TST,BUG: Explicitly allow running tests multiple times

[HaoZeke]

These are extracted from the meson port effort [1] but are not directly tied to a build system. The main fixes involve:

• Calling abort over stop
  • Equally well supported (in terms of language standards / compilation), but stop reports no errors (i.e. return code was always 0), and doesn't include a backtrace.
• Changing the test write status from new to replace
  • Allows tests to run multiple times without a clean step in between

[1] https://github.com/HaoZeke/OpenBLAS/tree/mesonBasic

[HaoZeke]

M1 failure seems unrelated..

[martin-frbg]

@HaoZeke any objections to replacing your call abort with a standard-conforming error stop for portability ?

[HaoZeke]

@HaoZeke any objections to replacing your call abort with a standard-conforming error stop for portability ?

I use error stop [err code] in newer code, and it would generally be a step up.

For OpenBLAS though, I'm not sure since it is a Fortran 2008 standard feature, and might not be as widely supported. call abort is an extension (GNU, Intel) but it has been around a lot longer AFAIK.

EDIT: My bad, I hadn't had a copy of NAG lying around (licensing fees) so I had forgotten it wasn't supported. The same comment stands though, as noted in #5002 a backfill function I'd be happier with, unless we're sure OpenBLAS is always compiled with Fortran 2008 compliant compilers.

EDIT-2: Actually I'd have just recommended f90_unix_proc in #5003 (comment)