False positives? - Re: TestBot job 104977 - results: [PATCH 1/8] d2d1/tests: Test StrokeContains when last segment is bezier and closes the path
On Freitag, 7. Januar 2022 09:50:08 CET you wrote:
VM Status Failures Command build success w7u_2qxl success 0 d2d1_test.exe d2d1 w7u_adm success 0 d2d1_test.exe d2d1 w7u_el success 0 d2d1_test.exe d2d1 w8 success 0 d2d1_test.exe d2d1 w8adm success 0 d2d1_test.exe d2d1 w864 success 0 d2d1_test.exe d2d1 w1064v1507 success 0 d2d1_test.exe d2d1 w1064v1809 success 0 d2d1_test.exe d2d1 w1064 success 0 d2d1_test.exe d2d1 w1064_tsign success 0 d2d1_test.exe d2d1 w10pro64 success 0 d2d1_test.exe d2d1 w864 success 0 d2d1_test64.exe d2d1 w1064v1507 success 0 d2d1_test64.exe d2d1 w1064v1809 success 0 d2d1_test64.exe d2d1 w1064 success 0 d2d1_test64.exe d2d1 w1064_2qxl success 0 d2d1_test64.exe d2d1 w1064_tsign success 0 d2d1_test64.exe d2d1 w10pro64 success 0 d2d1_test64.exe d2d1 w10pro64_ar success 0 d2d1_test64.exe d2d1 w10pro64_he success 0 d2d1_test64.exe d2d1 w10pro64_ja success 0 d2d1_test64.exe d2d1 w10pro64_zh_CN success 0 d2d1_test64.exe d2d1 debian11 failed 63 debian11 failed 16
You can also see the results at: https://testbot.winehq.org/JobDetails.pl?Key=104977
=== debian11 (32 bit report) ===
d2d1: d2d1.c:6664: Test failed: Figure does not match. d2d1.c:6672: Test failed: Figure does not match. d2d1.c:6680: Test failed: Figure does not match. d2d1.c:6688: Test failed: Figure does not match. d2d1.c:6672: Test failed: Figure does not match. d2d1.c:6680: Test failed: Figure does not match. d2d1.c:6688: Test failed: Figure does not match.
As this patch just adds another testcase, it seems somewhat unlikely it introduces new regressions in unrelated tests.
It seems like the my previous patch got applied (which would match the 4 regressions noted), but the patch.diff from the test bot tells something else. Is this possible?
I had tested this against rc4 and had no new test failures. As d2d was unchanged since rc1 this should be sufficient. Visual inspection of the results on my machine (teached the tests to dump result and reference as PBM) also looks good.
Unfortunately, the tests are only run with WINETEST_DEBUG=1, and the relevant output is suppressed in the logs.
Kind regards,
Stefan
On Freitag, 7. Januar 2022 17:07:17 CET you wrote:
On Freitag, 7. Januar 2022 09:50:08 CET you wrote:
[...]
You can also see the results at: https://testbot.winehq.org/JobDetails.pl?Key=104977
=== debian11 (32 bit report) ===
d2d1: d2d1.c:6664: Test failed: Figure does not match. d2d1.c:6672: Test failed: Figure does not match. d2d1.c:6680: Test failed: Figure does not match. d2d1.c:6688: Test failed: Figure does not match. d2d1.c:6672: Test failed: Figure does not match. d2d1.c:6680: Test failed: Figure does not match. d2d1.c:6688: Test failed: Figure does not match.
As this patch just adds another testcase, it seems somewhat unlikely it introduces new regressions in unrelated tests.
It seems like the my previous patch got applied (which would match the 4 regressions noted), but the patch.diff from the test bot tells something else. Is this possible?
I had tested this against rc4 and had no new test failures. As d2d was unchanged since rc1 this should be sufficient. Visual inspection of the results on my machine (teached the tests to dump result and reference as PBM) also looks good.
Unfortunately, the tests are only run with WINETEST_DEBUG=1, and the relevant output is suppressed in the logs.
Kind regards,
Stefan
I have rerun the test with WINETEST_DEBUG=2, see https://testbot.winehq.org/JobDetails.pl?Key=105001
This seems to be an issue on the Debian 11 CI system. It is a small rendering glitch, the (first)test is supposed to paint a plain rectangle, but on Debian both horizontal edges are thinned by one pixel (see attached PNGs or ASCII art below). All 4 joins are rendered correctly.
ASCII art - Win/TW || Debian:
*====* *____* | | | | | | | | *====* *____*
On my openSUSE Tumbleweed, the output matches the reference, so it may be a graphics driver issue.
Kind regards,
Stefan
On Tue, 11 Jan 2022 at 08:55, Stefan Brüns stefan.bruens@rwth-aachen.de wrote:
I have rerun the test with WINETEST_DEBUG=2, see https://testbot.winehq.org/JobDetails.pl?Key=105001
This seems to be an issue on the Debian 11 CI system. It is a small rendering glitch, the (first)test is supposed to paint a plain rectangle, but on Debian both horizontal edges are thinned by one pixel (see attached PNGs or ASCII art below). All 4 joins are rendered correctly.
ASCII art - Win/TW || Debian:
*====* *____* | | | | | | | | *====* *____*
On my openSUSE Tumbleweed, the output matches the reference, so it may be a graphics driver issue.
IIRC that particular machine is using llvmpipe, which unfortunately cares a bit more about speed than correctness. Sadly, I don't think a solution is likely to happen; sorry you ran into this.
-
Henri Verbeet -
Stefan Brüns