search

cbaggers / varjo

Lisp to GLSL Language Translator
BSD 2-Clause "Simplified" License
223 stars 23 forks source link

stemcells checks if symbol is bound which is undesirable for some use-cases #203

Closed mfiano closed 6 years ago

mfiano commented 6 years ago

suitable-symbol-for-stemcellp is checking for boundp, forcing me to use keyword symbols in vari for my uniform system to work correctly.

cbaggers commented 6 years ago

The stemcell thing was added to allow injection of bound 'global' vars as uniforms and to let a form's type be inferred from surrounding code (this is a bad way of doing it but that was the idea)

Allowing unbound wasnt part of the design so this is definitely needed in the current design.

If I allow unbound symbols then every variable typo becomes a stemcell which means that the compiler is way less useful as it never checks if a variable is bound, it just assumes it's a stemcell.

Previously we had the additional restriction that a injected variables would have to have earmuffs but that was removed at user request a while ago.

mfiano commented 6 years ago

What about (setf allow-unbound-stemcell-variables t)? The code would be tiny to support this.

djeis97 commented 6 years ago

Alternately, perhaps that check could be moved to CEPL? That would preserve the current behavior CEPL-side, where free variables are expected to refer to bound specials, but allow other wrappers around varjo to make their own decisions on this front.

djeis97 commented 6 years ago

If no stemcell type interence function is provided by a varjo user than any free variable is an error, IIRC, so it’s not like a direct varjo user could accidentally stumble into free variable weirdness without first learning something of how stemcells work, at which point it becomes a documentation problem.

cbaggers commented 6 years ago

Heya @djeis97, moving the logic sounds more sensible for sure. Maybe we can change it so that if the stemcell hook returns nil then varjo reports it as a unknown variable. I'd need to look into this as there is some old code there that might be relying on some dumb behavior.

Thanks for chipping in!

cbaggers commented 6 years ago

@djeis97 Done. Varjo no longer checks that the symbol is bound and instead will report it being undefined if the infer hook function returns nil.

Has been pushed to master. I hope this helps :)

cbaggers commented 6 years ago

Closing for now, but please reopen if this doesnt cover your use case