status

# set and get a key value
redis-cli set mykey myvalue
redis-cli get mykey

# interactive shell
redis-cli

Generate statistics of your webserver

sudo apt install goaccess

# check if installed
goaccess -v

# combine logs
zcat -f /var/log/nginx/access.log* > /var/log/nginx/access-all.log

# export to single html
goaccess \
  --log-file=/var/log/nginx/access-all.log \
  --log-format=COMBINED \
  --exclude-ip=0.0.0.0 \
  --ignore-crawlers \
  --real-os \
  --output=/var/www/html/stats.html

# cleanup afterwards
rm /var/log/nginx/access-all.log

Search for a given pattern in files

grep -r ‘pattern’ files

Find proccess ID for a specific program

pgrep nginx

Print name of current/working directory

pwd

Creates a blank new file

touch newfile.txt

Displays first lines in a file

# -n <x> presents the number of lines (10 by default)
head -n 20 somefile.txt

Displays last lines in a file

# -n <x> presents the number of lines (10 by default)
tail -n 20 somefile.txt

# -f follows the changes in file (doesn't closes)
tail -f somefile.txt

Count lines in a file

wc -l somefile.txt

Find all instances of the file

sudo apt install mlocate

locate somefile.txt

Find file names that begin with ‘index’ in /home folder

find /home/ -name "index"

Find files larger than 100MB in the home folder

find /home -size +100M

Displays block devices related information

lsblk

Displays free space on mounted systems

df -h

Displays free and used memory in the system

free -h

Displays all active listening ports

sudo apt install net-tools

netstat -pnltu

Kill a process violently

kill -9 <pid>

List files opened by user

lsof -u <user>

Execute "df -h", showing periodic updates

# -n 1 means every second
watch -n 1 df -h

Simple world clock with eInk display and Raspberry Pi Zero

Sat, 26 Jun 2021 12:00:00 +0200

Our team is spread across the world, from the USA all the way to Australia, so having some sort of world clock makes sense.

Currently, I am using an extension for Gnome called Timezone extension, and it serves the purpose quite well.

But I also have a bunch of electronics that I bought through the time, and I am not using any of them, and it's time to stop hording this stuff and use it in a project.

A while ago I bought a small eInk display Inky pHAT and I have a bunch of Raspberry Pi's Zero lying around that I really need to use.

Since the Inky Inky pHAT is essentially a HAT, it can easily be added on top of the Raspberry Pi Zero.

First, I installed the necessary software on Raspberry Pi with pip3 install inky.

And then I created a file clock.py in home directory /home/pi.

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import sys
import os
from inky.auto import auto
from PIL import Image, ImageFont, ImageDraw
from font_fredoka_one import FredokaOne

clocks = [
  'America/New_York',
  'Europe/Ljubljana',
  'Australia/Brisbane',
]

board = auto()
board.set_border(board.WHITE)
board.rotation = 90

img = Image.new('P', (board.WIDTH, board.HEIGHT))
draw = ImageDraw.Draw(img)

big_font = ImageFont.truetype(FredokaOne, 18)
small_font = ImageFont.truetype(FredokaOne, 13)

x = board.WIDTH / 3
y = board.HEIGHT / 3

idx = 1
for clock in clocks:
  ctime = os.popen('TZ="{}" date +"%a,%H:%M"'.format(clock))
  ctime = ctime.read().strip().split(',')
  city = clock.split('/')[1].replace('_', ' ')

  draw.text((15, (idx*y)-y+10), city, fill=board.BLACK, font=small_font)
  draw.text((110, (idx*y)-y+7), str(ctime[0]), fill=board.BLACK, font=big_font)
  draw.text((155, (idx*y)-y+7), str(ctime[1]), fill=board.BLACK, font=big_font)

  idx += 1

board.set_image(img)
board.show()

And because eInk displays are rather slow to refresh and the clock requires refreshing only once a minute, this can be done through cronjob.

Before we add this job to cron we need to make clock.py executable with chmod +x clock.py.

Then we add a cronjob with crontab -e.

* * * * * /home/pi/clock.py

So, we end up with a result like this.

You can download my STL file for the enclosure here, but make sure that dimensions make sense and also opening for USB port should be added or just use a drill and some hot glue to make it stick in the enclosure.

Using GoAccess with Nginx to replace Google Analytics

Mon, 25 Jan 2021 12:00:00 +0200

Introduction

I know! You cannot simply replace Google Analytics with parsing access logs and displaying a couple of charts. But to be honest, I actually never used Google Analytics to the fullest extent and was usually interested in seeing page hits and which pages were visited most often.

I recently moved my blog from Firebase to a VPS and also decided to remove Google Analytics tracking code from the site since its quite malicious and tracks users across other pages also and is creating a profile of a user, and I've had it. But I also need some insight of what is happening on a server and which content is being read the most etc.

I have looked at many existing solutions like:

But the more I looked at them the more I noticed that I am replacing one evil with another one. Don't get me wrong. Some of these solutions are absolutely fantastic but would require installation of databases and something like PHP or Node. And I was not ready to put those things on my fresh server. Also having Docker installed is out of the question.

Opting for log parsing

So, I defaulted to parsing already existing logs and generating HTML reports from this data.

I found this amazing software GoAccess which provides all the functionalities I need, and it's a single binary. Written in Go.

GoAccess can be used in two different modes.

Running in a terminal

Running in a browser

I, however, need this to run in a browser. So, the second option is the way to go. The Idea is to periodically run cronjob and export this report into a folder that gets then server by Nginx behind a Basic authentication.

Getting Nginx ready

I choose Ubuntu on DigitalOcean. First I installed Nginx, and Letsencrypt certbot and all the necessary dependencies.

# log in as root user
sudo su -

# first let's update the system
apt update && apt upgrade -y

# let's install
apt install nginx certbot python3-certbot-nginx apache2-utils

After all this is installed we can create a new configuration for a statistics. Stats will be available at stats.domain.com.

# creates directory where html will be hosted
mkdir -p /var/www/html/stats.domain.com

cp /etc/nginx/sites-available/default /etc/nginx/sites-available/stats.domain.com
nano /etc/nginx/sites-available/stats.domain.com

server {
  root /var/www/html/stats.domain.com;
  server_name stats.domain.com;

  index index.html;
  location / {
    try_files $uri $uri/ =404;
  }
}

Now we check if the configuration is ok. We can do this with nginx -t. If all is ok, we can restart Nginx with service nginx restart.

After all that you should add A record for this domain that points to IP of a droplet.

Before enabling SSL you should test if DNS records have propagated with curl stats.domain.com.

Now, it's time to provision TLS certificate. To achieve this, you execute command certbot --nginx. Follow the wizard and when you are asked about redirection always choose 2 (always redirect to HTTPS).

When this is done you can visit https://stats.domain.com and you should get 404 not found error which is correct.

Getting GoAccess ready

If you are using Debian like system GoAccess should be available in repository. Otherwise refer to the official website.

apt install goaccess

To enable Geo location we also need one additiona thing.

cd /var/www/html/stats.stats.com
wget https://github.com/P3TERX/GeoLite.mmdb/raw/download/GeoLite2-City.mmdb

Now we create a shell script that will be executed every 10 minutes.

nano /var/www/html/stats.domain.com/generate-stats.sh

Contents of this file should look like this.

#!/bin/sh

zcat -f /var/log/nginx/access.log* > /var/log/nginx/access-all.log

goaccess \
  --log-file=/var/log/nginx/access-all.log \
  --log-format=COMBINED \
  --exclude-ip=0.0.0.0 \
  --geoip-database=/var/www/html/stats.domain.com/GeoLite2-City.mmdb \
  --ignore-crawlers \
  --real-os \
  --output=/var/www/html/stats.domain.com/index.html

rm /var/log/nginx/access-all.log

Because after a while nginx creates multiple files with access logs we use zcat to extract Gziped contents and create a file that has all the access logs. After this file is used we delete it.

If you want to exclude your home IP's result look at the --exclude-ip option in script and instead of 0.0.0.0 add your own home IP address. You can find your home IP by executing curl ifconfig.me from your local machine and NOT from the droplet.

Test the script by executing sh /var/www/html/stats.domain.com/generate-stats.sh and then checking https://stats.domain.com. If you can see stats instead of 404 than you are set.

It's time to add this script to cron with cron -e.

*/10 * * * * sh /var/www/html/stats.domain.com/generate-stats.sh

Securing with Basic authentication

You probably don't want stats to be publicly available, so we should create a user and a password for Basic authentication.

First we create a password for a user stats with htpasswd -c /etc/nginx/.htpasswd stats.

Now we update config file with nano /etc/nginx/sites-available/stats.domain.com. You probably noticed that the file looks a bit different from before. This is because certbot added additional rules for SSL.

Your location portion the config file should now look like. You should add auth_basic and auth_basic_user_file lines to the file.

location / {
  try_files $uri $uri/ =404;
  auth_basic "Private Property";
  auth_basic_user_file /etc/nginx/.htpasswd;
}

Test if config is still ok with nginx -t and if it is you can restart Nginx with service nginx restart.

If you now visit https://stats.domain.com you should be prompted for username and password. If not, try reopening your browser.

That is all. You now have analytics for your server that gets refreshed every 10 minutes.

Replacing Dropbox in favor of DigitalOcean spaces

Sun, 24 Jan 2021 12:00:00 +0200

A few months ago I experimented with DigitalOcean spaces as my backup solution that could replace Dropbox eventually. That solution worked quite nicely, and I was amazed how smashing together a couple of existing solutions would work this fine.

I have been running that solution in the background for a couple of months now and kind of forgot about it. But recent developments around deplatforming and having us people hostages of technology and big companies speed up my goals to become less dependent on Google, Dropbox etc and take back some control.

I am not a conspiracy theory nut, but to be honest, what these companies are doing lately is out of control. It is a matter of principle at this point. I have almost completely degoogled my life all the way from ditching Gmail, YouTube and most of the services surrounding Google. And I must tell you, I feel so good. I haven't felt this way for a long time.

Anyways. Let's get to the meat of things.

Before you continue you should read my post about syncing to Dropbox.

Also to note, I am using Linux on my machine with Gnome desktop environment. This should work on MacOS too. To use this on Windows I suggest using Subsystem for Linux or Cygwin.

Folder structure

I liked structure from Dropbox. One folder where everything is located and synced. So, that's why adopted this also for my sync setup.

~/Vault
   ↳ backup
   ↳ bin
   ↳ documents
   ↳ projects

All of my code is located in ~/Vault/projects folder. And most of the projects are Git repositories. I do not use this sync method for backup per see but in case I reinstall my machine I can easily recreate all the important folder structure with one quick command. No external drives needed that can fail etc.

Sync script

My sync script is located in ~/Vault/bin/vault-backup.sh

#!/bin/bash

# dconf load /com/gexperts/Tilix/ < tilix.dconf
# 0 2 * * * sh ~/Vault/bin/vault-backup.sh

cd ~/Vault/backup/dotfiles

MACHINE=$(whoami)@$(hostname)
mkdir -p $MACHINE
cd $MACHINE

cp ~/.config/VSCodium/User/settings.json settings.json
cp ~/.s3cfg s3cfg
cp ~/.bash_extended bash_extended
cp ~/.ssh ssh -rf

codium --list-extensions > vscode-extension.txt
dconf dump /com/gexperts/Tilix/ > tilix.dconf

cd ~/Vault
s3cmd sync --delete-removed --exclude 'node_modules/*' --exclude '.git/*' --exclude '.venv/*' ./ s3://bucket-name/backup/

echo `date +"%D %T"` >> ~/.vault.log

notify-send \
	-u normal \
	-i /usr/share/icons/Adwaita/96x96/status/security-medium-symbolic.symbolic.png \
	"Vault sync succeded at `date +"%D %T"`"

This script also backups some of the dotfiles I use and sends notification to Gnome notification center. It is a straightforward solution. Nothing special going on.

One obvious benefit of this is that I can omit syncing Node's node_modules or Python's .venv and .git folders.

You can use this script in a combination with Cron.

0 2 * * * sh ~/Vault/bin/vault-backup.sh

When you start syncing your local stuff with a remote server you can review your items on DigitalOcean.

I have been using this script now for quite some time, and it's working flawlessly. I also uninstalled Dropbox and stopped using it completely.

All I need to do is write a Bash script that does the reverse and downloads from remote server to local folder. This could be another post.

Using Digitalocean Spaces to sync between computers

Wed, 09 Sep 2020 12:00:00 +0200

I've been using Dropbox for probably 10+ years now and I-ve became so used to it that it runs in the background that I don't even imagine a world without it. But it's not without problems.

At first I had problems with .venv environments for Python and the only solution for excluding synchronization for this folder was to manually exclude a specific folder which is not really scalable. FYI, my whole project folder is synced on Dropbox. This of course introduced a lot of syncing of files and folders that are not needed or even break things on other machines. In the case of Python, I couldn't use that on my second machine. I needed to delete .venv folder and pip it again which synced files again to the main machine. This was very frustrating. Nodejs handles this much nicer and I can just run the scripts without deleting node_modules again and reinstalling. However, node_modules is a beast of its own. It creates so many files that OS has a problem counting them when you check the folder contents for size.

