David Neto
David Neto
I socialized this at the 2022-08-23 meeting.
> I would recommend that we spec that the indexed object is materialized before the dynamic index. That makes sense. For example, in the vector case, there is one type...
Same probably applies to builtins. Eg pow(10000,10000) should be shader-creation error because it overflows f64
Yeah, the section is a conceptual spaghetti at the moment. Now that module-scope let is replaced by 'const', we can reorg to make things much more clear. - const decl...
Also applies to inverse trig functions.
Also, think about pulling constraints together. E.g. - vertex shader must have a builtin position output, but that rule is in Function restrictions section - compute shader must not have...
I think this patch has to say that these are predeclared types. So they are actually identifiers that do resolve in the normal way.
Note that this proposal would lead to some inconsistency between const-expr evaluation and runtime-expression evaluation. E.g. ``` fn foo() { var one: i32 = 1; let runtime_div_by_zero = one /...
I very much agree with @jimblandy 's "domain error" approach. This is very well aligned with existing practice in the C/C++ standard library. Eg. many descriptions for math functions on...
In the above, I did not distinguish between +0.0 and -0.0