Hi Stephan,
On 22/02/2023 10:54, Stephan Bergmann wrote:
<https://ci.libreoffice.org//job/lo_ubsan/2691> failed
CppunitTest_basegfx with
...
in testTrapezoidHelper that had originally been introduced in
<https://git.libreoffice.org/core/+/82b56544a7a53528970632d086c3cfd8ef879335%5E!/>
"basegfx: accelerate Trapezoid subdivision by avoiding allocations" and
apparently uses random data (via basegfx2d::getRandomOrdinal, defined in
basegfx/test/boxclipper.cxx), so that runs of CppunitTest_basegfx are
not reproducible (so that there's no good way to reproduce and fix the
above division by zero error). Can that test be made reproducible please?
Ooh - interesting; so would using the boxclipper's getRandomOrdinal -
which uses rand() and (I assume) could be consistently seeded ?
Or could we consistently seed:
comphelper::rng::uniform_size_distribution(
To produced a known pattern ? or we could hard-code a set of numbers
(which I guess) wouldn't provoke the problem into the test.
Would be lovely to find the sequence that kills it of course,
Thoughts much appreciated,
Michael.
--
michael.meeks@xxxxxxxxxxxxx <><, GM Collabora Productivity
Hangout: mejmeeks@xxxxxxxxx, Skype: mmeeks
(M) +44 7795 666 147 - timezone usually UK / Europe