I wanted something similar to Dropbox. I could without the instant syncing but it would need to be fast and had the option for me to exclude folders like node_modules, .venv, .git and folders like that.

I went on a hunt for an alternative to Dropbox and found:

You know, the usual list of suspects. I didn't include Google drive or One drive since they are even more draconian than Dropbox.

All this does not stem from me being paranoid but recently these companies have became more and more aggressive and they keep violating our privacy when they share our data with 3rd party services. It is getting out of control.

So, my main problem was still there. No way of excluding a specific folder from syncing. And before we go into "But you have git, isn't that enough?", I must say, that many of the files (PDFs, spreadsheets, etc) I have in a git repo don't get pushed upstream to Git and I still want to have them synced across my computers.

I initially wanted to use rsync but I would need to then have a remote VPS or transfer between my computers directly. I wanted a solution where all my files could be accessible to me without my machine.

WARNING: This solution will cost you money! DigitalOcean Spaces are $5 per month and there are some bandwidth limitations and if you go beyond that you get billed additionally.

Then I remembered that I could use something like S3 since it has versioning and is fully managed. I didn't want to go down the AWS rabbit hole with this so I choose DigitalOcean Spaces.

Then I needed a command-line tool to sync between source and target. I found this nice tool s3cmd and it is in the Ubuntu repositories.

sudo apt install s3cmd

After installation will I create a new Space bucket on DigitalOcean. Remember the zone you will choose because you will need it when you will configure s3cmd.

Then I visited Digitalocean Applications & API and generated Spaces access keys. Save both key and secret somewhere safe because when you will leave the page secret will not be available anymore to you and you will need to re-generate it.

# enter your key and secret and correct endpoint
# my endpoint is ams3.digitaloceanspaces.com because
# I created my bucket in Amsterdam regiin
s3cmd --configure

After that I played around with options for s3cmd and got to the following command.

# I executed this command from my projects folder
cd projects
s3cmd sync --delete-removed --exclude 'node_modules/*' --exclude '.git/*' --exclude '.venv/*' ./ s3://my-bucket-name/projects/

When syncing int he other direction you will need to change the order of the SOURCE and TARGET to s3://my-bucket-name/projects/ and ./.

Be sure that all the paths have trailing slash so that sync knows that this are directories.

I am planning to implement some sort of a .ignore file that will enable me to have a project-specific exclude options.

I am currently running this every hour as a cronjob which is perfectly fine for now when I am testing how this whole thing works and how it all will turn out.

I have also created a small Gnome extension which is still very unstable, but when/if this whole experiment pays of I will share on Github.

Fix bind warning in .profile on login in Ubuntu

Tue, 08 Sep 2020 12:00:00 +0200

Recently I moved back to bash as my default shell. I was previously using fish and got used to the cool features it has. But, regardless of that, I wanted to move to a more standard shell because I was hopping back and forth with exporting variables and stuff like that which got pretty annoying.

So I embarked on a mission to make bash more like fish and in the process found that I really missed autosuggest with TAB on changing directories.

I found a nice alternative that emulates zsh like autosuggestion and autocomplete so I added the following to my .bashrc file.

bind "TAB:menu-complete"
bind "set show-all-if-ambiguous on"
bind "set completion-ignore-case on"
bind "set menu-complete-display-prefix on"
bind '"\e[Z":menu-complete-backward'

I haven't noticed anything wrong with this and all was working fine until I restarted my machine and then I got this error.

When I pressed OK, I got into the Gnome shell and all was working fine, but the error was still bugging me. I started looking for the reason why this is happening and found a solution to this error on Remote SSH Commands - bash bind warning: line editing not enabled.

So I added a simple if [ -t 1 ] around bind statements to avoid running commands that presume the session is interactive when it isn't.

if [ -t 1 ]; then
  bind "TAB:menu-complete"
  bind "set show-all-if-ambiguous on"
  bind "set completion-ignore-case on"
  bind "set menu-complete-display-prefix on"
  bind '"\e[Z":menu-complete-backward'
fi

After logging out and back in the problem was gone.

Getting started with MicroPython and ESP8266

Sun, 06 Sep 2020 12:00:00 +0200

Introduction

A while ago I bought some ESP8266 and ESP32 dev boards to play around with and I finally found a project to try it out.

For my project, I used ESP32 but I could easily choose ESP8266. This guide contains which tools I use and how I prepared my workspace to code for ESP8266.

This guide covers:

flashing SOC
install proper tooling
deploying a simple script

Make sure that you are using a good USB cable. I had some problems with mine and once I replaced it everything started to work.

Flashing the SOC

Plug your ESP8266 to USB port and check if the device was recognized with executing dmesg | grep ch341-uart.

Then check if the device is available under /dev/ by running ls /dev/ttyUSB*.

Linux users: if a device is not available be sure you are in dialout group. You can check this by executing groups $USER. You can add a user to dialout group with sudo adduser $USER dialout.

After these conditions are meet go to the navigate to https://micropython.org/download/esp8266/ and download esp8266-20200902-v1.13.bin.

mkdir esp8266-test
cd esp8266-test

wget https://micropython.org/resources/firmware/esp8266-20200902-v1.13.bin

After obtaining firmware we will need some tooling to flash the firmware to the board.

sudo pip3 install esptool

You can read more about esptool at https://github.com/espressif/esptool/.

Before flashing the firmware we need to erase the flash on device. Substitute USB0 with the device listed in output of ls /dev/ttyUSB*.

esptool.py --port /dev/ttyUSB0 erase_flash

If flash was successfully erased it is now time to flash the new firmware to it.

esptool.py --port /dev/ttyUSB0 --baud 460800 write_flash --flash_size=detect 0 esp8266-20200902-v1.13.bin

If everything went ok you can try accessing MicroPython REPL with screen /dev/ttyUSB0 115200 or picocom /dev/ttyUSB0 -b115200.

Sometimes you will need to press ENTER in screen or picocom to access REPL.

When you are in REPL you can test if all is working properly following steps.

> import machine
> machine.freq()

This should output a number representing a frequency of the CPU (mine was 80000000).

When you are in screen or picocom these can help you a bit.

Key	Command
CTRL+d	preforms soft reboot
CTRL+a x	exits picocom
CTRL+a \	exits screen

Install better tooling

Now, to make our lives a little bit easier there are couple of additional tools that will make this whole experience a little more bearable.

There are twq cool ways of uploading local files to SOC flash.

ampy → https://github.com/scientifichackers/ampy
rshell → https://github.com/dhylands/rshell

ampy

# installing ampy
sudo pip3 install adafruit-ampy

Listed below are some common commands I used.

# uploads file to flash
ampy --delay 2 --port /dev/ttyUSB0 put boot.py

# lists file on flash
ampy --delay 2 --port /dev/ttyUSB0 ls

# outputs contents of file on flash
ampy --delay 2 --port /dev/ttyUSB0 cat boot.py

I added delay of 2 seconds because I had problems with executing commands.

rshell

Even though ampy is a cool tool I opted with rshell in the end since it's much more polished and feature rich.

# installing ampy
sudo pip3 install rshell

Now that rshell is installed we can connect to the board.

rshell --buffer-size=30 -p /dev/ttyUSB0 -a

This will open a shell inside bash and from here you can execute multiple commands. You can check what is supported with help once you are inside of a shell.

m@turing ~/Junk/esp8266-test
$ rshell --buffer-size=30 -p /dev/ttyUSB0 -a

Using buffer-size of 30
Connecting to /dev/ttyUSB0 (buffer-size 30)...
Trying to connect to REPL  connected
Testing if ubinascii.unhexlify exists ... Y
Retrieving root directories ... /boot.py/
Setting time ... Sep 06, 2020 23:54:28
Evaluating board_name ... pyboard
Retrieving time epoch ... Jan 01, 2000
Welcome to rshell. Use Control-D (or the exit command) to exit rshell.
/home/m/Junk/esp8266-test> help

Documented commands (type help <topic>):
========================================
args    cat  connect  date  edit  filesize  help  mkdir  rm     shell
boards  cd   cp       echo  exit  filetype  ls    repl   rsync

Use Control-D (or the exit command) to exit rshell.

Inside a shell ls will display list of files on your machine. To get list of files on flash folder /pyboard is remapped inside the shell. To list files on flash you must perform ls /pyboard.

Moving files to flash

To avoid copying files all the time I used rsync function from the inside of rshell.

rsync . /pyboard

Executing scripts

It is a pain to continuously reboot the device to trigger /pyboard/boot.py and there is a better way of testing local scripts on remote device.

Lets assume we have src/freq.py file that displays CPU frequency of a remote device.

# src/freq.py

import machine
print(machine.freq())

Now lets upload this and execute it.

# syncs files to remove device
rsync ./src /pyboard

# goes into REPL
repl

# we import file by importing it without .py extension and this will run the script
> import freq

# CTRL+x will exit REPL

Additional resources

Disable mouse wake from suspend with systemd service

Sat, 15 Aug 2020 12:00:00 +0200

I recently bought ThinkPad X220 just as a joke on eBay to test Linux distributions and play around with things and not destroy my main machine. Little to my knowledge I felt in love with it. Man, they really made awesome machines back then.

After changing disk that came with it to SSD and installing Ubuntu to test if everything works I noticed that even after a single touch of my external mouse the system would wake up from sleep even though the lid was shut down.

I wouldn't even noticed it if laptop didn't have LED sleep indicator. I already had a bad experience with Linux and it's power management. I had a Dell Inspiron 7537 laptop with a touchscreen and while traveling it decided to wake up and started cooking in my backpack to the point that the digitizer responsible for touch actually glue off and the whole screen got wrecked. So, I am a bit touchy about this.

I went on solution hunting and to my surprise there is no easy way to disable specific devices to perform wake up. Why is this not under the power management tab in setting is really strange.

After googling for a solution I found this nice article describing the solution that worked for me. The only problem with this solution was that he added his solution to .bashrc and this triggers sudo that asks for a password each time new terminal is opened, which get annoying quickly since I open a lot of terminals all the time.

I followed his instructions and got to solution sudo sh -c "echo 'disabled' > /sys/bus/usb/devices/2-1.1/power/wakeup".

I created a system service file sudo nano /etc/systemd/system/disable-mouse-wakeup.service and removed sudo and replaced sh with /usr/bin/sh and pasted all that in ExecStart.

[Unit]
Description=Disables wakeup on mouse event
After=network.target
StartLimitIntervalSec=0

[Service]
Type=simple
Restart=always
RestartSec=1
User=root
ExecStart=/usr/bin/sh -c "echo 'disabled' > /sys/bus/usb/devices/2-1.1/power/wakeup"

[Install]
WantedBy=multi-user.target

After that I enabled, started and checked status of service.

sudo systemctl enable disable-mouse-wakeup.service
sudo systemctl start disable-mouse-wakeup.service
sudo systemctl status disable-mouse-wakeup.service

This will permanently disable that device from wakeing up you computer on boot. If you have many devices you would like to surpress from waking up your machine I would create a shell script and call that instead of direclty doing it in service file.

Remote work and how it affects the daily lives of people

Tue, 05 May 2020 12:00:00 +0200

I have been working remotely for the past 5 years. I love it. Love the freedom and make your schedule thingy.

You work more not less

I've heard from people things like: "Oh, you are so lucky, working from home, having all the free time you want". It was obvious they had no clue what means working remotely. They had this romantic idea of remote work. You can watch TV whenever you like, you can go outside for a picnic if you want and stuff like that.

This may be true if you work a day or two in a week from home. But if you go completely remote all these changes completely. I take some time to acclimate but then you start feeling the consequences of going fully remote. And it's not all rainbows and unicorns. Rather the opposite.

Feeling lost

At first, I remembered I felt lost. I was not used to this kind of environment. It felt disoriented and a part of you that is used to procrastinate turns on. You start thinking of a workday as a whole day. And soon this idea of "I can do this later" starts creeping in. Well, I have the whole day ahead of me. I can do this a bit later.

Hyper-performance

As a direct result, you become more focused on your work since you don't have all the interruptions common in the workplace. And you can quickly get used to this hyper-performance. But this mode requires also a lot of peace and quiet.

And here we come to the ugly parts of all this. People rarely have the self-control to not waste other people's time. It is paralyzing when people start calling you, sending you chat messages, etc. The thing is, that when I achieve this hyper-performance mode I am completely embroiled in the problem I am solving and this kind of interruptions mess with your head. I need an hour at least to get back in the zone. Sometimes not achieving the same focus the whole day.

I know that life is not how you want it to be and takes its route but from what I've learned this kind of interruptions can be avoided in 90% of the case easily just by closing any chat programs and putting your phone in a drawer.

