GNOME Bugzilla – Bug 772991
gegl 0.3.8 test "test-image-compare" seen failing
Last modified: 2017-05-11 17:52:58 UTC
Hi! While it works on my machine, we had a report [1] that test "test-image-compare" was seen failing for someone else: [..] ./test-image-compare The max_diff property differs: 1.832876 instead of 1.832906. -> Comparison failed for the green channel case FAIL test-image-compare [..] The full log is attached at [1]. Best Sebastian [1] https://bugs.gentoo.org/show_bug.cgi?id=595332
It would be nice to know if such failures are currently being observed with git master. The addition of denser babl-fast path coverage might have made this issue disappear.
Hi! For Git master (8d725c049479f55dffa41e45be42d2ce45516738) .. * (I run into bug #780226 with "make all") * "FAIL gegl.xml" for "make check" * "FAIL test-serialize" for "make -C tests/simple/ check" test-image-compare has always passed for me (and still does). It failed for other people. I can ask for a re-test from others at https://bugs.gentoo.org/show_bug.cgi?id=595332 if you want. is this what we want them to run?: git clone https://git.gnome.org/browse/gegl cd gegl/ ./autogen.sh && ./configure && make all make check ; make -C tests/simple/ check Directly related: Is "make check" supposed to not run all tests anymore? Did I overlook an option to make it do that? I want to run them all :) Best, Sebastian
summary when something in a subdir has a failure.(In reply to Sebastian Pipping from comment #2) > git clone https://git.gnome.org/browse/gegl > cd gegl/ > ./autogen.sh && ./configure && make all > make check ; make -C tests/simple/ check Yep - This would be sufficient. > Directly related: Is "make check" supposed to not run all tests anymore? > Did I overlook an option to make it do that? I want to run them all :) The only thing that should've changed wrt which tests run on make check.. is the order the Makefile does subdirectories in, building aborts after the first subdir with a failure - when printing the subdirs summary. I agree that running all tests is desirable if multiple tests are failing.
(In reply to Øyvind Kolås from comment #3) > Yep - This would be sufficient. Done: https://bugs.gentoo.org/show_bug.cgi?id=595332#c7 Thanks for the explanation on subdirs.
The GIMP jenkins builds of GEGL also sometimes exhibit this behavior now, decreasing the threshold by an order of magnitude and closing the bug, if the bug-reappears, increase - or figure out which code path causes which numbers. commit 113a233ffd8e985247e1fe4bc3fdc22dd459301b Author: Øyvind Kolås <pippin@gimp.org> Date: Thu May 11 19:49:26 2017 +0200 tests: make test-image-compare less hairtriggered Something, probably under our control is causing this test to sometimes fail, possibly related to a performance race about different level of good enough babl-fast paths, hopefully fixes bug #772991