On 26/04/2022 21:14, Caolán McNamara wrote:
On Tue, 2022-04-26 at 17:46 +0200, Stephan Bergmann wrote:
and I have some findings in the comment at
<
https://gerrit.libreoffice.org/c/core/+/130493/7#message-e08f298e02448
997aec8c89adf128cfe5e0efae5>
"add a layout test to exercise UXTSOREL".
lets see if https://gerrit.libreoffice.org/c/core/+/133460 works to
resolve this
To conclude: My hypothesis is that the test would always have run to
completion, but just depending on which font happened to get substituted
for the requested Palatino Linotype on a given machine, that could have
taken obscenely long, esp. in --enable-dbgutil builds.
That would explain why some "Gerrit Linux clang/dbgutil" and "UBSAN
Linux Build" builds succeeded and others failed, even on the same
machine (as some happened to start the given test early enough so that
its long but finite run time overlapped with enough other build output
so as not to let the timeout killer kick in), and why this seemingly
never was an issue with non-debug "Gerrit Linux gcc/release" builds
(where the given test would always have succeeded quickly enough).
The take-away probably is that we should make sure to only use fonts in
our `make check` tests that are covered by --with-fonts.
