Hi, I think this is a useful mod for the ones that want to completely disable the frame and the black borders. I also made some cleaning in the function tsample removing the parameters offs and resolution because they're not used in function.

This PR adds full WebAssembly support to dos-like.

All the samples are available to be viewed/played on the github-io page of my repo: https://schellingb.github.io/dos-like/

Before merging, there are a few things worth to maybe have a discussion about.

There are no real threads in WebAssembly readily available so I chose to use coroutines to emulate the threaded nature of native dos-like to keep the API intact while being a regular single threaded JavaScript application on the web.

The way I kept the amount of #ifdef __wasm__ in dos.h as low as I could was to add a wasm platform to thread.h with mostly fake functions (where threads/mutex do nothing and atomics are just regular variables). I contemplated not including thread.h on the wasm build target and adding #ifndef __wasm__ around all the concurrency control code in dos.h. Might result in a cleaner implementation overall?

We talked on Twitter about how, due to the coroutine nature, the users main loop must somehow end up calling waitvbl from time to time to switch the context back to dos.h and how I added forced waitvbl calls to swapbuffers/readkeys/readchars. I can't really think of a better alternative so hopefully that's fine. Maybe as an improvement it could detect/remember that waitvbl isn't being called but swapbuffers is, and then do waitvbl every time instead of only every third time swapbuffers is called?

The JavaScript code embedded in the new WASM part of app.h is not adhering to the code style of the C code around it. Could be made matching the C style or could be left as is because it is not C after all.

Another thing is the "wasm" directory I added to the repository root. Could be considered annoying to people who aren't interested in it. As explained in the README addition, It needs a bunch of things (llvm, node, wajic) anyway so it could be kept external as well.

Speaking of, the README should probably link to the deployed web versions. I didn't want to add absolute links to my github-io pages or github repo so my changes use relative links only.

That's about it of things I can think of. If there's any request for change or concern, let me know!

I've tested all three supported types of builds against the changes in this pull request (building with tcc on Linux with Wine, building on Linux, and even building both Apple Silicon and Intel Mac builds), and tested that all types of builds run successfully.

I changed how the Linux and Mac build commands receive SDL2, because the previous way SDL2 was provided didn't work on Apple Silicon using Homebrew, but the change fixes that. And using sdl2-config is the "right way" to use SDL2 with manually-written compile commands, because it provides some stuff SDL2 needs beyond -lSDL2.

In the process of fixing it for Mac, I created a CMake build script to help with development, as using CMake makes it easier to develop in many IDEs. Plus, it compiles the base dos-like library separately from the samples, so all of the samples compile faster.

This crashes on Linux for me (using a Xorg session, not Wayland), the issue seems unrelated to #2:

X Error of failed request:  BadWindow (invalid Window parameter)
  Major opcode of failed request:  40 (X_TranslateCoords)
  Resource id in failed request:  0x6005c9
  Serial number of failed request:  270
  Current serial number in output stream:  270

Commenting out SDL_HideWindow( app->window ); in dos.h fixed the issue for me, but not sure if it's a reasonable fix.

Example programs segfault on startup.

I've tracked it down to (I think) the exit_flag field in internal_t being uninitialised at the time of SDL initialisation.

int app_run( int (*app_proc)( app_t*, void* ), void* user_data, void* memctx, void* logctx, void* fatalctx )
    {
    app_t* app = (app_t*) APP_MALLOC( memctx, sizeof( app_t ) );
    memset( app, 0, (int)sizeof( app_t ) );
    app->memctx = memctx;
    app->logctx = logctx;
    app->fatalctx = fatalctx;
    app->interpolation = APP_INTERPOLATION_LINEAR;
    app->screenmode = APP_SCREENMODE_FULLSCREEN;

    int result = 0xff;

    if( SDL_Init( SDL_INIT_EVERYTHING ) < 0 )
    {
        goto init_failed;
    }

.
.
.

    result = app_proc( app, user_data );

This is the full stack trace (line numbers are wrong - my GDB version is out of date by the looks of it)