Suggestion to all the new remote workers

Stop wasting other people's time. You don't bother people at their desks in the office either.
Do not replace daily chats in the hallways with instant messaging software. It will only interrupt people. Nothing good will come of it.
Set your working hours and try to not allow it to bleed outside these boundaries and maintain your routine.
Be prepared that hours will be longer regardless of your good intentions and your well thought of routine.
Try to be hyper-focused and do only one thing at the time. Multitasking is the enemy of progress.
Avoid long meetings and if possible eliminate them. Rather take time to write them out and allow others to respond in their own time. Meetings are usually a large waste of time and most of the people attending them are there just because the manager said so.
The software will not solve your problems. And throwing money at problems neither.
If you are in a managerial position don't supervise any single minute of workers. They are probably giving you more hours anyways. Track progress weekly not daily. You hired them and give them the benefit of the doubt that they will deliver what you agreed upon.

The strange case of Elasticsearch allocation failure

Sun, 29 Mar 2020 12:00:00 +0200

I've been using Elasticsearch in production for 5 years now and never had a single problem with it. Hell, never even known there could be a problem. Just worked. All this time. The first node that I deployed is still being used in production, never updated, upgraded, touched in anyway.

All this bliss came to an abrupt end this Friday when I got notification that Elasticsearch cluster went warm. Well, warm is not that bad right? Wrong! Quickly after that I got another email which sent chills down my spine. Cluster is now red. RED! Now, shit really hit the fan!

I tried googling what could be the problem and after executing allocation function noticed that some shards were unassigned and 5 attempts were already made (which is BTW to my luck the maximum) and that meant I am basically fucked. They also applied that one should wait for cluster to re-balance itself. So, I waited. One hour, two hours, several hours. Nothing, still RED.

The strangest thing about it all was, that queries were still being fulfilled. Data was coming out. On the outside it looked like nothing was wrong but everybody that would look at the cluster would know immediately that something was very very wrong and we were living on borrowed time here.

Please, DO NOT do what I did. Seriously! Please ask someone on official forums or if you know an expert please consult him. There could be million of reasons and these solution fit my problem. Maybe in your case it would disastrous. I had all the data backed up and even if I would fail spectacularly I would be able to restore the data. It would be a huge pain and I would loose couple of days but I had a plan B.

Executing allocation and told me what the problem was but no clear solution yet.

GET /_cat/allocation?format=json

I got a message that ALLOCATION_FAILED with additional info failed to create shard, failure ioexception[failed to obtain in-memory shard lock]. Well splendid! I must also say that our cluster is capable more than enough to handle the traffic. Also JVM memory pressure never was an issue. So what happened really then?

I tried also re-routing failed ones with no success due to AWS restrictions on having managed Elasticsearch cluster (they lock some of the functions).

POST /_cluster/reroute?retry_failed=true

I got a message that significantly reduced my options.

{
  "Message": "Your request: '/_cluster/reroute' is not allowed."
}

After that I went on a hunt again. I won't bother you with all the details because hours/days went by until I was finally able to re-index the problematic index and hoped for the best. Until that moment even re-indexing was giving me errors.

POST _reindex
{
  "source": {
    "index": "myindex"
  },
  "dest": {
    "index": "myindex-new"
  }
}

I needed to do this multiple times to get all the documents re-indexed. Then I dropped the original one with the following command.

DELETE /myindex

And re-indexed again new one in the original one (well by name only).

POST _reindex
{
  "source": {
    "index": "myindex-new"
  },
  "dest": {
    "index": "myindex"
  }
}

On the surface it looks like all is working but I have a long road in front of me to get all the things working again. Cluster now shows that it is in Green mode but I am also getting a notification that the cluster has processing status which could mean million of things.

Godspeed!

Create placeholder images with sharp Node.js image processing library

Fri, 27 Mar 2020 12:00:00 +0200

I have been searching for a solution to pre-generate some placeholder images for image server I needed to develop that resizes images on S3. I though this would be a 15min job and quickly found out how very mistaken I was.

Even though Node.js is not really the best way to do this kind of things (surely something written in C or Rust or even Golang would be the correct way to do this but we didn't need the speed in our case) I found an excellent library sharp - High performance Node.js image processing.

Getting things running was a breeze.

Fetch image from S3 and save resized

const sharp = require('sharp');
const aws = require('aws-sdk');

const x,y = 100;
const s3 = new aws.S3({});

aws.config.update({
  secretAccessKey: 'secretAccessKey',
  accessKeyId: 'accessKeyId',
  region: 'region'
});

const originalImage = await s3.getObject({
  Bucket: 'some-bucket-name',
  Key: 'image.jpg',
}).promise();

const resizedImage = await sharp(originalImage.Body)
  .resize(x, y)
  .jpeg({ progressive: true })
  .toBuffer();

s3.putObject({
  Bucket: 'some-bucket-name',
  Key: `optimized/${x}x${y}/image.jpg`,
  Body: resizedImage,
  ContentType: 'image/jpeg',
  ACL: 'public-read'
}).promise();

All this code was wrapped inside a web service with some additional security checks and defensive coding to detect if key is missing on S3.

And at that point I needed to return placeholder images as a response in case key is missing or x,y are not allowed by the server etc. I could have created PNG in Gimp and just serve them but I wanted to respect aspect ratio and I didn't want to return some mangled images.

Main problem with finding a clean solution I could copy and paste and change a bit was a task. API is changing constantly and there weren't clear examples or I was unable to find them.

Generating placeholder images using SVG

What I ended up was using SVG to generate text and created image with sharp and used composition to combine both layers. Response returned by this function is a buffer you can use to either upload to S3 or save to local file.

const generatePlaceholderImageWithText = async (width, height, message) => {
  const overlay = `<svg width="${width - 20}" height="${height - 20}">
    <text x="50%" y="50%" font-family="sans-serif" font-size="16" text-anchor="middle">${message}</text>
  </svg>`;

  return await sharp({
    create: {
      width: width,
      height: height,
      channels: 4,
      background: { r: 230, g: 230, b: 230, alpha: 1 }
    }
  })
    .composite([{
      input: Buffer.from(overlay),
      gravity: 'center',
    }])
    .jpeg()
    .toBuffer();
}

That is about it. Nothing more to it. You can change the color of the image by changing background and if you want to change text styling you can adapt SVG to your needs.

Also be careful about the length of the text. This function positions text at the center and adds 20px padding on all sides. If text is longer than the image it will get cut.

Simple Server-Sent Events based PubSub Server

Sun, 22 Mar 2020 12:00:00 +0200

Before we continue ...

Publisher Subscriber model is nothing new and there are many amazing solutions out there, so writing a new one would be a waste of time if other solutions wouldn't have quite complex install procedures and weren't so hard to maintain. But to be fair, comparing this simple server with something like Kafka or RabbitMQ is laughable at the least. Those solutions are enterprise grade and have many mechanisms there to ensure messages aren't lost and much more. Regardless of these drawbacks, this method has been tested on a large website and worked until now without any problems. So now, that we got that cleared up, let's continue.

Wiki definition: Publish/subscribe messaging, or pub/sub messaging, is a form of asynchronous service-to-service communication used in serverless and microservices architectures. In a pub/sub model, any message published to a topic is immediately received by all the subscribers to the topic.

General goals

provide a simple server that relays messages to all the connected clients,
messages can be posted on specific topics,
messages get sent via Server-Sent Events to all the subscribers.

How exactly does the pub/sub model work?

The easiest way to explain this is with diagram bellow. Basic function is simple. We have subscribers that receive messages, and we have publishers that create and post messages. Similar model is also well know pattern that works on a premise of consumers and producers, and they take similar roles.

These are some naive characteristics we want to achieve:

producer is publishing messages to subscribe topic,
consumer is receiving messages from subscribed topic,
servers is also known as Broker,
broker does not store messages or tracks success,
broker uses FIFO method for delivering messages,
if consumer wants to receive messages from a topic, producer and consumer topics must match,
consumer can subscribe to multiple topics,
producer can publish to multiple topics,
each message has a messageId.

Known drawbacks:

messages will not be stored in a persistent queue or unreceived messages like DeadLetterQueue so old messages could be lost on server restart,
Server-Sent Events opens a long-running connection between the client and the server so make sure if your setup is load balanced that the load balancer in this case can have long opened connection,
no system moderation due to the dynamic nature of creating queues.

Server-Sent Events

Read more about it on official specification page.

Current browser support

Check https://caniuse.com/#feat=eventsource for latest information about browser support.

Known issues

Firefox 52 and below do not support EventSource in web/shared workers
In Firefox prior to version 36 server-sent events do not reconnect automatically in case of a connection interrupt (bug)
Reportedly, CORS in EventSource is currently supported in Firefox 10+, Opera 12+, Chrome 26+, Safari 7.0+.
Antivirus software may block the event streaming data chunks.

Source: https://caniuse.com/#feat=eventsource

Message format

The simplest message that can be sent is only with data attribute:

data: this is a simple message
<blank line>

You can send message IDs to be used if the connection is dropped:

id: 33
data: this is line one
data: this is line two
<blank line>

And you can specify your own event types (the above messages will all trigger the message event):

id: 36
event: price
data: 103.34
<blank line>

Server requirements

The important thing is how you send headers and which headers are sent by the server that triggers browser to threat response as a EventStream.

Headers responsible for this are:

Content-Type: text/event-stream
Cache-Control: no-cache
Connection: keep-alive

Debugging with Google Chrome

Google Chrome provides build-in debugging and exploration tool for Server-Sent Events which is quite nice and available from Developer Tools under Network tab.

You can debug only client side events that get received and not the server ones. For debugging server events add console.log to server.js code and print out events.

Server implementation

For the sake of this example we will use Node.js with Express as our router since this is the easiest way to get started and we will use already written SSE library for node sse-pubsub so we don't reinvent the wheel.

npm init --yes

npm install express
npm install body-parser
npm install sse-pubsub

Basic implementation of a server (server.js):

const express = require('express');
const bodyParser = require('body-parser');
const SSETopic = require('sse-pubsub');

const app = express();
const port = process.env.PORT || 4000;

// topics container
const sseTopics = {};

app.use(bodyParser.json());

// open for all cors
app.all('*', (req, res, next) => {
  res.header('Access-Control-Allow-Origin', '*');
  res.header('Access-Control-Allow-Headers', 'X-Requested-With, Content-Type');
  next();
});

// preflight request error fix
app.options('*', async (req, res) => {
  res.header('Access-Control-Allow-Origin', '*');
  res.header('Access-Control-Allow-Headers', 'X-Requested-With, Content-Type');
  res.send('OK');
});

// serve the event streams
app.get('/stream/:topic', async (req, res, next) => {
  const topic = req.params.topic;

  if (!(topic in sseTopics)) {
    sseTopics[topic] = new SSETopic({
      pingInterval: 0,
      maxStreamDuration: 15000,
    });
  }

  // subscribing client to topic
  sseTopics[topic].subscribe(req, res);
});

// accepts new messages into topic
app.post('/publish', async (req, res) => {
  let body = req.body;
  let status = 200;

  console.log('Incoming message:', req.body);

  if (
    body.hasOwnProperty('topic') &&
    body.hasOwnProperty('event') &&
    body.hasOwnProperty('message')
  ) {
    const topic = req.body.topic;
    const event = req.body.event;
    const message = req.body.message;

    if (topic in sseTopics) {
      // sends message to all the subscribers
      sseTopics[topic].publish(message, event);
    }
  } else {
    status = 400;
  }

  res.status(status).send({
    status,
  });
});

// returns JSON object of all opened topics
app.get('/status', async (req, res) => {
  res.send(sseTopics);
});

// health-check endpoint
app.get('/', async (req, res) => {
  res.send('OK');
});

// return a 404 if no routes match
app.use((req, res, next) => {
  res.set('Cache-Control', 'private, no-store');
  res.status(404).end('Not found');
});

// starts the server
app.listen(port, () => {
  console.log(`PubSub server running on http://localhost:${port}`);
});

Our custom message format

Each message posted on a server must be in a specific format that out server accepts. Having structure like this allows us to have multiple separated type of events on each topic.

With this we can separate streams and only receive events that belong to the topic.

One example would be, that we have index page and we want to receive messages about new upvotes or new subscribers but we don't want to follow events for other pages. This reduces clutter and overall network. And structure is much nicer and maintanable.

{
  "topic": "sample-topic",
  "event": "sample-event",
  "message": { "name": "John" }
}

Publisher and subscriber clients

Publisher and subscriber in action

You can download the code and follow along.

Publisher

As talked about above publisher is the one that send messages to the broker/server. Message inside the payload can be whatever you want (string, object, array). I would however personally avoid send large chunks of data like blobs and such.

