std.experimental.allocator.building_blocks.allocator_list: Flaky unittests since v2.111

[kinke]

I've seen this module's unittests newly fail sporadically, i.e., not consistently. At least twice:

• DMD FreeBSD CI (https://github.com/dlang/dmd/actions/runs/14202895666/job/39793758312?pr=21133#step:3:7803):

  ****** FAIL debug64 std.experimental.allocator.building_blocks.allocator_list
  core.exception.AssertError@std/experimental/allocator/building_blocks/allocator_list.d(500): Assertion failure
  

• LDC macOS x86_64 CI (https://github.com/ldc-developers/ldc/actions/runs/14049565198/job/39337280925?pr=4877#step:12:3605, https://github.com/ldc-developers/ldc/actions/runs/14419064983/job/40439352050)

  The following tests FAILED:
  	732 - std.experimental.allocator.building_blocks.allocator_list-debug (Timeout)
  

  (with a generous 2-minutes timeout)

I guess introduced in #10594.

[kinke]

I've seen it hang on macOS arm64 too, but don't have the link to the log anymore. So far, I think I've really only seen it hang on FreeBSD and macOS, not on Linux or Windows.

[LightBender]

Thankfully, these are experimental so we can break them if we need to. But this is yet another in the plus column for the "Delete Allocators in V3" argument.

[rainers]

It seems the allocator_list has issues when running multi-threaded. An assert is triggered, but not shown because the program freezes due to a lock that is never released by unwinding as that is not generated for nothrow functions. See also #10876 (comment)

[rikkimax]

Removing the eliding of exception handling code for nothrow blocks is approved. I haven't got to it yet.

…

On Fri, 3 Oct 2025, 20:52 Rainer Schuetze, ***@***.***> wrote: *rainers* left a comment (dlang/phobos#10730) <#10730 (comment)> It seems the allocator_list has issues when running multi-threaded. An assert is triggered, but not shown because the program freezes due to a lock that is never released by unwinding as that is not generated for nothrow functions. See also #10876 (comment) <#10876 (comment)> — Reply to this email directly, view it on GitHub <#10730 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAHSL427EYEX5CQ7ZSUEAW33VYTNNAVCNFSM6AAAAAB2HYBLGOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNRUGY2TQOJYHE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>