#0  thread_atomic_int_load (atomic=0x40) at source/libs/thread.h:999
#1  0x000055555555c522 in shutdown () at source/dos.h:441
#2  0x00007ffff771e8f2 in xcb_disconnect () from /usr/lib/x86_64-linux-gnu/libxcb.so.1
#3  0x00007ffff784fc2f in XCloseDisplay () from /usr/lib/x86_64-linux-gnu/libX11.so.6
#4  0x00007ffff7f217b9 in X11_Available () at /home/steve/src_local/SDL2-2.0.10/src/video/x11/SDL_x11video.c:101
#5  0x00007ffff7eee6d6 in SDL_VideoInit_REAL (driver_name=<optimized out>, [email protected]=0x0)
    at /home/steve/src_local/SDL2-2.0.10/src/video/SDL_video.c:501
#6  0x00007ffff7e3fcf7 in SDL_InitSubSystem_REAL (flags=62001) at /home/steve/src_local/SDL2-2.0.10/src/SDL.c:170
#7  0x00005555555648ab in app_run (app_proc=0x555555561c02 <app_proc>, user_data=0x7fffffffd580, memctx=0x0, 
    logctx=0x0, fatalctx=0x0) at source/libs/app.h:3374
#8  0x0000555555582267 in main (argc=1, argv=0x7fffffffd678) at source/dos.h:2864

If I alter shutdown() in dos.h to return 0, rather than atomic loading internal->exit_flag then the programs run file but that is obviously a poor solution.

Hello, I am trying to get some bindings for this going with LuaJIT. I've compiled dos.h using this:

gcc -shared -fPIC -o dos.so source/dos.c `sdl2-config --libs --cflags` -lGLEW -lGL -lm -lpthread

and I am trying to test some initial graphics functions with the following LuaJIT code:

local ffi = require "ffi"
ffi.cdef[[
   void circle( int x, int y, int r );
   enum videomode_t {
    videomode_40x25_8x8,
    videomode_40x25_9x16,
    videomode_80x25_8x8,
    videomode_80x25_8x16,
    videomode_80x25_9x16,
    videomode_80x43_8x8,
    videomode_80x50_8x8,
    videomode_320x200,
    videomode_320x240,
    videomode_320x400,
    videomode_640x200,
    videomode_640x350,
    videomode_640x400,
    videomode_640x480,
   };

   void setvideomode( enum videomode_t mode );
   void setdoublebuffer( int enabled );
]]

local dos = ffi.load("~/dos-like/libdos.so", true)

This works, but trying to call any dos functions fails with a segfault, assumedly because the initialization stuff isn't being called. Do you have any suggestions for connecting LuaJIT to the app_proc function?

This is mostly a copy'n'paste of loadwav(), calling drwav_open_memory_and_read_pcm_frames_s16() instead of drwav_open_file_and_read_pcm_frames_s16(). Might make sense to refactor and have only one internal implementation that just uses a buffer and read the file contents like is done in installuserfont() (and that "load the complete file into memory" might make sense to be in a separate function), but I'll leave that up to you :)

Without a license file or licensing information in README.md it is hard to know if how this can be used to create own games/demos/content. Some of the source files contains licensing info but it is conflicting and on a file-by-file basis. Would be good to clarify.

I was running a dos like game (stranded) and it was really loud, but I found that my volume up and volume down buttons on my keyboard did not work. This is the first time something like this has happened with my keyboard in 5 years of having it. Not sure how easy it is to test if you don't have media buttons on your own keyboard.

Just reporting it here in case it helps the project.

Can I have both FPIDE and Lazarus bindings for at least 64 bit Windows 10 as well as a target for 64-bit Linux as well for those willing to go with Free Pascal for linux? This should be a treat for those still having problems with the plain Free Pascal SDL2 package on Github.

I very much appreciate it.

Hi Mattias,

I fixed the readchars() SDL problem so I could extinguish the fire in the burn example program. The sound example (mentioned in #29) works now too. Tested on Ubuntu.

Sam

In the libs/app.h event processing for SDL, APP_INPUT_CHAR events were not being generated so dos.h's readchars() would never return any data. As a result, the burn and sound example programs would not respond to user input.

The SDL_TEXTINPUT SDL event appears to be the appropriate SDL event to drive the APP_INPUT_CHAR app events.

Should hopefully resolve #23.

• Documentation was added to each public facing symbol in dos.h. Most of it had already been written when I was working on dos-like-rs. Still, some validation on their accuracy would be much appreciated.
• Added Doxyfile configuration file, for easy documentation generation. All you have to do is install Doxygen and run doxygen.
• Updated .gitignore to ignore generated docs

Hi, love the library but it's kind of difficult to work out what all the functions do in dos.h without experimentation sometimes (yes - even with the source). It'd be nice to have comments in the file, or maybe even a standalone document that shows all the functions and their parameters.