<!DOCTYPE html>
<html lang="en">

  <head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Publisher</title>
  </head>

  <body>

    <h1>Publisher</h1>

    <fieldset>
      <p>
        <label>Server:</label>
        <input type="text" id="server" value="http://localhost:4000">
      </p>
      <p>
        <label>Topic:</label>
        <input type="text" id="topic" value="sample-topic">
      </p>
      <p>
        <label>Event:</label>
        <input type="text" id="event" value="sample-event">
      </p>
      <p>
        <label>Message:</label>
        <input type="text" id="message" value='{"name": "John"}'>
      </p>
      <p>
        <button type="button" id="button">Publish message to topic</button>
      </p>
    </fieldset>

    <script>

      const button = document.querySelector('#button');
      const server = document.querySelector('#server');
      const topic = document.querySelector('#topic');
      const event = document.querySelector('#event');
      const message = document.querySelector('#message');

      button.addEventListener('click', async (evt) => {
        const req = await fetch(`${server.value}/publish`, {
          method: 'post',
          headers: {
            'Accept': 'application/json',
            'Content-Type': 'application/json',
          },
          body: JSON.stringify({
            topic: topic.value,
            event: event.value,
            message: JSON.parse(message.value),
          }),
        });

        const res = await req.json();
        console.log(res);
      });

    </script>

  </body>

</html>

Subscriber

Subscriber is responsible for receiving new messages that come from server via publisher. The code bellow is very rudimentary but works and follows the implementation guidelines for EventSource.

You can use either Developer Tools Console to see incoming messages or you can defer to Debugging with Google Chrome section above to see all EventStream messages.

Don't be alarmed if the subscriber gets disconnected from the server every so often. The code we have here resets connection every 15s but it automatically get reconnected and fetches all messages up to last received message id. This setting can be adjusted in server.js file; search for the maxStreamDuration variable.

<!DOCTYPE html>
<html lang="en">

  <head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>Subscriber</title>
    <link rel="stylesheet" href="style.css">
  </head>

  <body>

    <h1>Subscriber</h1>

    <fieldset>
      <p>
        <label>Server:</label>
        <input type="text" id="server" value="http://localhost:4000">
      </p>
      <p>
        <label>Topic:</label>
        <input type="text" id="topic" value="sample-topic">
      </p>
      <p>
        <label>Event:</label>
        <input type="text" id="event" value="sample-event">
      </p>
      <p>
        <button type="button" id="button">Subscribe to topic</button>
      </p>
    </fieldset>

    <script>

      const button = document.querySelector('#button');
      const server = document.querySelector('#server');
      const topic = document.querySelector('#topic');
      const event = document.querySelector('#event');

      button.addEventListener('click', async (evt) => {

        let es = new EventSource(`${server.value}/stream/${topic.value}`);

        es.addEventListener(event.value, function (evt) {
          console.log(`incoming message`, JSON.parse(evt.data));
        });

        es.addEventListener('open', function (evt) {
          console.log('connected', evt);
        });

        es.addEventListener('error', function (evt) {
          console.log('error', evt);
        });

      });

    </script>

  </body>

</html>

Reading further

Using sentiment analysis for clickbait detection in RSS feeds

Sat, 19 Oct 2019 12:00:00 +0200

Initial thoughts

One of the things that interested me for a while now is if major well established news sites use click bait titles to drive additional traffic to their sites and generate additional impressions.

Goal is to see how article titles and actual content of article differ from each other and see if titles are clickbaited.

Preparing and cleaning data

For this example I opted to just use RSS feed from a new website and decided to go with The Guardian World news. While this gets us limited data (~40) articles and also description (actual content) is trimmed this really doesn't reflect the actual article contents.

To get better content I could use web scraping and use RSS as link list and fetch contents directly from website, but for this simple example this will suffice.

There are couple of requirements we need to install before we continue:

pip3 install feedparser (parses RSS feed from url)
pip3 install vaderSentiment (does sentiment polarity analysis)
pip3 install matplotlib (plots chart of results)

So first we need to fetch RSS data and sanitize HTML content from description.

import re
import feedparser

feed_url = "https://www.theguardian.com/world/rss"
feed = feedparser.parse(feed_url)

# sanitize html
for item in feed.entries:
    item.description = re.sub('<[^<]+?>', '', item.description)

Perform sentiment analysis

Since we now have cleaned up data in our feed.entries object we can start with performing sentiment analysis.

There are many sentiment analysis libraries available that range from rule-based sentiment analysis up to machine learning supported analysis. To keep things simple I decided to use rule-based analysis library vaderSentiment from C.J. Hutto. Really nice library and quite easy to use.

from vaderSentiment.vaderSentiment import SentimentIntensityAnalyzer
analyser = SentimentIntensityAnalyzer()

sentiment_results = []
for item in feed.entries:
    sentiment_title = analyser.polarity_scores(item.title)
    sentiment_description = analyser.polarity_scores(item.description)
    sentiment_results.append([sentiment_title['compound'], sentiment_description['compound']])

Now that we have this data in a shape that is compatible with matplotlib we can plot results to see the difference between title and description sentiment of an article.

import matplotlib.pyplot as plt

plt.rcParams['figure.figsize'] = (15, 3)
plt.plot(sentiment_results, drawstyle='steps')
plt.title('Sentiment analysis relationship between title and description (Guardian World News)')
plt.legend(['title', 'description'])
plt.show()

Results and assets

Because of the small sample size further conclusions are impossible to make.
Rule-based approach may not be the best way of doing this. By using deep learning we would be able to get better insights.
Next step would be to periodically fetch RSS items and store them over a longer period of time and then perform analysis again and use either machine learning or deep learning on top of it.

Figure above displays difference between title and description sentiment for specific RSS feed item. 1 means positive and -1 means negative sentiment.

» Download Jupyter Notebook

Going further

Encoding binary data into DNA sequence

Thu, 03 Jan 2019 12:00:00 +0200

Initial thoughts

Imagine a world where you could go outside and take a leaf from a tree and put it through your personal DNA sequencer and get data like music, videos or computer programs from it. Well, this is all possible now. It was not done on a large scale because it is quite expensive to create DNA strands but it's possible.

Encoding data into DNA sequence is relatively simple process once you understand the relationship between binary data and nucleotides and scientists have been making large leaps in this field in order to provide viable long-term storage solution for our data that would potentially survive our specie if case of global disaster. We could imprint all the world's knowledge into plants and ensure the survival of our knowledge.

More optimistic usage for this technology would be easier storage of ever growing data we produce every day. Once machines for sequencing DNA become fast enough and cheaper this could mean the next evolution of storing data and abandoning classical hard and solid state drives in data warehouses.

As we currently stand this is still not viable but it is quite an amazing and cool technology.

My interests in this field are purely in encoding processes and experimental testing mainly because I don't have the access to this expensive machines. My initial goal was to create a toolkit that can be used by everybody to encode their data into a proper DNA sequence.

Glossary

deoxyribose A five-carbon sugar molecule with a hydrogen atom rather than a hydroxyl group in the 2′ position; the sugar component of DNA nucleotides.

double helix The molecular shape of DNA in which two strands of nucleotides wind around each other in a spiral shape.

nitrogenous base A nitrogen-containing molecule that acts as a base; often referring to one of the purine or pyrimidine components of nucleic acids.

phosphate group A molecular group consisting of a central phosphorus atom bound to four oxygen atoms.

RGB The RGB color model is an additive color model in which red, green and blue light are added together in various ways to reproduce a broad array of colors.

GCC The GNU Compiler Collection is a compiler system produced by the GNU Project supporting various programming languages.

Data encoding

TL;DR: Encoding involves the use of a code to change original data into a form that can be used by an external process.

Encoding is the process of converting data into a format required for a number of information processing needs, including:

Program compiling and execution
Data transmission, storage and compression/decompression
Application data processing, such as file conversion

Encoding can have two meanings:

In computer technology, encoding is the process of applying a specific code, such as letters, symbols and numbers, to data for conversion into an equivalent cipher.
In electronics, encoding refers to analog to digital conversion.

Quick history of DNA

1869 - Friedrich Miescher identifies "nuclein".
1900s - The Eugenics Movement.
1900 – Mendel's theories are rediscovered by researchers.
1944 - Oswald Avery identifies DNA as the 'transforming principle'.
1952 - Rosalind Franklin photographs crystallized DNA fibres.
1953 - James Watson and Francis Crick discover the double helix structure of DNA.
1965 - Marshall Nirenberg is the first person to sequence the bases in each codon.
1983 - Huntington's disease is the first mapped genetic disease.
1990 - The Human Genome Project begins.
1995 - Haemophilus Influenzae is the first bacterium genome sequenced.
1996 - Dolly the sheep is cloned.
1999 - First human chromosome is decoded.
2000 – Genetic code of the fruit fly is decoded.
2002 – Mouse is the first mammal to have its genome decoded.
2003 – The Human Genome Project is completed.
2013 – DNA Worldwide and Eurofins Forensic discover identical twins have differences in their genetic makeup.

What is DNA?

Deoxyribonucleic acid, a self-replicating material which is present in nearly all living organisms as the main constituent of chromosomes. It is the carrier of genetic information.

The nitrogen in our DNA, the calcium in our teeth, the iron in our blood, the carbon in our apple pies were made in the interiors of collapsing stars. We are made of starstuff. -- Carl Sagan, Cosmos

The nucleotide in DNA consists of a sugar (deoxyribose), one of four bases (cytosine (C), thymine (T), adenine (A), guanine (G)), and a phosphate. Cytosine and thymine are pyrimidine bases, while adenine and guanine are purine bases. The sugar and the base together are called a nucleoside.

DNA (a) forms a double stranded helix, and (b) adenine pairs with thymine and cytosine pairs with guanine. (credit a: modification of work by Jerome Walker, Dennis Myts)

Encode binary data into DNA sequence

As an input file you can use any file you want:

ASCII files,
Compiled programs,
Multimedia files (MP3, MP4, MVK, etc),
Images,
Database files,
etc.

Note: If you would copy all the bytes from RAM to file or pipe data to file you could encode also this data as long as you provide file pointer to the encoder.

Basic Encoding

As already mentioned, the Basic Encoding is based on a simple mapping. Since DNA is composed of 4 nucleotides (Adenine, Cytosine, Guanine, Thymine; usually referred using the first letter). Using this technique we can encode

using a single nucleotide. In this way, we are able to use the 4 bases that compose the DNA strand to encode each byte of data.

Two bits	Nucleotides
00	A (Adenine)
10	G (Guanine)
01	C (Cytosine)
11	T (Thymine)

With this in mind we can simply encode any data by using two-bit to Nucleotides conversion.

{ Algorithm 1: Naive byte array to DNA encode }
procedure EncodeToDNASequence(f) string
begin
  enc string
  while not eof(f) do
    c byte := buffer[0]                             { Read 1 byte from buffer }
    bin integer := sprintf('08b', c)                { Convert to string binary }
    for e in range[0, 2, 4, 6] do
      if e[0] == 48 and e[1] == 48 then             { 0x00 - A (Adenine) }
        enc += 'A'
      else if e[0] == 48 and e[1] == 49 then        { 0x01 - G (Guanine) }
        enc += 'G'
      else if e[0] == 49 and e[1] == 48 then        { 0x10 - C (Cytosine) }
        enc += 'C'
      else if e[0] == 49 and e[1] == 49 then        { 0x11 - T (Thymine) }
        enc += 'T'
  return enc                                        { Return DNA sequence }
end

Another encoding would be Goldman encoding. Using this encoding helps with Nonsense mutation (amino acids replaced by a stop codon) that occurs and is the most problematic during translation because it leads to truncated amino acid sequences, which in turn results in truncated proteins.

Where to store big data? In DNA: Nick Goldman at TEDxPrague

FASTA file format

In bioinformatics, FASTA format is a text-based format for representing either nucleotide sequences or peptide sequences, in which nucleotides or amino acids are represented using single-letter codes. The format also allows for sequence names and comments to precede the sequences. The format originates from the FASTA software package, but has now become a standard in the field of bioinformatics.

The first line in a FASTA file started either with a ">" (greater-than) symbol or, less frequently, a ";" (semicolon) was taken as a comment. Subsequent lines starting with a semicolon would be ignored by software. Since the only comment used was the first, it quickly became used to hold a summary description of the sequence, often starting with a unique library accession number, and with time it has become commonplace to always use ">" for the first line and to not use ";" comments (which would otherwise be ignored).

