Race Conditions and Deadlock in qarray Test With Thread Sanitizer

[insertinterestingnamehere]

The qarray test hangs with thread sanitizer (x86-64, nemesis, clang17, no topology detection).

Prior to hanging though, it also emits various thread sanitizer errors:

atomic write:

qthreads/src/ds/qarray.c

Line 738 in 619afe9

qthread_incr(arg->donecount, 1);

non-atomic read of same variable:

qthreads/src/ds/qarray.c

Line 1033 in 619afe9

while (donecount < maxsheps) {

In the test itself, non-atomic read:

qthreads/test/features/qarray.c

Line 110 in 619afe9

if (count != ELEMENT_COUNT) {

atomic write:

qthreads/test/features/qarray.c

Line 28 in 619afe9

qthread_incr(&count, 1);

Similar:

qthreads/test/features/qarray.c

Line 124 in 619afe9

if (elem[j] != 1) {

qthreads/test/features/qarray.c

Line 27 in 619afe9

memset(arg, 1, sizeof(bigobj));

Similar:

qthreads/test/features/qarray.c

Line 90 in 619afe9

double elem = *(double *)qarray_elem_nomigrate(a, i);

qthreads/test/features/qarray.c

Line 20 in 619afe9

*(double *)arg = 1.0;

Somewhat similar:
Non-atomic write:

qthreads/test/features/qarray.c

Line 104 in 619afe9

count = 0;

Atomic write:

qthreads/test/features/qarray.c

Line 21 in 619afe9

qthread_incr(&count, 1);

Non-atomic write:

qthreads/src/ds/qarray.c

Line 545 in 619afe9

FREE(a, sizeof(qarray));

Non-atomic read:

qthreads/src/ds/qarray.c

Line 706 in 619afe9

void *ptr = qarray_elem_nomigrate(arg->a, count + inpage_offset);

Note the read occurs inside the called function qarray_elem_nomigrate at:

qthreads/include/qthread/qarray.h

Line 114 in 619afe9

return a->base_ptr + ((segment_num * a->segment_bytes) + ((index - segment_num * a->segment_size) * a->unit_size));

Non-atomic write:

qthreads/src/ds/qarray.c

Line 982 in 619afe9

struct qarray_func_wrapper_args qfwa = { { NULL }, a, NULL, &donecount, startat, stopat };

Atomic write to same address:

qthreads/src/ds/qarray.c

Line 738 in 619afe9

qthread_incr(arg->donecount, 1);

[insertinterestingnamehere]

Debugged some more. It's not actually a deadlock. It's just that that particular test hits the thread sanitizer performance penalty really hard. Adjusting the problem size gets it down to a reasonable runtime.

[insertinterestingnamehere]

Closing in favor of #303