multicoretests Add gc tests

This PR adds initial tests of the Gc module - and in particular testing of Gc.compact.

Exercising Gc.counters already triggered a known issue locally, which has recently been fixed on trunk: https://github.com/ocaml/ocaml/pull/13370 I therefore expect the parallel test to fail on 5.2.0 and earlier.

This can be extended (and improved) in many ways adding (big)arrays, Weak, Ephemeron, custom blocks, finalizers, and dynamic Gc_ctrl changes.

Aug 27 '24 16:08 jmid

CI summary for ca8cf93:

11 5.2 workflows failed as expected due to the Gc test
- 32bit 5.2 failed with a segfault
- Bytecode aborted with Fatal error: allocation failure during minor GC
- Cygwin 5.2 aborted with Fatal error: allocation failure during minor GC
- FP 5.2 aborted with Fatal error: allocation failure during minor GC
- Linux 5.2 aborted with Fatal error: allocation failure during minor GC
- Linux 5.2 debug aborted with Fatal error: allocation failure during minor GC
- MinGW 5.2 exited with code -1073741819
- MinGW bytecode 5.2 exited with code -1073741819
- macOS-ARM64 5.2 failed with a segfault
- macOS-intel 5.2 got signal BUS
- freebsd-amd64-5.2 failed with a segfault
3 Cygwin/MinGW trunk workflows failed with a version mismatch after the 5.3 branching
- Cygwin trunk
- MinGW bytecode trunk
- MinGW trunk
Linux trunk debug failed with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live #470

Out of 33 workflows 15 failed with 12 genuine issues and 3 ci-setup issues

Aug 27 '24 19:08 jmid

I removed the stress test, as the positive parallel test is already stress testing. I've also rebased based on #471 which should take care of the MinGW+Cygwin version mismatching.

I still expect the 5.2 workflows to trigger ocaml/ocaml#13370 (unsure whether the cmd should be disabled under 5.2 testing :thinking: )

Aug 29 '24 09:08 jmid

CI summary for 9ef90b5:

7 5.2 workflows failed as expected due to the Gc test
- 32bit 5.2 failed the parallel STM Gc test with Allocated_bytes : -7.47499188671e+196
- Bytecode 5.2 aborted with Fatal error: allocation failure during minor GC
- Cygwin 5.2 aborted with Fatal error: allocation failure during minor GC
- FP 5.2 aborted with Fatal error: allocation failure during minor GC
- MinGW 5.2 failed with Fatal error: allocation failure during minor GC
- macOS-intel 5.2 failed with BUS error
- linux-s390x-5.2 - aborted with Fatal error: allocation failure during minor GC
3 debug runtime runs failed with an assertion error:
- Linux 5.2 debug aborted with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live
- Linux 5.3 debug failed with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live
- Linux trunk debug failed with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live

Out of 45 workflows 10 failed, all with genuine issues

Sep 03 '24 19:09 jmid

CI summary for 6b2669b

10 5.2 workflows failed as expected due to the Gc test #474
- 32bit 5.2 failed STM Gc test parallel with a negative allocated_bytes
- Cygwin 5.2 aborted with Fatal error: allocation failure during minor GC
- FP 5.2 aborted with Fatal error: allocation failure during minor GC
- Linux 5.2 debug aborted with Fatal error: allocation failure during minor GC
- MinGW bytecode 5.2 failed with Fatal error: allocation failure during minor GC
- macOS-ARM64 5.2 failed with signal BUS
- macOS-intel 5.2 failed with signal BUS
- linux-arm64-5.2 aborted with Fatal error: allocation failure during minor GC
- linux-s390x-5.2 aborted with Fatal error: allocation failure during minor GC
- freebsd-amd64-5.2 aborted with Fatal error: allocation failure during minor GC
2 debug runtime workflows failed with an assertion error #470
- Linux 5.3 debug failed with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live
- Linux trunk debug failed with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live
linux-ppc64le-5.2 found an unexpected STM Sys test parallel counterexample #466

Out of 45 workflows 13 failed with with 12 genuine issues and 1 false alarm

Sep 03 '24 20:09 jmid

CI summary for e9c9a6a

all 3 Linux debug runs aborted with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live #470
- Linux 5.2 debug
- Linux 5.3 debug
- Linux trunk debug
linux-s390x-5.2 failed STM Gc test parallel with a quick_stat output mismatch

Out of 45 workflows 4 failed with genuine issues

Sep 25 '24 14:09 jmid

CI summary for 6c1252f

9 workflows failed due to minor_heap_size page size rounding
- 32bit 5.2, 5.3, trunk
- Cygwin 5.2, 5.3, trunk
- macOS-ARM64 5.2, trunk
- linux-ppc64le-5.2
all 3 Linux debug runs aborted with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live` #470
- Linux 5.2, 5.3, trunk debug
linux-s390x-5.2 timed out during STM Gc stress test parallel #421
macOS-ARM64 5.3 segfaulted in STM Gc stress test parallel #480

Out of 45 workflows 14 failed with 13 genuine issues and 1 CI infrastructure issue. The macOS-ARM64 5.3 crash is disturbing. fd0a3fd should address the pagesize rounding... :crossed_fingers:

Sep 25 '24 14:09 jmid

CI summary for fd0a3fd:

3 Linux debug workflows failed with runtime/shared_heap.c; line 1392 ### Assertion failed: local->stats.pool_live_words == pool_stats.live #470
- Linux 5.2, 5.3, trunk debug
4 MSVC workflows failed to compile due to a header issue fatal error C1189: #error: "No Target Architecture"
- MSVC 5.3, trunk
- MSVC bytecode 5.3, trunk

Out of 45 workflows, 3 failed with a genuine error and 4 due to a buggy Windows header in the test-suite. 1ebdda4 should address that, leaving only genuine errors.