;LCBO - Prolactin precursor - Bovine
; a sample sequence in FASTA format
MDSKGSSQKGSRLLLLLVVSNLLLCQGVVSTPVCPNGPGNCQVSLRDLFDRAVMVSHYIHDLSS
EMFNEFDKRYAQGKGFITMALNSCHTSSLPTPEDKEQAQQTHHEVLMSLILGLLRSWNDPLYHL
VTEVRGMKGAPDAILSRAIEIEEENKRLLEGMEMIFGQVIPGAKETEPYPVWSGLPSLQTKDED
ARYSAFYNLLHCLRRDSSKIDTYLKLLNCRIIYNNNC*

>MCHU - Calmodulin - Human, rabbit, bovine, rat, and chicken
ADQLTEEQIAEFKEAFSLFDKDGDGTITTKELGTVMRSLGQNPTEAELQDMINEVDADGNGTID
FPEFLTMMARKMKDTDSEEEIREAFRVFDKDGNGYISAAELRHVMTNLGEKLTDEEVDEMIREA
DIDGDGQVNYEEFVQMMTAK*

>gi|5524211|gb|AAD44166.1| cytochrome b [Elephas maximus maximus]
LCLYTHIGRNIYYGSYLYSETWNTGIMLLLITMATAFMGYVLPWGQMSFWGATVITNLFSAIPYIGTNLV
EWIWGGFSVDKATLNRFFAFHFILPFTMVALAGVHLTFLHETGSNNPLGLTSDSDKIPFHPYYTIKDFLG
LLILILLLLLLALLSPDMLGDPDNHMPADPLNTPLHIKPEWYFLFAYAILRSVPNKLGGVLALFLSIVIL
GLMPFLHTSKHRSMMLRPLSQALFWTLTMDLLTLTWIGSQPVEYPYTIIGQMASILYFSIILAFLPIAGX
IENY

FASTA format was extended by FASTQ format from the Sanger Centre in Cambridge.

PNG encoded DNA sequence

Nucleotides	RGB	Color name
A ➞ Adenine	(0,0,255)	Blue
G ➞ Guanine	(0,100,0)	Green
C ➞ Cytosine	(255,0,0)	Red
T ➞ Thymine	(255,255,0)	Yellow

With this in mind we can create a simple algorithm to create PNG representation of a DNA sequence.

{ Algorithm 2: Naive DNA to PNG encode from FASTA file }
procedure EncodeDNASequenceToPNG(f)
begin
  i image
  while not eof(f) do
    c char := buffer[0]                             { Read 1 char from buffer }
    case c of
      'A': color := RGB(0, 0, 255)                  { Blue }
      'G': color := RGB(0, 100, 0)                  { Green }
      'C': color := RGB(255, 0, 0)                  { Red }
      'T': color := RGB(255, 255, 0)                { Yellow }
    drawRect(i, [x, y], color)
  save(i)                                           { Save PNG image }
end

Encoding text file in practice

In this example we will take a simple text file as our input stream for encoding. This file will have a quote from Niels Bohr and saved as txt file.

How wonderful that we have met with a paradox. Now we have some hope of making progress. ― Niels Bohr

First we encode text file into FASTA file.

./dnae-encode -i quote.txt -o quote.fa
2019/01/10 00:38:29 Gathering input file stats
2019/01/10 00:38:29 Starting encoding ...
 106 B / 106 B [==================================] 100.00% 0s
2019/01/10 00:38:29 Saving to FASTA file ...
2019/01/10 00:38:29 Output FASTA file length is 438 B
2019/01/10 00:38:29 Process took 987.263µs
2019/01/10 00:38:29 Done ...

Output of quote.fa file contains the encoded DNA sequence in ASCII format.

>SEQ1
GACAGCTTGTGTACAAGTGTGCTTGCTCGCGAGCGGGTACGCGCGTGGGCTAACAAGTGA
GCCAGCAGGTGAACAAGTGTGCGGACAAGCCAGCAGGTGCGCGGACAAGCTGGCGGGTGA
ACAAGTGTGCCGGTGAGCCAACAAGCAGACAAGTAAGCAGGTACGCAGGCGAGCTTGTCA
ACTCACAAGATCGCTTGTGTACAAGTGTGCGGACAAGCCAGCAGGTGCGCGGACAAGTAT
GCTTGCTGGCGGACAAGCCAGCTTGTAAGCGGACAAGCTTGCGCACAAGCTGGCAGGCCT
GCCGGCTCGCGTACAAATTCACAAGTAAGTACGCTTGCGTGTACGCGGGTATGTATACTC
AACCTCACCAAACGGGACAAGATCGCCGGCGGGCTAGTATACAAGAACGCTTGCCAGTAC
AACC

Then we encode FASTA file from previous operation to encode this data into PNG.

./dnae-png -i quote.fa -o quote.png
2019/01/10 00:40:09 Gathering input file stats ...
2019/01/10 00:40:09 Deconstructing FASTA file ...
2019/01/10 00:40:09 Compositing image file ...
 424 / 424 [==================================] 100.00% 0s
2019/01/10 00:40:09 Saving output file ...
2019/01/10 00:40:09 Output image file length is 1.1 kB
2019/01/10 00:40:09 Process took 19.036117ms
2019/01/10 00:40:09 Done ...

After encoding into PNG format this file looks like this.

The larger the input stream is the larger the PNG file would be.

Compiled basic Hello World C program with GCC would look like.

// gcc -O3 -o sample sample.c
#include <stdio.h>

main() {
  printf("Hello, world!\n");
  return 0;
}

Toolkit for encoding data

I have created a toolkit with two main programs:

dnae-encode (encodes file into FASTA file)
dnae-png (encodes FASTA file into PNG)

Toolkit with full source code is available on github.com/mitjafelicijan/dna-encoding.

dnae-encode

> ./dnae-encode --help
usage: dnae-encode --input=INPUT [<flags>]

A command-line application that encodes file into DNA sequence.

Flags:
      --help             Show context-sensitive help (also try --help-long and --help-man).
  -i, --input=INPUT      Input file (ASCII or binary) which will be encoded into DNA sequence.
  -o, --output="out.fa"  Output file which stores DNA sequence in FASTA format.
  -s, --sequence=SEQ1    The description line (defline) or header/identifier line, gives a name and/or a unique identifier for the sequence.
  -c, --columns=60       Row characters length (no more than 120 characters). Devices preallocate fixed line sizes in software.
      --version          Show application version.

dnae-png

> ./dnae-png --help
usage: dnae-png --input=INPUT [<flags>]

A command-line application that encodes FASTA file into PNG image.

Flags:
      --help              Show context-sensitive help (also try --help-long and --help-man).
  -i, --input=INPUT       Input FASTA file which will be encoded into PNG image.
  -o, --output="out.png"  Output file in PNG format that represents DNA sequence in graphical way.
  -s, --size=10           Size of pairings of DNA bases on image in pixels (lower resolution lower file size).
      --version           Show application version.

Benchmarks

First we generate some binary sample data with dd.

dd if=<(openssl enc -aes-256-ctr  -pass pass:"$(dd if=/dev/urandom bs=128 count=1 2>/dev/null | base64)" -nosalt < /dev/zero) of=1KB.bin bs=1KB count=1 iflag=fullblock

Our freshly generated 1KB file looks something like this (its full of garbage data as intended).

We create following binary files:

1KB.bin
10KB.bin
100KB.bin
1MB.bin
10MB.bin
100MB.bin

After this we create FASTA files for all the binary files by encoding them into DNA sequence.

./dnae-encode -i 100MB.bin -o 100MB.fa

Then we GZIP all the FASTA files to see how much the can be compressed.

gzip -9 < 10MB.fa > 10MB.fa.gz

The speed increase that occurs when encoding to FASTA format.

Size of the out file after encoding.

Download CSV file with benchmarks.

References

Using DigitalOcean Spaces Object Storage with FUSE

Tue, 16 Jan 2018 12:00:00 +0200

Couple of months ago DigitalOcean introduced new product called Spaces which is Object Storage very similar to Amazon's S3. This really peaked my interest, because this was something I was missing and even the thought of going over the internet for such functionality was in no interest to me. Also in fashion with their previous pricing this also is very cheap and pricing page is a no-brainer compared to AWS or GCE. Prices are clearly and precisely defined and outlined. You must love them for that :)

Initial requirements

Is it possible to use them as a mounted drive with FUSE? (tl;dr YES)
Will the performance degrade over time and over different sizes of objects? (tl;dr NO&YES)
Can storage be mounted on multiple machines at the same time and be writable? (tl;dr YES)

Let me be clear. This scripts I use are made just for benchmarking and are not intended to be used in real-life situations. Besides that, I am looking into using this approaches but adding caching service in front of it and then dumping everything as an object to storage. This could potentially be some interesting post of itself. But in case you would need real-time data without eventual consistency please take this scripts as they are: not usable in such situations.

Is it possible to use them as a mounted drive with FUSE?

Well, actually they can be used in such manor. Because they are similar to AWS S3 many tools are available and you can find many articles and Stackoverflow items.

To make this work you will need DigitalOcean account. If you don't have one you will not be able to test this code. But if you have an account then you go and create new Droplet. If you click on this link you will already have preselected Debian 9 with smallest VM option.

Please be sure to add you SSH key, because we will login to this machine remotely.
If you change your region please remember which one you choose because we will need this information when we try to mount space to our machine.

Instuctions on how to use SSH keys and how to setup them are available in article How To Use SSH Keys with DigitalOcean Droplets.

After we created Droplet it's time to create new Space. This is done by clicking on a button Create (right top corner) and selecting Spaces. Choose pronounceable Unique name because we will use it in examples below. You can either choose Private or Public, it doesn't matter in our case. And you can always change that in the future.

When you have created new Space we should generate Access key. This link will guide to the page when you can generate this key. After you create new one, please save provided Key and Secret because Secret will not be shown again.

Now that we have new Space and Access key we should SSH into our machine.

# replace IP with the ip of your newly created droplet
ssh root@IP

# this will install utilities for mounting storage objects as FUSE
apt install s3fs

# we now need to provide credentials (access key we created earlier)
# replace KEY and SECRET with your own credentials but leave the colon between them
# we also need to set proper permissions
echo "KEY:SECRET" > .passwd-s3fs
chmod 600 .passwd-s3fs

# now we mount space to our machine
# replace UNIQUE-NAME with the name you choose earlier
# if you choose different region for your space be careful about -ourl option (ams3)
s3fs UNIQUE-NAME /mnt/ -ourl=https://ams3.digitaloceanspaces.com -ouse_cache=/tmp

# now we try to create a file
# once you mount it may take a couple of seconds to retrieve data
echo "Hello cruel world" > /mnt/hello.txt

After all this you can return to your browser and go to DigitalOcean Spaces and click on your created space. If file hello.txt is present you have successfully mounted space to your machine and wrote data to it.

I choose the same region for my Droplet and my Space but you don't have to. You can have different regions. What this actually does to performance I don't know.

Additional information on FUSE:

Will the performance degrade over time and over different sizes of objects?

For this task I didn't want to just read and write text files or uploading images. I actually wanted to figure out if using something like SQlite is viable in this case.

Measurement experiment 1: File copy

# first we create some dummy files at different sizes
dd if=/dev/zero of=10KB.dat bs=1024 count=10 #10KB
dd if=/dev/zero of=100KB.dat bs=1024 count=100 #100KB
dd if=/dev/zero of=1MB.dat bs=1024 count=1024 #1MB
dd if=/dev/zero of=10MB.dat bs=1024 count=10240 #10MB

# now we set time command to only return real
TIMEFORMAT=%R

# now lets test it
(time cp 10KB.dat /mnt/) |& tee -a 10KB.results.txt

# and now we automate
# this will perform the same operation 100 times
# this will output results into separated files based on objecty size
n=0; while (( n++ < 100 )); do (time cp 10KB.dat /mnt/10KB.$n.dat) |& tee -a 10KB.results.txt; done
n=0; while (( n++ < 100 )); do (time cp 100KB.dat /mnt/100KB.$n.dat) |& tee -a 100KB.results.txt; done
n=0; while (( n++ < 100 )); do (time cp 1MB.dat /mnt/1MB.$n.dat) |& tee -a 1MB.results.txt; done
n=0; while (( n++ < 100 )); do (time cp 10MB.dat /mnt/10MB.$n.dat) |& tee -a 10MB.results.txt; done

Files of size 100MB were not successfully transferred and ended up displaying error (cp: failed to close '/mnt/100MB.1.dat': Operation not permitted).

As I suspected, object size is not really that important. Sadly I don't have the time to test performance over periods of time. But if some of you would do it please send me your data. I would be interested in seeing results.

Here are plotted results

You can download raw result here. Measurements are in seconds.

As far as these tests show, performance is quite stable and can be predicted which is fantastic. But this is a small test and spans only over couple of hours. So you should not completely trust them.

