On 29. 5. 2026 00:17, Jun Omae wrote:
Hi,
On 2026/05/29 4:37, Branko Čibej wrote:
This intermittent failure shows up sometimes on both Linux and macOS:
FAIL: ra-test: Unknown test failure (-11); see tests.log.
That's a crash, 11 is SIGSEGV on both OSes. Sometimes the same happens in
repos-test, too, and it doesn't matter if its local or DAV or svnserve. I
suspect that we have a wonderful little bug in our code. I've been trying, on
and off, to trace this down either with a debugger or with clang's address
sanitiser or with APR's pool debugging enabled, but it's an elusive little
bxtrd and I've not yet been able to track it down. Not even so far as to figure
out whether it's hiding in FSFS or somewhere else.
It's been a few years and I think I remember seeing it on the 1.14 branch, too.
On the other hand, I don't recall any bug reports that could be linked to this
observation. We're seeing this failure in the CI tests, too, at least with
autotools. I haven't seen this with CMake, but failures in those workflows are
far more often related to vcpkg or other environmental stuff, so it's a bit
hard to find.
If anyone has any idea where to look without diving into a line-by-line review
of the code, please share your thoughts. This is starting to get on my nerves,
just a little bit.
The issue occurs in serf_default_destroy_and_data() via
apr_pool_cleanup_for_exec() from the child process after apr_proc_create() to
open a tunnel with multi-threaded. I assume that libsvn_ra_serf and/or serf
have something wrong. Also, I guess that the same issue might occur when a hook
is invoked with SVNMasterURI enabled on Apache mpm event or worker.
It happens without DAV, too, and with prefork mpm – the only one that
even works on macOS, so I always use that for testing.
[...]
-- Brane
