53427 2011-01-31 08:48:46 -0800 [Qt][WK2] Canvas tests make css2.1 ones failing 2012-02-03 08:28:56 -0800 1 1 1 Unclassified WebKit WebKit2 528+ (Nightly build) PC All RESOLVED FIXED LayoutTestFailure, Qt, QtTriaged P3 Normal --- 1 kbalazs zherczeg cshu jesus laszlo.gombos ossy pierre.rossi zecke oldest_to_newest 342594 0 kbalazs 2011-01-31 08:48:46 -0800 Pretty annoying bug. If one runs canvas tests than all of the css2.1 tests failing with WTR. Without the canvas tests there is only 3 failure in css2.1. The whole story happening only on the bot's 32 bit Debian environment. On the same machine in a 64 bit chroot there is no such problem. 343147 1 kbalazs 2011-02-01 04:44:24 -0800 I have reproduced the problem on a different 32 bit Linux environment. Since the css2.1 tests are testing more basic and so important parts, I think it would be better to skip the canvas tests for now. I am going to upload the patch. 343150 2 kbalazs 2011-02-01 05:00:20 -0800 After a look of the number of the css2.1 and the canvas tests, and some discussion with Ossy I have rejected the idea of skipping the canvas tests. 343151 3 kbalazs 2011-02-01 05:03:50 -0800 Some snippet from the output of run-webkit-tests -2 --verbose running canvas/philip/tests/type.extend.html -> succeeded running canvas/philip/tests/type.name.html -> succeeded running canvas/philip/tests/type.replace.html -> succeeded running css1/box_properties/float_on_text_elements.html -> failed running css1/classification/white_space.html -> failed running css1/units/zero-duration-without-units.html -> succeeded running css2.1/t09-c5526c-display-00-e.html -> failed running css2.1/t0402-c71-fwd-parsing-00-f.html -> failed running css2.1/t0402-c71-fwd-parsing-01-f.html -> failed running css2.1/t0402-c71-fwd-parsing-02-f.html -> failed running css2.1/t0402-c71-fwd-parsing-03-f.html -> failed running css2.1/t0402-c71-fwd-parsing-04-f.html -> failed ... (lot of failing css2.1 tests) running css2.1/t1204-root-e.html -> failed running css2.1/t1205-c561-list-displ-00-b.html -> failed | Closing DumpTool | | Opening DumpTool | running css2.1/t1205-c563-list-type-00-b.html -> succeeded running css2.1/t1205-c563-list-type-01-b.html -> succeeded running css2.1/t1205-c564-list-img-00-b-g.html -> succeeded running css2.1/t1205-c565-list-pos-00-b.html -> succeeded ... (almost all the css2.1 tests runs successfully from here) 411379 4 zherczeg 2011-05-27 06:14:53 -0700 QApplication::fontMetrics().averageCharWidth() doubles during the running of the test. Something affects this member. Do you have any idea? 411381 5 zherczeg 2011-05-27 06:24:35 -0700 (Besides QApplication::setFont() which is not happen) 412193 6 zherczeg 2011-05-30 05:36:25 -0700 Big progress! The increase happen during: WebCore/bindings/js/GCController.cpp: gcTimerFired() Don't ask how I found it ;) I suspect a Qt specific object do bad things during its destructor. 412616 7 zherczeg 2011-05-31 07:42:34 -0700 Another big step: Source/WebCore/html/HTMLCanvasElement.cpp HTMLCanvasElement::~HTMLCanvasElement() m_context.clear(); Can cause this. Tee conditions are not yet known. 414033 8 zherczeg 2011-06-02 04:54:30 -0700 FontCustomPlatformDataQt.cpp QFontDatabase::removeApplicationFont(m_handle); - is actually successful in case of WebKit2's DRT, but fails with normal DRT. I think the normal DRT has an extra reference to the font, which keeps it alive (so the fail hides the issue). 