Measurement experiment 2: SQLite performanse

I was unable to use database file directly from mounted drive so this is a no-go as I suspected. So I executed code below on a local disk just to get some benchmarks. I inserted 1000 records with DROPTABLE, CREATETABLE, INSERTMANY, FETCHALL, COMMIT for 1000 times to generate statistics. As you can see performance of SQLite is quite amazing. You could then potentially just copy file to mounted drive and be done with it.

import time
import sqlite3
import sys

if len(sys.argv) < 3:
  print("usage: python sqlite-benchmark.py DB_PATH NUM_RECORDS REPEAT")
  exit()

def data_iter(x):
  for i in range(x):
    yield "m" + str(i), "f" + str(i*i)

header_line = "%s\t%s\t%s\t%s\t%s\n" % ("DROPTABLE", "CREATETABLE", "INSERTMANY", "FETCHALL", "COMMIT")
with open("sqlite-benchmarks.tsv", "w") as fp:
  fp.write(header_line)

start_time = time.time()
conn = sqlite3.connect(sys.argv[1])
c = conn.cursor()
end_time = time.time()
result_time = CONNECT = end_time - start_time
print("CONNECT: %g seconds" % (result_time))

start_time = time.time()
c.execute("PRAGMA journal_mode=WAL")
c.execute("PRAGMA temp_store=MEMORY")
c.execute("PRAGMA synchronous=OFF")
result_time = PRAGMA = end_time - start_time
print("PRAGMA: %g seconds" % (result_time))

for i in range(int(sys.argv[3])):
  print("#%i" % (i))

  start_time = time.time()
  c.execute("drop table if exists test")
  end_time = time.time()
  result_time = DROPTABLE = end_time - start_time
  print("DROPTABLE: %g seconds" % (result_time))

  start_time = time.time()
  c.execute("create table if not exists test(a,b)")
  end_time = time.time()
  result_time = CREATETABLE = end_time - start_time
  print("CREATETABLE: %g seconds" % (result_time))

  start_time = time.time()
  c.executemany("INSERT INTO test VALUES (?, ?)", data_iter(int(sys.argv[2])))
  end_time = time.time()
  result_time = INSERTMANY = end_time - start_time
  print("INSERTMANY: %g seconds" % (result_time))

  start_time = time.time()
  c.execute("select count(*) from test")
  res = c.fetchall()
  end_time = time.time()
  result_time = FETCHALL = end_time - start_time
  print("FETCHALL: %g seconds" % (result_time))

  start_time = time.time()
  conn.commit()
  end_time = time.time()
  result_time = COMMIT = end_time - start_time
  print("COMMIT: %g seconds" % (result_time))

  print
  log_line = "%f\t%f\t%f\t%f\t%f\n" % (DROPTABLE, CREATETABLE, INSERTMANY, FETCHALL, COMMIT)
  with open("sqlite-benchmarks.tsv", "a") as fp:
    fp.write(log_line)

start_time = time.time()
conn.close()
end_time = time.time()
result_time = CLOSE = end_time - start_time
print("CLOSE: %g seconds" % (result_time))

You can download raw result here. And again, these results are done on a local block storage and do not represent capabilities of object storage. With my current approach and state of the test code these can not be done. I would need to make Python code much more robust and check locking etc.

Can storage be mounted on multiple machines at the same time and be writable?

Well, this one didn't take long to test. And the answer is YES. I mounted space on both machines and measured same performance on both machines. But because file is downloaded before write and then uploaded on complete there could potentially be problems is another process is trying to access the same file.

Observations and conslusion

Using Spaces in this way makes it easier to access and manage files. But besides that you would need to write additional code to make this one play nice with you applications.

Nevertheless, this was extremely simple to setup and use and this is just another excellent product in DigitalOcean product line. I found this exercise very valuable and am thinking about implementing some sort of mechanism for SQLite, so data can be stored on Spaces and accessed by many VM's. For a project where data doesn't need to be accessible in real-time and can have couple of minutes old data this would be very interesting. If any of you find this proposal interesting please write in a comment box below or shoot me an email and I will keep you posted.

Simple IOT application supported by real-time monitoring and data history

Fri, 11 Aug 2017 12:00:00 +0200

Initial thoughts

I have been developing these kind of application for the better part of my last 5 years and people keep asking me how to approach developing such application and I will give a try explaining it here.

IOT applications are really no different than any other kind of applications. We have data that needs to be collected and visualized in some form of tables or charts. The main difference here is that most of the times these data is collected by some kind of device foreign to developer that mainly operates in web domain. But fear not, it's not that different than writing some JavaScript.

There are many devices able to transmit data via wireless or wired network by default but for the sake of example we will be using commonly known Arduino with wireless module already on the board → Arduino MKR1000.

In order to make this little project as accessible to others as possible I will try to make it as inexpensive as possible. And by this I mean that I will avoid using hosted virtual servers and will be using my own laptop as a server. But you must buy Arduino MKR1000 to follow steps below. But if you would want to deploy this software I would suggest using DigitalOcean → smallest VPS is only per month making this one of the most affordable option out there. Please notice that this software will not run on stock web hosting that only supports LAMP (Linux, Apache, MySQL, and PHP).

But before we begin please take notice that this is strictly experimental code and not well optimized and there are much better ways in handling some aspects of the application but that requires much deeper knowledge of technology that is not needed for an example like this.

Development steps

Simple Python API that will receive and store incoming data.
Prototype C++ code that will read "sensor data" and transmit it to API.
Data visualization with charts → extends Python web application.

Step 1. and 3. will share the same web application. One route will be dedicated to API and another to serving HTML with chart.

Schema below represents what we will try to achieve and how different parts correlates to each other.

Simple Python API

I have always been a fan of simplicity so we will be using Bottle: Python Web Framework. It is a single file web framework that seriously simplifies working with routes, templating and has built-in web server that satisfies our need in this case.

First we need to install bottle package. This can be done by downloading bottle.py and placing it in the root of your application or by using pip software pip install bottle --user.

If you are using Linux or MacOS then Python is already installed. If you will try to test this on Windows please install Python for Windows. There may be some problems with path when you will try to launch python webapp.py so please take care of this before you continue.

Basic web application

Most basic bottle application is quite simple. Paste code below in webapp.py file and save.

# -*- coding: utf-8 -*-

import bottle

# initializing bottle app
app = bottle.Bottle()

# triggered when / is accessed from browser
# only accepts GET → no POST allowed
@app.route("/", method=["GET"])
def route_default():
  return "howdy from python"

# starting server on http://0.0.0.0:5000
if __name__ == "__main__":
  bottle.run(
    app = app,
    host = "0.0.0.0",
    port = 5000,
    debug = True,
    reloader = True,
    catchall = True,
  )

To run this simple application you should open command prompt or terminal on your machine and go to the folder containing your file and type python webapp.py. If everything goes ok then open your web browser and point it to http://0.0.0.0:5000.

If you would like change the port of your application (like port 80) and not use root to run your app this will present a problem. The TCP/IP port numbers below 1024 are privileged ports → this is a security feature. So in order of simplicity and security use a port number above 1024 like I have used port 5000.

If this fails at any time please fix it before you continue, because nothing below will work otherwise.

We use 0.0.0.0 as default host so that this app is available over your local network. If you find your local ip ifconfig and try accessing this site with your phone (if on same network/router as your machine) this should work as well (example of such ip http://192.168.1.15:5000). This is a must have because Arduino will be accessing this application to send it's data.

Web application security

There is a lot to be said about security and is a topic of many books. Of course all this can not be written here but to just establish some basic security → you should always use SSL with your application. Some fantastic free certificates are available by Let's Encrypt - Free SSL/TLS Certificates. With SSL certificate installed you should then make use of HTTP headers and send your "API key" via a header. If your key is send via header then this key is encrypted by SSL and send encrypted over the network. Never send your api keys by GET parameter like http://example.com/?api_key=somekeyvalue. The problem that this kind of sending presents is that this key is visible in logs and by network sniffers.

There is a fantastic article describing some aspects about security: 11 Web Application Security Best Practices. Please check it out.

Simple API for writing data-points

We will now be using boilerplate code from example above and extend it to be SQLite3 because it plays well with Python and can store quite large amount of able to write data received by API to local storage. For example use I will use data. I have been using it to collect gigabytes of data in a single database without any corruption or problems → your experience may vary.

To avoid learning SQLite I will be using Dataset: databases for lazy people. This package abstracts SQL and simplifies writing and reading data from database. You should install this package with pip software pip install dataset --user.

Because API will use POST method I will be testing if code works correctly by using Restlet Client for Google Chrome. This software also allows you to set headers → for basic security with API_KEY.

To quickly generate passwords or API keys I usually use this nifty website RandomKeygen.

Copy and paste code below over your previous code in file webapp.py.

# -*- coding: utf-8 -*-

import time
import bottle
import random
import dataset

# initializing bottle app
app = bottle.Bottle()

# connects to sqlite database
# check_same_thread=False allows using it in multi-threaded mode
app.config["dsn"] = dataset.connect("sqlite:///data.db?check_same_thread=False")

# api key that will be used in Arduino code
app.config["api_key"] = "JtF2aUE5SGHfVJBCG5SH"

# triggered when /api is accessed from browser
# only accepts POST → no GET allowed
@app.route("/api", method=["POST"])
def route_default():
  status = 400
  ts = int(time.time()) # current timestamp
  value = bottle.request.body.read() # data from device
  api_key = bottle.request.get_header("Api_Key") # api key from header

  # outputs to console received data for debug reason
  print ">>> {} :: {}".format(value, api_key)

  # if api_key is correct and value is present
  # then writes attribute to point table
  if api_key == app.config["api_key"] and value:
    app.config["dsn"]["point"].insert(dict(ts=ts, value=value))
    status = 200

  # we only need to return status
  return bottle.HTTPResponse(status=status, body="")

# starting server on http://0.0.0.0:5000
if __name__ == "__main__":
  bottle.run(
    app = app,
    host = "0.0.0.0",
    port = 5000,
    debug = True,
    reloader = True,
    catchall = True,
  )

To run this simply go to folder containing python file and run python webapp.py from terminal. If everything goes ok you should have simple API available via POST method on /api route.

After testing the service with Restlet Client you should be able to view your data in a database file data.db.

You can also check the contents of new database file by using desktop client for SQLite → DB Browser for SQLite.

Table structure is as simple as it can be. We have ts (timestamp) and value (value from Arduino). As you can see timestamp is generated on API side. If you would happen to have atomic clock on Arduino it would be then better to generate and send timestamp with the value. This would be particularity useful if we would be collecting sensor data at a higher frequency and then sending this data in bulk to API.

If you will deploy this app with uWSGI and multi-threaded, use DSN (Data Source Name) url with ?check_same_thread=False.

Ok, now that we have some sort of a working API with some basic security so unwanted people can not post data to your database can we proceed further and try to program Arduino to send data to API.

Sending data to API with Arduino MKR1000

First of all you should have MKR1000 module and microUSB cable to proceed. If you have ever done any work with Arduino you should know that you also need Arduino IDE. On provided link you should be able to download and install IDE. Once that task is completed and you have successfully run blink example you should proceed to the next step.

In order to use wireless capabilities of MKR1000 you need to first install WiFi101 library in Arduino IDE. Please check before you install, you may already have it installed.

Code below is a working example that sends data to API. Before you try to test your code make sure you have run Python web application. Then change settings for wifi, api endpoint and api_key. If by some reason code bellow doesn't work for you please leave a comment and I'll try to help.

Once you have opened IDE and copied this code try to compile and upload it. Then open "Serial monitor" to see if any output is presented by Arduino.

#include <WiFi101.h>

// wifi settings
char ssid[] = "ssid-name";
char pass[] = "ssid-password";

// api server enpoint
char server[] = "192.168.6.22";
int port = 5000;

// api key that must be the same as the one in Python code
String api_key = "JtF2aUE5SGHfVJBCG5SH";

// frequency data is sent in ms - every 5 seconds
int timeout = 1000 * 5;

int status = WL_IDLE_STATUS;

void setup() {

  // initialize serial and wait for port to open:
  Serial.begin(9600);
  delay(1000);

  // check for the presence of the shield
  if (WiFi.status() == WL_NO_SHIELD) {
    Serial.println("WiFi shield not present");
    while (true);
  }

  // attempt to connect to wifi network
  while (status != WL_CONNECTED) {
    Serial.print("Attempting to connect to SSID: ");
    Serial.println(ssid);
    status = WiFi.begin(ssid, pass);
    // wait 10 seconds for connection
    delay(10000);
  }

  // output wifi status to serial monitor
  Serial.print("SSID: ");
  Serial.println(WiFi.SSID());

  IPAddress ip = WiFi.localIP();
  Serial.print("IP Address: ");
  Serial.println(ip);

  long rssi = WiFi.RSSI();
  Serial.print("signal strength (RSSI):");
  Serial.print(rssi);
  Serial.println(" dBm");
}

void loop() {
  WiFiClient client;

  if (client.connect(server, port)) {

    // I use random number generator for this example
    // but you can use analog or digital inputs from arduino
    String content = String(random(1000));

    client.println("POST /api HTTP/1.1");
    client.println("Connection: close");
    client.println("Api-Key: " + api_key);
    client.println("Content-Length: " + String(content.length()));
    client.println();
    client.println(content);

    delay(100);
    client.stop();
    Serial.println("Data sent successfully ...");

  } else {
    Serial.println("Problem sending data ...");
  }

  // waits for x seconds and continue looping
  delay(timeout);
}

As seen from example you can notice that Arduino is generating random integer between [ 0 .. 1000 ]. You can easily replace this with a temperature sensor or any other kind of sensor.

Now that we have API under the hood and Arduino is sending demo data we can now focus on data visualization.

