Provide aliases for patchES300

[vorg]

Current approach is verbose and error prone

ctx.pipeline({
   vert: parser.patchES300("..."),
   frag: parser.patchES300("...", "fragment")
})

The alternative approach is much more readable

ctx.pipeline({
   vert: parser.patchVert("..."),
   frag: parser.patchFrag("...")
})

Downsides:

• assumes glsl (could be wglsl one day)
• function name doesn't imply that it is for webgl2/ES300

But for comparision i don't think this is better

ctx.pipeline({
   vert: parser.patchVertES300("..."),
   frag: parser.patchFragES300("...")
})

Maybe it should be the parser that is typed?

import { parser } from "pex-shaders";

to

import { es3000Parser as parser } from "pex-shaders";

then you could switch language by switching import (in my dreams..)

[vorg]

One could as why do we even have to bother and why pex-context is not auto-upgrading it for us?

[vorg]

Another option build a online tool when we can upgrade our code by copy paste and stop using parser.

What do you think @dmnsgn

[dmnsgn]

I think parser.patchES300 is purposely verbose as it should be a relic from the past by now and we are willing to remove it entirely.

The current solution (as of Jan 2024) is:

• provide a shader written in GLSL 1.00 (WebGL 1 only) with output.vert/frag chunks that allows the shader to be used seamlessly with GLSL version 3 (WebGL2) https://github.com/pex-gl/pex-shaders/blob/6e2cd8fe97fbda9331e0e8f1b2e6ef6275468f45/shaders/chunks/output.glsl.js#L7-L13
• use parser.build(ctx, src, defines = [], extensions = {}) to add #version, defines and extension

That works beautifully in pex-renderer v4 alpha and has the following pros:

• one source of truth, write shader once
• familiarity of attribute/varying
• no string replace except adding defines/extension in parser.build (which means greater performance improvement when diffing shader code): what you write is what you get
• MRT (WebGL2 multiple color attachments) made easy with define (again perf improvement by reducing GPU memory pressure) https://github.com/pex-gl/pex-shaders/blob/6e2cd8fe97fbda9331e0e8f1b2e6ef6275468f45/shaders/chunks/output.glsl.js#L56-L58
• easy to see what code is there for #if (VERSION >= 300) and check new https://github.com/pex-gl/pex-shaders/blob/6e2cd8fe97fbda9331e0e8f1b2e6ef6275468f45/shaders/chunks/output.glsl.js#L27-L36

And cons:

• written in "old school" language (although so many more snippets out there are written in 1 and not 3), not all features (3d textures, transform feedback etc.)
• shaders code slightly bigger (although we use template strings so not a biggy)
• WGSL doesn't support #define (at least last time I checked)

That's for current status.

We reached the conclusion that –now that Safari caught up– WebGL 2 has enough worldwide support so we should entirely drop WebGL1. In a soonish future that means:

• ditching all the tricks above (parser output chunks, VERSION checks...)
• rewrite shaders to use in/out instead of attribute/varying and only use that syntax
• rely on non WebGL fallback if any device still doesn't support WebGL2
• enjoy

PS: I still don't see any obvious GLSL -> WGSL bridge without DSL extracting shader variables and providing different body like I did in dgel:

https://github.com/dmnsgn/dgel/blob/b3a213d636a3327eb80a89a93344a4f852c25db2/examples/cube.js#L159-L194

[vorg]

Ok. I missed that defines SOTA. It kind of brings back the question of Shift from framework to library #9 as those defines are in final shaders that could be in pex-renderer instead of here.

Re. glsl/wgsl we could got further into thi.ng/umbrella and copy ThreeJS approach with DSL in JS/TS. https://github.com/thi-ng/umbrella/tree/develop/packages/shader-ast. There is no WGSL front end yet, only glsl.

const sinc = (x: FloatTerm, k: FloatTerm) =>
    div(sin(mul(x,k)), mul(x, k));

Not sure how does it feel to write whole PBR shader in this though... Seems like 6 months project.