414034 9 zherczeg 2011-06-02 04:57:09 -0700 Zecke, you made this code before. Maybe you have some idea. 414070 10 zherczeg 2011-06-02 06:32:42 -0700 If I comment out the mentioned line, the tests are passed. FontCustomPlatformData::~FontCustomPlatformData() { - QFontDatabase::removeApplicationFont(m_handle); +// QFontDatabase::removeApplicationFont(m_handle); } 417170 11 zherczeg 2011-06-08 03:59:35 -0700 Now I know whz the tests are passed in DRT: DumpRenderTree::open() calls QFontDatabase::removeAllApplicationFonts() this call hides this particular bug. Since the Qt port moves to QRawFont, QFontDatabase will be dropped from the source, it is enough to mimic the DRT behaviour in WKTR for now, even if it is wrong. I have no better idea at the moment. 417870 12 96555 zherczeg 2011-06-08 23:58:20 -0700 Created attachment 96555 patch Ugly, but this is how DRT works, and just hides the bug. We need to refactor and fix the platform font management code in the future. 417892 13 96555 ossy 2011-06-09 01:37:41 -0700 Comment on attachment 96555 patch View in context: https://bugs.webkit.org/attachment.cgi?id=96555&action=review You're right, it is a little ugly, but necessary workaround to make WK2 tests work. r=me to go ahead with some typo fix. Is there any bug report for moving to QRawFont? > LayoutTests/ChangeLog:9 > + * platform/qt-wk2/Skipped: > + + Unskip canvas tests. > Tools/ChangeLog:8 > + Adding a workaround for this issue by mimicking the behaviour of DumpRenderTree. s/mimicking/mimicing 417899 14 zherczeg 2011-06-09 01:49:13 -0700 Landed in http://trac.webkit.org/changeset/88435 500707 15 ossy 2011-11-11 05:31:57 -0800 Reopen, because this bug is appeared again after migrating from Qt 4.8 to Qt 5.0, because workaround doesn't work anymore. 503959 16 kbalazs 2011-11-16 09:15:25 -0800 (In reply to comment #15) > Reopen, because this bug is appeared again after migrating from Qt 4.8 to Qt 5.0, because workaround doesn't work anymore. Actually we still have FontDatabase::removeAllApplicationFonts - maybe the Qt5 implementation is not sufficient, need to check - but it came out that we have a much more basic issue which is that we don't use the test fonts with Qt5. Tracked in https://bugs.webkit.org/show_bug.cgi?id=72513. 548692 17 jesus 2012-02-03 07:16:04 -0800 Is this related to the work Pierre has being doing with raw fonts?! 548742 18 kbalazs 2012-02-03 08:28:56 -0800 (In reply to comment #17) > Is this related to the work Pierre has being doing with raw fonts?! Not really. This bug has been solved long time ago thanks to Zoltan. The new failures was not a reappearing of that but another bug that has also been fixed. 96555 2011-06-08 23:58:20 -0700 2011-11-11 05:23:05 -0800 patch 0001-font-bug.patch text/plain 3825 zherczeg RnJvbSBiNmMxMjE3ZDgwMTkzYmQzY2JhNTliN2U4YjRiMGI3MzlkMjVhN2NjIE1vbiBTZXAgMTcg MDA6MDA6MDAgMjAwMQpGcm9tOiBab2x0YW4gSGVyY3plZyA8emhlcmN6ZWdAaW5mLnUtc3plZ2Vk Lmh1PgpEYXRlOiBXZWQsIDggSnVuIDIwMTEgMjM6NTE6MzIgLTA3MDAKU3ViamVjdDogW1BBVENI XSBmb250LWJ1ZwoKLS0tCiBMYXlvdXRUZXN0cy9DaGFuZ2VMb2cgICAgICAgICAgICAgICAgICAg ICAgICAgICAgICB8ICAgIDkgKysrKysrKysrCiBMYXlvdXRUZXN0cy9wbGF0Zm9ybS9xdC13azIv