DOS backend

[badlogic]

After adding the Web backend, I figured why not try DOS as well. This PR adds support for DOS, to be more specific, minifb applications compiled with DJGPP, a popular 32-bit DOS compiler based on GCC, running on DOS hardware or emulators that support VESA.

The backend has a couple of limitations:

• There can only ever be one window open at a time.
• Window sizes provided to mfb_window_open() must match a VESA mode available on the "machine" the app is running on. 320x240, 640x480, 800x600 work in DOSBox, but may not work on real hardware depending on the graphics card.
• Character input is limited to ASCII characters. A US keyboard layout is assumed.
• Key input is limited to the keys as found here: https://github.com/badlogic/minifb/blob/dos-pr/src/dos/DOSMiniFB.c#L18
• The buffer dimensions passed to mfb_update_ex() must either be a perfect match of the window dimensions, or 1/2 the window dimensions. E.g.:
  • Window 320x240, buffer 320x240: works
  • Window 640x480, buffer 320x240: works
  • Window 640x480, buffer 512x512: fails silently
• Viewports are not supported.

The current minifb tests do not adhere to these limitations, so I added tests/dos/dos.c.

To compile and run this test, you'll need DJGPP and DOSBox-x. The tests/dos/tools/download-dos-tools.sh script will download all the dependencies (and also a GDB version that understand COFF binaries) to tests/dos/tools/:

./tests/dos/tools/download-dos-tools.sh 

You can then build for DOS via:

cmake -DCMAKE_TOOLCHAIN_FILE=$(pwd)/tests/dos/tools/toolchain-djgpp.cmake -S . -B build
cmake --build build

You can run then run thedos.c test via DOSBox-x:

./tests/dos/tools/dosbox-x/dosbox-x -fastlaunch -exit build/dos.exe

This is what it looks like in DOSBox :)

I know this has a super low probability to be merged, but I figured someone else who wants to use minifb for little DOS experiments may find this PR and use it.

[Darky-Lucera]

I’m wondering why is this
Window 640x480, buffer 512x512: fails silently
happening?

[Darky-Lucera]

There are functions to scale content, with and without bilinear filtering

[badlogic]

The silent fail is merely due to me having only implemented a 1:1 blit and a very bad 1:2 blit.

You mean minifb has functions to scale buffers? I shall amend the PR then!

[badlogic]

Oh, how did I not see these before? Thanks for the pointer!

https://github.com/emoon/minifb/blob/master/src/MiniFB_internal.c

[badlogic]

Alrighty, I fixed it up using stretch_image(). Also adjusted dos.c to test if it actually works. noise.c, input_events and input_events_cpp work too now. The other tests do not, as they require multi-window support.

[Darky-Lucera]

Nice to know.

I'm a little afraid of maintaining so many backends in the future 😅

[badlogic]

I totally understand. For what it's worth, I'm actively using minifb in a bunch of projects that also require the web and DOS backend to keep working. I'd be happy to maintain them. If you want a track record of my OSS work, I'm the creator of libGDX, which has a track record of 12 years of maintenance :)

[Darky-Lucera]

Nice to know.
Thanks for the track record but I was sure you knew what you were doing.

[Darky-Lucera]

Let me know when this PR is finished (solve the conflicts)

[badlogic]

I've fixed up the merge conflicts. I still need to modify the README.md file to document usage and limitations, and also do some testing on Linux. I'll ping you here once I think it's ready.

[Darky-Lucera]

Ok

[badlogic]

Update README.md with (hopefully) extensive instructions. Tested everything on Windows, Linux, and macOS, including the debugging. Looks good to go on my end.

Note: on Windows you need Ninja as the build generator again, just like with Emscripten. Wished MSBuild could handle it :/

[badlogic]

Alright, I think I've fixed it all up well. Also tested on 86Box in vanilla MS-DOS 6.22, which does actual cycle accurate hardware emulation. Works well! I believe this is ready, if you think it's a nice addition to MiniFB.

[Darky-Lucera]

Excuse me for the long delay.
Could you please fix the problem in MakeLists.txt to merge this PR?

Thanks in advance!