Fixed clippy errors and applied the suggestion from @teoxoy. Will await consensus from @jimblandy before moving forward with whether to keep the flag. I don't mind either way, but the idea to have the flag stemmed from concerns in #4330 where OP seemed to have reservations about the injected polyfill being "quite long".

@jimblandy Made the emitted polyfills start with a double underscore and further namespaced them as __naga_polyfill. I've had to add an exception to wgsl-in to allow __naga_polyfill in the lexer otherwise wgsl-out would output WGSL that wgsl-in would reject.

@chyyran No, you're right, double underscore isn't something that the code that we generate can define. We should be producing WGSL that other WGSL implementations would accept. Sorry for the bad advice.

I think instead you should change the polyfill names back, so they start with _naga again, and just add "_naga" to the list of reserved prefixes that the WGSL back end passes to Namer::reset. The namer will take care of consistently adjusting any identifiers coming from the naga::Module, so that should just work. Then we can use _naga in the WGSL backend for any future injected stuff.

@jimblandy dropped the changes to add the double underscore and added _naga to the list of reserved prefixes in Namer in 8309067

@chyyran it looks like the snapshots need updating.

Looks like its sensitive to insertion order. I’ll switch it to use an indexset.

@teoxoy switched required_polyfills to an indexset in a0a81ac, could you rerun workflows?