On 02/09/2026 17.37, Noah Meyerhans via dovecot wrote:
On Tue, Sep 01, 2026 at 05:57:21PM -0400, Noah Meyerhans via dovecot wrote:
On loong64, the test suite fails in test-lib with:
ostream multiplex stream (backpressure): sub-process ended properly .. : FAILED Warning: test: Sub-process forcibly terminated with signal 9 ostream multiplex stream (backpressure) .............................. : FAILED
I'm not quite sure what's going on yet, but the child process in this test is hanging in the write() call in o_stream_file_writev().
I've noticed that if I increase the output buffer size from 4096 to 8192 in test_stream_backpressure_child() then the test passes. See below for the diff. If the buffer was too large and the entire input was consumed, that would still result in a test failure, so I believe this change still exercises the code paths as expected. I'm not entirely sure why this is happening only on loong64, though.
Hi, we've hit the same issues on s390x and ppc64le archs.
I've checked the PR and it solved the issue for s390x.
As for the ppc64le issue it seems that the test expects to hit pipe buffer full situation, so write stops. The problem seems to be that pipe buffer on linux ppc64le systems is much bigger (default 1 MB) so it never hits buffer full and finishes writing instead. I've used patch bellow that sets pipe buffer size to 64k (same as x86_64, s390x...)
Cheers, Michal Hlavinka