Use-before-def problem in queried argument values (when array types are involved?) about chapel

[damianmoz]

This works:

proc fftplanfactors(n : uint(?w), ufr : [?uD] uint(w))
{
   // use uint(w) within the routine
...
}

This fails:

proc fftplanfactors(ufr : [?uD] uint(?w), n : uint(w))
{
   // use uint(w) within the routine
...
}

Hope this helps.

[dlongnecke-cray]

It's kind of silly, but I'm just going to extend the current cloning we do today to apply to [?d] int(?w) as well. In the future when the new frontend is the resolver we won't have to do any of this nonsense.

[dlongnecke-cray]

It looks like fixupArrayFormal in normalize is just straight up deleting the type query?! What the heck...

It's a strange order of operations error... fixupArrayFormal is supposed to replace any mentions of the array element type with arr.eltType, but that deletes the type query...

We currently do this strange "stamping out" in the production compiler where any time we encounter int(?w) we stamp out overloads where w is replaced by each int width. However, this doesn't seem to happen properly when the int(?w) is an array formal's element type e.g., [?d] int(?w). Is this just because this was a pattern we couldn't write in previous releases?

The question is whether or not we stamp out overloads for this array type case as well or if we just leave it generic and see what happens...

[bradcray]

Thanks for filing this @damianmoz, and for improving the state of things, @dlongnecke-cray! Damian, if you get a chance to try out the fix in your code, please let us know if you're seeing the expected behavior now.

[damianmoz]

Next week. Thanks