zig 0.17 support

[Yinameah]

Hello,

I'm writing because I've been descending DVUI's dependency list to try the improvements in incremental compilation with latest zig's "pre-release".

DVUI depends on raylib which depend on zemcripten.

In case you are not aware, upcoming zig release reworked consequently the build system. This is for the better, of course 😉 but in zemscripten case it's a non trivial change because your build script relies on direct (as []const u8) path manipulation and doing so in the latest build system iteration will not be feasible anymore (or a huge hack)

Noticeably, b.getInstallPath is no more, so this means not only it will be breaking for zemscripten consumers in any cases.

I've been doing build.zig updates quite a bit in the last few days so I want to propose my assistance on the matter. But since it's gonna be a bit intrusive changes, I wanted to reach out first before throwing some code at you ...

[hazeycode]

Thanks for reaching out. It's difficult to stay on top of maintenance churn of all the zig-gamedev libraries so any help is very welcome.
If you want to discuss anything in a more informal forum please feel free to join us on Discord https://discord.gg/JrYcW5sUD

[Yinameah]

Excellent.

Maybe you have a recommandation for a simple project "consuming" zemscripten that I could use as test vector.

I've done some preliminary work transitioning to Lazy Paths but raylib has a pretty heavy build script that is not ideal to hack on not being super familiar with emscripten in the first place.

My plan would be to make it built on latest zig (and hopefully compile incrementally) and then make raylib do the same. That way everybody wins.

[Yinameah]

But one concrete question maybe :

I'm confused with activateEmsdkStep function in the build script.

It does 3 things :

1. "register" calls to emsdk update/install/activate
2. "register" chmod for the two utilities used later ( emcc & emrun)
3. return a "top level step"

What confuses me is :

a) Is there a specific reason this "top level empty step" is declared in such a custom way ?
Shouldn't this just be const step = b.step("Activate EMSDK","nice description")

b) the dependency chain seems odd. I had to draw it to be sure I followed, and what I read is :

So I guess the activate step should depends on the main step, and the chmod of the main script be on the bottom of this graph. I'm not sure zig's build system garentee that the chmod will be run before the update otherwise. Does that seem correct ?

PS :
I tend to prefer taking time to think and write longer messages, and I haven't found Discord to encourage this in me. But if you feel this kind of questions would be bettor of there, let me know, I adapt gladly.

[Yinameah]

I have something that I think is working in #23.

I answered my own questions with the following :

a) I guess the step was declared that way to allow consumer to call the function multiple times and avoid running into "duplicated step" error.
But I could not find a way to translate this for 0.17 (.custom step is gone) and honestly I don't think the benefits outweigh the extra complexity. If the client script call this function, they get back a honest-to-got step that they can call independently, that seems fair to me.

b) I just assume it was mostly a typo that never caused issues. The new graph looks like this in my PR :

I did not manage (yet ?) to make raylib compile with theses changes so it's effectively not tested. Raylib's compile script is fairly complicated, and I belive it uses the API of zemscripten a bit backwards. I'll see if I manage to sort out something.

If you can point me to a demo repo that uses zemscripten, I'll test this stuff a bit further.

Cheers

[Yinameah]

Update :

I finally managed to make raylib work using the changes in #23

The issue was related with emsdk version bump lately done here while raylib was tracking a prior version, so unrelated to zig 0.17

So I belive it at least prove the concept.

I'm not sure I'll come back to this given my original goal is reached, but feel absolutely free to ping me if / when you have questions on the matter.

[Yinameah]

Gosh, I just noticed that I got the dependency tree wrong. Sorry for the noise.

Without cached files, the chmod of emrun & emcc.py faile because the files are not there yet. Will fix

[Yinameah]

Ok, the graph now looks like this :

Which hopefully is now correct