LAGraph: Re-run failing tests with more verbose output and continue.

[mmuetzel]

Re-run failing tests with --rerun-failed --output-on-failure to get more verbose output that might help in tracking down the issue on 32-bit platforms.

Also continue even if ctest fails to make it behave more similar to the other projects.

[DrTimothyAldenDavis]

I see the bug. LAGraph_Malloc has a test that tries to avoid calling malloc when the request is too large:

SuiteSparse/LAGraph/src/utility/LAGraph_Malloc.c

Lines 42 to 48 in fd163d9

	size_t size ;
	bool ok = LG_Multiply_size_t (&size, nitems, size_of_item) ;
	if (!ok || nitems > GrB_INDEX_MAX || size_of_item > GrB_INDEX_MAX)
	{
	// overflow
	return (GrB_OUT_OF_MEMORY) ;
	}

but GrB_INDEX_MAX is always 2^60 on all platforms. On a 32-bit platform, size_t is 4 bytes, however, and the test_Malloc script does this on line 39-40, which is supposed to trigger that test and return p = NULL.

SuiteSparse/LAGraph/src/test/test_Malloc.c

Lines 39 to 40 in fd163d9

	LAGraph_Malloc ((void **) &p, GrB_INDEX_MAX + 1, sizeof (char), msg) ;
	TEST_CHECK (p == NULL) ;

That input parameter, GrB_INDEX_MAX+1, gets typecasted from uint64_t to size_t, which truncates it, probably to the value 1. So the call to LAGraph_Malloc succeeds, when it is supposed to fail.

I will need to look through all of LAGraph to make sure there's no other broken typecasting like this with size_t.