Errors thrown in parallel loop occasionally result in `SIGSEGV(11)/SIGBUS(10)` about chapel HOT 9 CLOSED

@pierce314159 : I need to apologize, this issue arrived at a busy time and simply slipped between the cracks. Can you let me know if it's a blocking issue for you?

@bmcdonald3 : Have you happened to have a chance to look into this issue at all?

from chapel.

@bradcray no worries! Nope this issue is non-blocking

I was able to workaround it by identifying the condition causing the error (an illegal cast from string to int when values are missing) and setting a boolean variable hadError to true in the loop iterations that would result in the bad cast. So after adding a with ( | reduce hadError) to the parallel loop, I can throw the error after the parallel loop completed

from chapel.

@bradcray no, I haven't looked into this yet, can give it a shot today.

from chapel.

Maybe not necessary since Pierce described it as non-blocking... I was just curious.

from chapel.

This crash is coming from this block of code in CSVmsg:

                            forall x in intersection {
                                var row = lines[x-offsets[file_idx]+data_offset].split(col_delim);
                                A[x] = row[colidx]: t;
                            }

Where the cast is calling Chapel's string casting with row[colidx]: t, which appears to be causing the issue.

When switching to a for instead of a forall, I am unable to reproduce the crash (but can consistently when it's a forall).

In the casting code, we are doing something like this:

...
var localX = x.localize();
...
retVal = c_string_to_int64_t(localX.c_str(), isErr);
...

So, I believe we are passing something we shouldn't be to these extern functions and that is causing the issue, but still speculation at this point.

from chapel.

The issue here seems to be unrelated to the string casting problem, I think it is actually throwing a TaskErrors from within a coforall. I've opened up #23905 to capture that.

Edit: I now realize that this was what Pierce was thinking all along, I was the one slow on the uptake here...

from chapel.

Potentially related issues (both using throws within foralls, I believe):

• #21058
• #18742

from chapel.

@vasslitvinov / @bmcdonald3 : Was this issue resolved by #23992 or does more remain here?

from chapel.

I confirmed that this was resolved, closing now.

from chapel.