Note to self: Having removed v=0 in debug mode triggers v=63 and adds a bit of noise to the logs for the Gc test...

Sep 25 '24 16:09 jmid

CI summary for 1ebdda4:

3 Linux debug workflows aborted STM Gc stress test parallel with Assertion failed: local->stats.pool_live_words == pool_stats.live
- Linux 5.2, Linux 5.3, trunk debug
macOS-ARM64 5.2 segfaulted in STM Gc stress test parallel #480

Out of 45 workflows 4 failed - all with genuine issues

Sep 27 '24 12:09 jmid

CI summary for e5b93d3

3 32bit workflows all segfaulted in STM implicit Gc stress test parallel or STM implicit Gc test parallel
- 32bit 5.2, 5.3, trunk
3 Linux debug workflows aborted STM Gc stress test parallel with ### Assertion failed: local->stats.pool_live_words == pool_stats.live
- 5.2, 5.3, trunk debug
- In addition, Linux 5.2 debug also failed STM implicit Gc test parallel with a negative allocated_bytes result
macOS-ARM64 trunk segfaulted in STM Gc stress test parallel

Out of 45 workflows 7 failed all with genuine errors

Sep 30 '24 08:09 jmid

CI summary for 6a86f20

32bit 5.2 segfaulted in STM implicit Gc test parallel
32bit 5.3 segfaulted in STM implicit Gc stress test parallel
32bit trunk segfaulted in STM implicit Gc stress test parallel
Bytecode 5.2 aborted STM implicit Gc test parallel with allocation failure during minor GC
Bytecode 5.3 segfaulted in STM implicit Gc stress test parallel
Bytecode trunk segfaulted in STM implicit Gc test parallel
Linux 5.2 debug
- aborted STM Gc stress test parallel with Assertion failed: local->stats.pool_live_words == pool_stats.live #470 and
- failed STM implicit Gc test parallel with a negative allocated_bytes result
Linux 5.3 debug failed STM Gc stress test parallel with Assertion failed: local->stats.pool_live_words == pool_stats.live #470
Linux trunk debug failed STM Gc stress test parallel with Assertion failed: local->stats.pool_live_words == pool_stats.live #470
linux-s390x-5.2 timedout in Lin Bytes test with Thread #421
linux-ppc64le-5.2 found an unexpected counterexample in STM Sys test parallel #466

Out of 45 workflows, 11 failed with 9 genuine issues, 1 CI timeout, and 1 false alarm

Sep 30 '24 13:09 jmid

I've just pushed to trigger a fresh run, now that #13553 has been merged to trunk and cherry-picked to 5.3.

In other news, I've investigated the previously observed FreeBSD 5.2 crash. This was easy to reproduce, but turned out to be a known one, namely db586e0b27 - Merge pull request #13370 from DemiMarie/fix-missing-gc-rule

Oct 23 '24 12:10 jmid

CI summary for 4a93462:

Linux 5.2 debug failed with an assertion error #470
Linux 5.3 debug failed with an assertion error #470
Linux trunk debug failed with an assertion error #470
macOS-intel 5.2 crashed in STM Gc stress test parallel #480

Out of 41 workflows, 4 failed - all with genuine issues :tada:

(For some reason, the force push didn't trigger multicoretests-ci's four 5.2.0 workflows)

Oct 23 '24 15:10 jmid

CI summary for f7b2fa9

Linux 5.3 debug failed with assertion errors #470
Linux trunk debug failed with assertion errors #470

Out of 31 workflows 2 failed with genuine errors

Jan 06 '25 14:01 jmid

CI summary for 04f66ce

Linux 5.3 debug failed with assertion errors #470
Linux trunk debug failed with assertion errors #470

Out of 31 workflows 2 failed with genuine errors

Jan 06 '25 14:01 jmid

CI summary for 0374c67

Linux 5.3 debug failed with an assertion error #470
Linux trunk debug failed with an assertion error #470
MSVC bytecode 5.3 failed to trigger an STM bigarray parallel counterexample #467
macOS-ARM64 5.3 crashed in stm_tests_par_stress #480

Out of 33 workflows, 4 failed with 3 genuine issues and 1 false alarm.

Jan 06 '25 14:01 jmid

The experiment in 0374c67 replacing Gc.compact in cleanup worked fine - the previously seen issues are still triggering.

Jan 08 '25 13:01 jmid

First, I've rebased this on main.

Secondly, to make progress this PR d7ab558 temporarily omits Gc.compact to avoid triggering the two remaining issues: #470 and #480. If this is CI green, I plan to merge and handle reenabling of Gc.compact in a separate one-line PR.

Jan 08 '25 13:01 jmid

CI summary for d7ab558: all 33 workflows passed.

I'll merge this then :tada:

Jan 08 '25 16:01 jmid

CI summary for merge to main: All 34 workflows passed.

Jan 08 '25 21:01 jmid