Data visualization

Before we continue we should examine our project folder structure. Currently we only have two files in our project:

simple-iot-app/

webapp.py
data.db

We will now add HTML template that will contain CSS and JavaScript code inline for the simplicity reason. And for the bottle framework to be able to scan root application folder for templates we will add bottle.TEMPLATE_PATH.insert(0, "./") in webapp.py. By default bottle framework uses views/ subfolder to store templates. This is not the ideal situation and if you will use bottle to develop web applications you should use native behavior and store templates in it's predefined folder. But for the sake of example we will over-ride this. Be careful to fully replace your code with new code that is provided below. Avoid partially replacing code in file :) Also new code for reading data-points is provided in Python example below.

First we add new route to our web application. It should be trigger when browser hits root of application http://0.0.0.0:5000/. This route will do nothing more than render frontend.html template. This is done by return bottle.template("frontend.html"). Check code below to further examine how exactly this is done.

Now we will expand /api route and use different methods to write or read data-points. For writing data-point we will use POST method and for reading points we will use GET method. GET method will return JSON object with latest readings and historical data.

There is a fantastic JavaScript library for plotting time-series charts called MetricsGraphics.js that is based on D3.js library for visualizing data.

Data schema required by MetricsGraphics.js → to achieve this we need to transform data from database into this format:

[
  {
    "date": "2017-08-11 01:07:20",
    "value": 933
  },
  {
    "date": "2017-08-11 01:07:30",
    "value": 743
  }
]

Web application is now complete and we only need frontend.html that we will develop now. If you would try to start web app now and go to root app this will return error because we don't have frontend.html yet.

# -*- coding: utf-8 -*-

import time
import bottle
import json
import datetime
import random
import dataset

# initializing bottle app
app = bottle.Bottle()

# adds root directory as template folder
bottle.TEMPLATE_PATH.insert(0, "./")

# connects to sqlite database
# check_same_thread=False allows using it in multi-threaded mode
app.config["db"] = dataset.connect("sqlite:///data.db?check_same_thread=False")

# api key that will be used in Arduino code
app.config["api_key"] = "JtF2aUE5SGHfVJBCG5SH"

# triggered when / is accessed from browser
# only accepts GET → no POST allowed
@app.route("/", method=["GET"])
def route_default():
  return bottle.template("frontend.html")

# triggered when /api is accessed from browser
# accepts POST and GET
@app.route("/api", method=["GET", "POST"])
def route_default():

  # if method is POST then we write datapoint
  if bottle.request.method == "POST":
    status = 400
    ts = int(time.time()) # current timestamp
    value = bottle.request.body.read() # data from device
    api_key = bottle.request.get_header("Api-Key") # api key from header

    # outputs to console recieved data for debug reason
    print ">>> {} :: {}".format(value, api_key)

    # if api_key is correct and value is present
    # then writes attribute to point table
    if api_key == app.config["api_key"] and value:
      app.config["db"]["point"].insert(dict(ts=ts, value=value))
      status = 200

      # we only need to return status
      return bottle.HTTPResponse(status=status, body="")

  # if method is GET then we read datapoint
  else:
    response = []
    datapoints = app.config["db"]["point"].all()

    for point in datapoints:
      response.append({
        "date": datetime.datetime.fromtimestamp(int(point["ts"])).strftime("%Y-%m-%d %H:%M:%S"),
        "value": point["value"]
      })

    bottle.response.content_type = "application/json"
    return json.dumps(response)

# starting server on http://0.0.0.0:5000
if __name__ == "__main__":
  bottle.run(
    app = app,
    host = "0.0.0.0",
    port = 5000,
    debug = True,
    reloader = True,
    catchall = True,
  )

And now finally we can implement frontend.html. Create file with this name and copy code below. When you are done you can start web application. Steps for this part are listed below the code.

<!DOCTYPE html>
<html>

  <head>
    <meta charset="utf-8">
    <title>Simple IOT application</title>
  </head>

  <body>

    <h1>Simple IOT application</h1>

    <div class="chart-placeholder">
      <div id="chart"></div>
    </div>

    <!-- application main script -->
    <script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/3.2.1/jquery.min.js"></script>
    <script src="https://cdnjs.cloudflare.com/ajax/libs/d3/4.10.0/d3.min.js"></script>
    <script src="https://cdnjs.cloudflare.com/ajax/libs/metrics-graphics/2.11.0/metricsgraphics.min.js"></script>
    <script>
      function fetch_and_render() {
        d3.json("/api", function(data) {
          data = MG.convert.date(data, "date", "%Y-%m-%d %H:%M:%S");
          MG.data_graphic({
            data: data,
            chart_type: "line",
            full_width: true,
            height: 270,
            target: document.getElementById("chart"),
            x_accessor: "date",
            y_accessor: "value"
          });
        });
      }
      window.onload = function() {
        // initial call for rendering
        fetch_and_render();

        // updates chart every 5 seconds
        setInterval(function() {
          fetch_and_render();
        }, 5000);
      }
    </script>

    <!-- application styles -->
    <style>
      body {
        font: 13px sans-serif;
        padding: 20px 50px;
      }
      .chart-placeholder {
        border: 2px solid #ccc;
        width: 100%;
        user-select: none;
      }
      /* chart styles */
      .mg-line1-color {
        stroke: red;
        stroke-width: 2;
      }
      .mg-main-area, .mg-main-line {
        fill: #fff;
      }
      .mg-x-axis line, .mg-y-axis line {
        stroke: #b3b2b2;
        stroke-width: 1px;
      }
    </style>

  </body>

</html>

Now the folder structure should look like:

simple-iot-app/

webapp.py
data.db
frontend.html

Ok, lets now start application and start feeding it data.

python webapp.py
connect Arduino MKR1000 to power source
open browser and go to http://0.0.0.0:5000

If everything goes well you should be seeing new data-points rendered on chart every 5 seconds.

If you navigate to http://0.0.0.0:5000 you should see rendered chart as shown on picture below.

Complete application with all the code is available for download.

Conclusion

I hope this clarifies some aspects of IOT application development. Of course this is a minimal example and is far from what can be done in real life with some further dive into other technologies.

If you would like to continue exploring IOT world here are some interesting resources for you to examine:

Any comment or additional ideas are welcomed in comments below.

Profiling Python web applications with visual tools

Fri, 21 Apr 2017 12:00:00 +0200

I have been profiling my software with KCachegrind for a long time now and I was missing this option when I am developing API's or other web services. I always knew that this is possible but never really took the time and dive into it.

Before we begin there are some requirements. We will need to:

implement cProfile into our web app,
convert output to callgrind format with pyprof2calltree,
visualize data with KCachegrind or Profiling Viewer.

If you are using MacOS you should check out Profiling Viewer or MacCallGrind.

We will be dividing this post into two main categories:

writing simple web-service,
visualize profile of this web-service.

Simple web-service

Let's use virtualenv so we won't pollute our base system. If you don't have virtualenv installed on your system you can install it with pip command.

# let's install virtualenv globally
$ sudo pip install virtualenv

# let's also install pyprof2calltree globally
$ sudo pip install pyprof2calltree

# now we create project
$ mkdir demo-project
$ cd demo-project/

# now let's create folder where we will store profiles
$ mkdir prof

# now we create empty virtualenv in venv/ folder
$ virtualenv --no-site-packages venv

# we now need to activate virtualenv
$ source venv/bin/activate

# you can check if virtualenv was correctly initialized by
# checking where your python interpreter is located
# if command bellow points to your created directory and not some
# system dir like /usr/bin/python then everything is fine
$ which python

# we can check now if all is good ➜ if ok couple of
# lines will be displayed
$ pip freeze
# appdirs==1.4.3
# packaging==16.8
# pyparsing==2.2.0
# six==1.10.0

# now we are ready to install bottlepy ➜ web micro-framework
$ pip install bottle

# you can deactivate virtualenv but you will then go
# under system domain ➜ for now don't deactivate
$ deactivate

We are now ready to write simple web service. Let's create file app.py and paste code bellow in this newly created file.

# -*- coding: utf-8 -*-

import bottle
import random
import cProfile

app = bottle.Bottle()

# this function is a decorator and encapsulates function
# and performs profiling and then saves it to subfolder
# prof/function-name.prof
# in our example only awesome_random_number function will
# be profiled because it has do_cprofile defined
def do_cprofile(func):
  def profiled_func(*args, **kwargs):
    profile = cProfile.Profile()
    try:
      profile.enable()
      result = func(*args, **kwargs)
      profile.disable()
      return result
    finally:
      profile.dump_stats("prof/" + str(func.__name__) + ".prof")
  return profiled_func


# we use profiling over specific function with including
# @do_cprofile above function declaration
@app.route("/")
@do_cprofile
def awesome_random_number():
  awesome_random_number = random.randint(0, 100)
  return "awesome random number is " + str(awesome_random_number)

@app.route("/test")
def test():
  return "dummy test"

if __name__ == '__main__':
  bottle.run(
    app = app,
    host = "0.0.0.0",
    port = 4000
  )

# run with 'python app.py'
# open browser 'http://0.0.0.0:4000'

When browser hits awesome_random_number() function profile is created in prof/ subfolder.

Visualize profile

Now let's create callgrind format from this cProfile output.

$ cd prof/
$ pyprof2calltree -i awesome_random_number.prof
# this creates 'awesome_random_number.prof.log' file in the same folder

This file can be opened with visualizing tools listed above. In this case we will be using Profilling Viewer under MacOS. You can open image in new tab. As you can see from this example there is hierarchy of execution order of your code.

Make sure you convert output of the cProfile output every time you want to refresh and take a look at your possible optimizations because cProfile updates .prof file every time browser hits the function.

This is just a simple example but when you are developing real-life applications this can be very illuminating, especially to see which parts of your code are bottlenecks and need to be optimized.

Update 2017-04-22

Reddit user mvt also recommended this awesome web based profile visualizer SnakeViz that directly takes output from cProfile module.

Comment from discussion Profiling Python web applications with visual tools.

# let's install it globally as well
$ sudo pip install snakeviz

# now let's visualize
$ cd prof/
$ snakeviz awesome_random_number.prof
# this automatically opens browser window and
# shows visualized profile

Reddit user ccharles suggested a better way for installing pip software by targeting user level instead of using sudo.

Comment from discussion Profiling Python web applications with visual tools.

# now we need to add this path to our $PATH variable
# we do this my adding this line at the end of your
# ~/.bashrc file
PATH=$PATH:$HOME/.local/bin/

# in order to use this new configuration you can close
# and reopen terminal or reload .bashrc file
$ source ~/.bashrc

# now let's test if new directory is present in $PATH
$ echo $PATH

# now we can install on user level by adding --user
# without use of sudo
$ pip install snakeviz --user

Or as suggested by mvt you can use pipsi.

What I've learned developing ad server

Mon, 17 Apr 2017 12:00:00 +0200

For the past year and half I have been developing native advertising server that contextually matches ads and displays them in different template forms on variety of websites. This project grew from serving thousands of ads per day to millions.

The system is made from couple of core components:

API for serving ads,
Utils - cronjobs and queue management tools,
Dashboard UI.

Initial release was using MongoDB for full-text search but was later replaced by Elasticsearch for better CPU utilization and better search performance. This provided us with many amazing functionalities of Elasticsearch. You should check it out if you do any search related operations.

Because the premise of the server is to provide native ad experience, they are rendered on the client side via simple templating engine. This ensures that ads can be displayed number of different ways based on the visual style of the page. And this makes JavaScript client library quite complex.

So now that you know basic information about the product lets get into the lessons we learned.

Aggregate everything

After beta version was released everything (impressions, clicks, etc) was written in nanosecond resolution in the database. At that time we were using PostgreSQL and database quickly grew way above 200GB in disk space. And that was problematic. Statistics took disturbingly long time to aggregate. Also using indexes on stats table in database was no help after we reached 500 million datapoints.

There is a marketing product information and there is real life experience. And the tend to be quite the opposite.

This was the reason that now everything is aggregated on daily basis and this data is then fed to Elastic in form of daily summary. With this we achieved we can now track many more dimensions such as zone, channel and platform information. And with this information we can now adapt occurrences of ads on specific places more precisely.

We have also adapted Redis as a full-time citizen in our stack. Because Redis also stores information on a local disk we have some sort of backup if server would accidentally suffer some failure.

All the real-time statistics for ad serving and redirecting is presented as counters in Redis instance and daily extracted and pushed to Elastic.

Measure everything

The thing about software is that we really don't know how well it is performing under load until such load is presented. When testing locally everything is fine but when on production things tend to fall apart.

As a solution for this we are measuring everything we can. Function execution time (by encapsulating functions with timers), server performance (cpu, memory, disk, etc), Nginx and uWSGI performance. We sacrifice a bit of performance for the sake of this information. And we store all this information for later analysis.

