CHPL_UNWIND output stops at task boundaries

[bradcray]

When using CHPL_UNWIND, the stack trace stops once we hit a Chapel concept that introduces a task, such as a forall loop or an on-clause. This limits is usefulness when halt()s occur anywhere other than in the original task, which seems unfortunate in a parallel language. I'm wondering what it would take to have it report past those boundaries.

[ghost]

Hi @bradcray, while I was working on stack traces I had the same realisation: since tasks are (often) separated thread with their own stack, we don't have a way to find the stack frames from the parent task. One solution could be including, when we create a new task, the parent's stack trace but I don't know about the practicability of this idea (it shouldn't be too hard and memory-expensive, since we don't need the procedures names themselves, just their codes. I'm not entirely sure about performances however).

[bradcray]

I was speaking with @ronawho about this at some point (who works most closely with our tasking runtime these days) and he may have had some ideas, though I've forgotten what they are.

[ronawho]

I remember talking about this, but for I can't remember if we had any insights or ideas.