Bug#904018: gcc-8: FTBFS on x32: hangs in the testsuite (zombie child)
On Sat, 13 Apr 2019 at 08:27, Laurence Parry <greenreaper@gmail.com> wrote:
>
> I think the problem may be that code to use libcurl is compiled into libphobos (std.net.curl) and tests for it are executed which do not correctly handle its absence.
>
> Neither libgphobos-8-dev nor libgphobos76 appear to depend on, recommend or suggest libcurl4 and it does not appear in the build logs; 'curl' is suggested but not installed.
>
> In the absence of correct libcurl libraries, these tests were known to time out:
> https://issues.dlang.org/show_bug.cgi?id=18519
> In doing so there might be a deadlock.
>
> The issue was raised initially on FreeBSD and Windows, but Iain Buclaw
> says there that it occurs on "Linux64 with MODEL=32"
> (although this may be a lack of regular x86 32-bit libraries on a amd64 system).
>
> This was meant to be fixed by a patch in March 2018, however it appears to have been ineffective:
> https://github.com/dlang/phobos/commit/35e9807bbec392658ecdc8b4ed93d1f6c02f2845
>
> Iain committed further patches to resolve this in January (now in dmd v2.085.1):
> https://github.com/dlang/phobos/commit/1dcb92aade1c82c356e964c1ca15c51e7de8bf76
> https://github.com/dlang/phobos/commit/d36b28c3fc90919a2a3a98249ad2d65393e28507
> I'm not sure if this would go automatically into a 8.4, or maybe it can be cherry-picked?
>
> I've no knowledge of D and other patches here (or elsewhere?) may also be relevant:
> https://github.com/dlang/phobos/commit/c6bd6743254584e292f82cbcbac644d1705d14b0
> https://github.com/dlang/phobos/commit/8d51519f3f11a5f9b538cd690521b61e9fbbf5df
> https://github.com/dlang/phobos/commits/master/std/net/curl.d in general
>
> Just noticed this failure popping up regularly in the x32 build logs and thought I'd have a look, since I'm hoping to build a home server on it shortly.
>
I don' think the hang should be happening anymore on trunk (only
tested -m32 recently).
I will be shortly be adding a
`check_effective_target_libcurl_available` guard in the testsuite to
SKIP the test rather than FAIL, or worse, hang.
See also:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88654
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89254
--
Iain
Reply to: