Switch to SDL2, drop winit ASAP. Point numbering broken, console broken on Linux due to winit kbd implementation being bad.

[ctrlcctrlv]

This says it all:

https://github.com/rust-windowing/winit/blob/386ead15a3e3fef5c0c02bd7491d3cc9a7a6e0aa/src/platform_impl/linux/x11/events.rs#L178

//ffi::XK_numbersign => VirtualKeyCode::Numbersign

🤦‍♂️_{🤦‍♂️🤦‍♂️} 😱❗❗❗

Issue discovered by @eliheuer in his Twitch stream. Confirmed and cause found just now by @iwsfutcmd.

I don't trust any software with code like this commented out period. They have no idea what they're doing in my humble opinion. The fact that Subtract and Minus were only recently unified is shocking as well.

So, this is an urgent issue. I had hoped not to need to rewrite main.rs again, but I don't mind given the severity of their keyboard handling.

SDL2 supports Unicode input, so console will still work without much worry about hooking OS APIs for IME's etc.

[ctrlcctrlv]

Cf. rust-windowing/winit#1656, rust-windowing/winit#812, btw. These are years old issues, not new bugs.

[iwsfutcmd]

One suggestion is, when monitoring text input events, make sure to track SDL_TEXTINPUT rather than SDL_TEXTEDITING, especially for the console. Nothing's worse than a program ignoring my IME because it thinks it knows what to do with my keystrokes better than I do.

[ctrlcctrlv]

Yup, I made sure Japanese IME worked in current winit console and will do the same for SDL2. 😃

[ctrlcctrlv]

OK, I've sort of gotten it working. Sorry for the delay, was focused on Spiro, which is a 1.0 feature per #10. (Cf. fontforge/libspiro#27, fontforge/libspiro#28 (bottom))

[2020-09-21T11:07:57Z INFO  Qglif] KeyDown { timestamp: 16780, window_id: 1, keycode: Some(RShift), scancode: Some(RShift), keymod: RSHIFTMOD | NUMMOD, repeat: false }
[2020-09-21T11:07:57Z INFO  Qglif] KeyDown { timestamp: 16943, window_id: 1, keycode: Some(Num3), scancode: Some(Num3), keymod: RSHIFTMOD | NUMMOD, repeat: false }
[2020-09-21T11:07:57Z INFO  Qglif] TextInput { timestamp: 16943, window_id: 1, text: "#" }
[2020-09-21T11:07:57Z INFO  Qglif] KeyUp { timestamp: 17065, window_id: 1, keycode: Some(Num3), scancode: Some(Num3), keymod: RSHIFTMOD | NUMMOD, repeat: false }
[2020-09-21T11:07:57Z INFO  Qglif] KeyUp { timestamp: 17095, window_id: 1, keycode: Some(RShift), scancode: Some(RShift), keymod: NUMMOD, repeat: false }

Much better—seems consistent too.

I need to fix ImGui: the copied support.rs assumes winit. I also need to rewrite our event matching code, which shouldn't be too hard, and make en masse tweaks to the event handlers, which should be doable with a few vim commands.

I wonder if @aclysma would be interested in a support.rs for SDL2? Skulpin seems to be in maintenance mode so I'm not sure.

[ctrlcctrlv]

Oh, and this seems to be working too @iwsfutcmd :-)

[2020-09-21T11:19:49Z INFO  Qglif] TextInput { timestamp: 169569, window_id: 1, text: "アイ・エム・イー" }

[ctrlcctrlv]

The war is over! https://twitter.com/fr_brennan/status/1373849754715705344

@MatthewBlanchard figured out that we could solve all of our switchover problems by just drawing Imgui with Skia. He's writing a Skia-based renderer for Imgui, which we'll use. Then we can drop all of the hacks which allow us to draw @ocornut's stock Imgui Vulkan renderer on top of Skulpin (Skia+Vulkan).

This allows us to flip over to SDL extremely easily. And because SDL can be used from WebAssembly…that means that MFEKglif could, some day, be compiled for the web browser. Really! Amazing work.