Example of function execution time

{
  "get_final_filtered_ads": {
    "counter": 1931250,
    "avg": 0.0066143431,
    "elapsed": 12773.9500310003
  },
  "store_keywords_statistics": {
    "counter": 1931011,
    "avg": 0.0004605267,
    "elapsed": 889.2821669996
  },
  "match_by_context": {
    "counter": 1931011,
    "avg": 0.0055960716,
    "elapsed": 10806.0758889999
  },
  "match_by_high_performance": {
    "counter": 262,
    "avg": 0.0152770229,
    "elapsed": 4.00258
  },
  "store_impression_stats": {
    "counter": 1931250,
    "avg": 0.0006189991,
    "elapsed": 1195.4419869999
  }
}

We have also started profiling with cProfile and then visualizing with KCachegrind. This provides much more detailed look into code execution.

Cache control is your friend

Because we use Javascript library for rendering ads we rely on this script extensively and when in need we need to be able to change behavior of the script quickly.

In our case we can not simply replace javascript url in html code. It usually takes a day or two for the guys who maintain sites to change code or add ?ver=xxx attribute. And this makes rapid deployment and testing very difficult and time consuming. There is a limitation of how much you can test locally.

We are now in the process of integrating Google Tag Manager but couple of websites are developed on ASP.net platform that have some problems with tag manager. With a solution below we are certain that we are serving latest version of the script.

And it only takes one mistake and users have the script cached and in case of caching it for 1 year you probably know where the problem is.

# nginx ➜ /etc/nginx/sites-available/default
location /static/ {
  alias /path-to-static-content/;
  autoindex off;
  charset utf-8;
  gzip on;
  gzip_types text/plain application/javascript application/x-javascript text/javascript text/xml text/css;
  location ~* \.(ico|gif|jpeg|jpg|png|woff|ttf|otf|svg|woff2|eot)$ {
    expires 1y;
    add_header Pragma public;
    add_header Cache-Control "public";
  }
  location ~* \.(css|js|txt)$ {
    expires 3600s;
    add_header Pragma public;
    add_header Cache-Control "public, must-revalidate";
  }
}

Also be careful when redirecting to url in your python code. We noticed that if we didn't precisely setup cache control and expire headers in response we didn't get the request on the server and therefore couldn't measure clicks. So when redirecting do as follows and there will be no problems.

# python ➜ bottlepy web micro-framework
response = bottle.HTTPResponse(status=302)
response.set_header("Cache-Control", "no-store, no-cache, must-revalidate")
response.set_header("Expires", "Thu, 01 Jan 1970 00:00:00 GMT")
response.set_header("Location", url)
return response

Cache control in browsers is quite aggressive and you need to be precise to avoid future problems. We learned that lesson the hard way.

Learn NGINX

When deciding on a web server we went with Nginx as a reverse proxy for our applications. We adapted micro-service oriented architecture early in the project to ensure when we scale we can easily add additional servers to our cluster. And Nginx was crucial to perform load balancing and static content delivery.

At first our config file was quite simple and later grew larger. After patching and adding new settings I sat down and learned more about the guts of Nginx. This proved to be very useful and we were able to squeeze much more out of our setup. So I advise you to take your time and read through the documentation. This saved us a lot of headache. Googling for solutions only goes so far.

Use Redis/Memcached

As explained above we are using caching basically for everything. It is the corner stone of our services. At first we were very careful about the quantity of things we stored in Redis. But we later found out that the memory footprint is very low even when storing large amount of data in it.

So we gradually increased our usage to caching whole HTML outputs of dashboard. This improved our performance in order of magnitude. And by using native TTL support this goes hand in hand with our needs.

The reason why we choose Redis over Memcached was the nature of scalability of Redis out of the box. But all this can be achieved with Memcached.

Conclusion

There are a lot more details that could have been written and every single topic in here deserves it's own post but you probably got the idea about the problems we faced.

Golang profiling simplified

Tue, 07 Mar 2017 12:00:00 +0200

Many posts have been written regarding profiling in Golang and I haven’t found proper tutorial regarding this. Almost all of them are missing some part of important information and it gets pretty frustrating when you have a deadline and are not finding simple distilled solution.

Nevertheless, after searching and experimenting I have found a solution that works for me and probably should also for you.

Where are my pprof files?

By default pprof files are generated in /tmp/ folder. You can override folder where this files are generated programmatically in your golang code as we will see below in example.

Why is my CPU profile empty?

I have found out that sometimes CPU profile is empty because program was not executing long enough. Programs, that execute too quickly don’t produce pprof file in my cases. Well, file is generated but only contains 4KB of information.

Profiling

As you can see from examples we are executing dummy_benchmark functions to ensure some sort of execution. Memory profiling can be done without such a “complex” function. But CPU profiling needs it.

Both memory and CPU profiling examples are almost the same. Only parameters in main function when calling profile.Start are different. When we set profile.ProfilePath(“.”) we tell profiler to store pprof files in the same folder as our program.

Memory profiling

package main

import (
  "fmt"
  "time"
  "github.com/pkg/profile"
)

func dummy_benchmark() {

  fmt.Println("first set ...")
  for i := 0; i < 918231333; i++ {
    i *= 2
    i /= 2
  }

  <-time.After(time.Second*3)

  fmt.Println("sencond set ...")
  for i := 0; i < 9182312232; i++ {
    i *= 2
    i /= 2
  }
}

func main() {
  defer profile.Start(profile.MemProfile, profile.ProfilePath("."), profile.NoShutdownHook).Stop()
  dummy_benchmark()
}

CPU profiling

package main

import (
  "fmt"
  "time"
  "github.com/pkg/profile"
)

func dummy_benchmark() {

  fmt.Println("first set ...")
  for i := 0; i < 918231333; i++ {
    i *= 2
    i /= 2
  }

  <-time.After(time.Second*3)

  fmt.Println("sencond set ...")
  for i := 0; i < 9182312232; i++ {
    i *= 2
    i /= 2
  }
}

func main() {
  defer profile.Start(profile.CPUProfile, profile.ProfilePath("."), profile.NoShutdownHook).Stop()
  dummy_benchmark()
}

Generating profiling reports

# memory profiling
go build mem.go
./mem
go tool pprof -pdf ./mem mem.pprof > mem.pdf

# cpu profiling
go build cpu.go
./cpu
go tool pprof -pdf ./cpu cpu.pprof > cpu.pdf

This will generate PDF document with visualized profile.

Wireless sensor networks

Thu, 24 Oct 2013 12:00:00 +0200

Zigbee networks have this wonderful capability to self-heal, which means they can reorder connections between them if one of them is inoperable. This works our of the box when you deploy them. But you have to have in mind that achieving this is not as easy as you would think. None of it is plug&play. So to make your life a bit easier, here are some pointers which, I hope, will help you.

Be careful when you are ordering your equipment abroad. There are many rules and regulations you need to comply before you get your Xbee radios. What they do is they wait until you prove that you won’t use the technology for some kind of evil take over control of the world project :). For this, they have EAR (Export Administration Regulations) which basically means “This product may require a license to export from the United States.”.
I don’t know if this applies for every country, but when we purchased our Xbee radios from Mouser, this was mandatory! What we needed to do was to print out a form and write information about our company and send them a copy via email. With this document, we proved that we are a legitimate company.
When you complete your purchase and send all the documentation, you are not clear yet. Then customs will take it from there :). There will be some additional costs. Before purchasing, make sure you have as much information about costs as possible. Because it can get costly in the end.
I suggest you use companies from your country. You can seriously cut your costs. Here in Slovenia, the best option so far as I know is Farnell. And based on my personal experience, they rock! All I need to say!
Make plans when ordering larger quantities. Do not, I say, do not make your orders in December! :) Believe me! You will have problems with stock they can provide for you. So, we were forced to buy some things from Mouser, which was extremely painful because of all the regulations you need to obey when importing goods from the USA.
Make sure that firmware version on your Xbee radios is exactly the same! Do not get creative!!! I propose using templates. You can get template by exporting settings/profile in X-CTU application. Make sure you have enabled “Upgrade firmware” so you can be sure each radio has the same firmware.
And again: make plans! Plan everything! In months advanced! You will thank me later :)
Test, test, test. Wireless networks can be tricky.

If you are serious, I suggest you buy this book, Building Wireless Sensor Networks. You will get a glimpse of how networks work in lumens terms. It is a good starting point for everybody who wants to build wireless networks.

Additional resources:

LED technology might not be as eco-friendly as you think

Fri, 09 Mar 2012 12:00:00 +0200

There is a lot of talk about LED technology. It is beginning to infiltrate industry at a fast rate, and it’s a challenge for designers and also engineers. I wondered when a weakness will be revealed. Then I stomped on an article talking about harm in using LED technology. It looks like this magical technology is not so magical and eco-friendly.

A new study from the University of California indicates that LED lights contain toxic metals, and should be produced, used and disposed of carefully. Besides the lead and nickel, the bulbs and their associated parts were also found to contain arsenic, copper, and other metals that have been linked to different cancers, neurological damage, kidney disease, hypertension, skin rashes and other illnesses in humans, and to ecological damage in waterways.

Since then, I haven’t yet found any regulation for disposal of LED lights or any other regulation or standard. This might be a problem in the future. And it is a massive drawback. This might have quite an impact on consumer market.

Nevertheless, there is a potential, and I am sure the market will adapt. I also hope I will be reading documents regarding solution for this concern soon.

Additional resources:

Most likely to succeed in the year of 2011

Thu, 13 Jan 2011 12:00:00 +0200

The year of 2010 was definitely the year of Geo-location. The market responded beautifully and lots of very cool services were launched. We all have to thank the mobile market for such extensive adoption. With new generations of mobile phones that are not only buffed with high-tech hardware but are also affordable. We can now manage tasks that were not so long time ago, almost Star Trek’ish. And all this had and has great influence on the destination to which we are going now.

Reading all this articles about new innovation about new thriving technologies makes me wonder what’s the next step. The future is the mesh, like Lisa Gansky said in her book The Mesh.

Many still have conservative views on distributed systems. The problems with security of information. Fear of not controlling every aspect of information flow. I am very opened to distributed systems and heterogeneous applications, and I think this is the correct and best way to proceed.

This year will definitely be about communication platforms. Mobile to mobile. Machine to mobile and vice versa. All the tech is available and ready to put into action. Wireless is today’s new mantra. And the concept of semantic web is now ready for industry.

Applications and developers now can gain access to new layers of systems and can prepare and build solutions to meet the high quality needs of market. The speed is everything now.

My vote goes to “Machine to Machine” and “Embedded Systems”!

Mitja Felicijan's posts

Using Address Sanitizer with clang

What -fsanitize=address does

An example how to use it

Change default applications on Linux from terminal

Location and structure of .desktop files

Query current associations

Set default application

Interfacing with C

Reading material

Integrated GDB Debugging in Vim with TermDebug

Motivation

Main goals and requirements

A quick demonstration

Main ~/.vimrc

Project ~/project/foo/.vimrc

Why use :term and :TermDebug

Isn't this setup a bit elaborate?

Minimal desktop setup with cwm and conky

Change Alt key to Win key under Xfce4

Lua performance in different environments

Naive and intentionally slow implementation of Fibonacci sequence

Embedded Lua primer

Quick results

Test script

Embedding game assets within your executable binary

Why?

Project structure

Main game loop

Converting assets to C header files

Embedding and compiling

Honorable mention: C23-embed-directive

Credits

Currated list of Vim ALE linters

Reading material

Linters

Male and female body proportion reference images

Use option key as meta in Alacritty under macOS

Write and read structs to/from files in C

Embedding resources into binary with C

Sending signals to C programs

Calling assembly functions from C

The abysmal state of GNU/Linux and a case against shared object libraries

Personal critique

Debian

Top 20 packages by size

Number of packages installed

Number of shared object libraries on system (*.so)

Shared objects sorted by size

Fedora

Top 20 packages by size

Number of packages installed

Number of shared object libraries on system (*.so)

Shared objects sorted by size

Quick observation

So, now what?

Using custom software with Github Actions to deploy a site

Converting Valgrind callgrinds to SVG format

Uninstall Ollama from a Linux box

Sane default for Dungeon Crawl Stone Soup Online edition

Extract lines from a file with sed

Personal sane Vim defaults

Add lazy loading of images in Jekyll posts

Using ffmpeg to combine videos side by side

Compile drawterm on Fedora 38

AWS EB PyYAML fix

Make B/W SVG charts with matplotlib

Set color temperature of displays on i3

Fix screen tearing on Debian 12 Xorg and i3

Online radio streaming with MPV from terminal

60's IBM Computers Commercial

10/GUI 10 Finger Multitouch User Interface

Alacritty open links with modifier

Development environments with Nix

Making cgit look nicer

Simple presentations with Markdown

Bulk thumbnails

Edsger W. Dijkstra Manuscripts ebook

Extending dte editor

Grep to Less that maintain colors

Location and structure of `.desktop` files

Main `~/.vimrc`

Project `~/project/foo/.vimrc`

Why use `:term` and `:TermDebug`