U2tpcHBlZCAgICAgICAgICAgICAgICB8ICAgIDMgLS0tCiBUb29scy9DaGFuZ2VMb2cgICAgICAg ICAgICAgICAgICAgICAgICAgICAgICAgICAgICB8ICAgMTUgKysrKysrKysrKysrKysrCiAuLi4v SW5qZWN0ZWRCdW5kbGUvcXQvTGF5b3V0VGVzdENvbnRyb2xsZXJRdC5jcHAgICB8ICAgMTIgKysr KysrKysrKysrCiA0IGZpbGVzIGNoYW5nZWQsIDM2IGluc2VydGlvbnMoKyksIDMgZGVsZXRpb25z KC0pCgpkaWZmIC0tZ2l0IGEvTGF5b3V0VGVzdHMvQ2hhbmdlTG9nIGIvTGF5b3V0VGVzdHMvQ2hh bmdlTG9nCmluZGV4IDE1ZmI5OTEuLjJmNTYzNWMgMTAwNjQ0Ci0tLSBhL0xheW91dFRlc3RzL0No YW5nZUxvZworKysgYi9MYXlvdXRUZXN0cy9DaGFuZ2VMb2cKQEAgLTEsMyArMSwxMiBAQAorMjAx MS0wNi0wOCAgWm9sdGFuIEhlcmN6ZWcgIDx6aGVyY3plZ0BpbmYudS1zemVnZWQuaHU+CisKKyAg ICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChPT1BTISkuCisKKyAgICAgICAgW1F0XVtXSzJdIENh bnZhcyB0ZXN0cyBtYWtlIGNzczIuMSBvbmVzIGZhaWxpbmcKKyAgICAgICAgaHR0cHM6Ly9idWdz LndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTUzNDI3CisKKyAgICAgICAgKiBwbGF0Zm9ybS9x dC13azIvU2tpcHBlZDoKKwogMjAxMS0wNi0wNyAgWm9sdGFuIEhlcmN6ZWcgIDx6aGVyY3plZ0Bp bmYudS1zemVnZWQuaHU+CiAKICAgICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChPT1BTISkuCmRp ZmYgLS1naXQgYS9MYXlvdXRUZXN0cy9wbGF0Zm9ybS9xdC13azIvU2tpcHBlZCBiL0xheW91dFRl c3RzL3BsYXRmb3JtL3F0LXdrMi9Ta2lwcGVkCmluZGV4IDZiOTBhYzkuLjQwYWI1ZDUgMTAwNjQ0 Ci0tLSBhL0xheW91dFRlc3RzL3BsYXRmb3JtL3F0LXdrMi9Ta2lwcGVkCisrKyBiL0xheW91dFRl c3RzL3BsYXRmb3JtL3F0LXdrMi9Ta2lwcGVkCkBAIC0yMjc0LDkgKzIyNzQsNiBAQCBodHRwL3Rl c3RzL2hpc3RvcnkvcG9wc3RhdGUtZmlyZXMtd2l0aC1wZW5kaW5nLXJlcXVlc3RzLmh0bWwKICMg cmVxdWlyZXMgdmlkZW8uYnVmZmVyZWQgdG8gYmUgYWJsZSB0byByZXR1cm4gbXVsdGlwbGUgdGlt ZXJhbmdlcwogaHR0cC90ZXN0cy9tZWRpYS92aWRlby1idWZmZXJlZC5odG1sCiAKLSMgY2FudmFz IHRlc3RzIHBhc3MsIGJ1dCBicmVhayB+MTAwIGNzczIuMSB0ZXN0cywgaXQgbWlnaHQgYmUgYSBi dWcgaW4gV2ViS2l0VGVzdFJ1bm5lcgotY2FudmFzCi0KICMgVGhlc2UgdGVzdHMgYWx3YXlzIGZh aWwgKHdpdGggYW5kIHdpdGhvdXQgLS1zaW5nbHkpLiBJdCBuZWVkcyBpbnZlc3RpZ2F0aW9uLgog YW5pbWF0aW9ucy9hbmltYXRpb24tY29udHJvbGxlci1kcnQtYXBpLmh0bWwKIGRvbS94aHRtbC9s ZXZlbDIvaHRtbC9IVE1MRG9jdW1lbnQxNy54aHRtbApkaWZmIC0tZ2l0IGEvVG9vbHMvQ2hhbmdl TG9nIGIvVG9vbHMvQ2hhbmdlTG9nCmluZGV4IDZjM2NmNWIuLmEwYzhmYjEgMTAwNjQ0Ci0tLSBh L1Rvb2xzL0NoYW5nZUxvZworKysgYi9Ub29scy9DaGFuZ2VMb2cKQEAgLTEsMyArMSwxOCBAQAor MjAxMS0wNi0wOCAgWm9sdGFuIEhlcmN6ZWcgIDx6aGVyY3plZ0BpbmYudS1zemVnZWQuaHU+CisK KyAgICAgICAgUmV2aWV3ZWQgYnkgTk9CT0RZIChPT1BTISkuCisKKyAgICAgICAgW1F0XVtXSzJd IENhbnZhcyB0ZXN0cyBtYWtlIGNzczIuMSBvbmVzIGZhaWxpbmcKKyAgICAgICAgaHR0cHM6Ly9i dWdzLndlYmtpdC5vcmcvc2hvd19idWcuY2dpP2lkPTUzNDI3CisKKyAgICAgICAgQWRkaW5nIGEg d29ya2Fyb3VuZCBmb3IgdGhpcyBpc3N1ZSBieSBtaW1pY2tpbmcgdGhlIGJlaGF2aW91ciBvZiBE dW1wUmVuZGVyVHJlZS4KKyAgICAgICAgU2luY2UgdGhlIFF0IHBvcnQgc2hvdWxkIGV2ZW50dWFs bHkgZ28gdG8gUVJhd0ZvbnQsIHRoaXMgaXNzdWUgd2lsbCBiZSByZXNvbHZlZAorICAgICAgICBp biB0aGUgZnV0dXJlLiBCb3RoIGNsZWFyTWVtb3J5Q2FjaGVzKCkgYW5kIHJlbW92ZUFsbEFwcGxp Y2F0aW9uRm9udHMoKSBhcmUKKyAgICAgICAgbmVlZGVkIGZvciB0aGUgd29ya2Fyb3VuZC4KKwor ICAgICAgICAqIFdlYktpdFRlc3RSdW5uZXIvSW5qZWN0ZWRCdW5kbGUvcXQvTGF5b3V0VGVzdENv bnRyb2xsZXJRdC5jcHA6CisgICAgICAgIChXVFI6OkxheW91dFRlc3RDb250cm9sbGVyOjpwbGF0 Zm9ybUluaXRpYWxpemUpOgorCiAyMDExLTA2LTAyICBEaXJrIFByYW5rZSAgPGRwcmFua2VAY2hy b21pdW0ub3JnPgogCiAgICAgICAgIFJldmlld2VkIGJ5IFRvbnkgQ2hhbmcuCmRpZmYgLS1naXQg YS9Ub29scy9XZWJLaXRUZXN0UnVubmVyL0luamVjdGVkQnVuZGxlL3F0L0xheW91dFRlc3RDb250 cm9sbGVyUXQuY3BwIGIvVG9vbHMvV2ViS2l0VGVzdFJ1bm5lci9JbmplY3RlZEJ1bmRsZS9xdC9M YXlvdXRUZXN0Q29udHJvbGxlclF0LmNwcAppbmRleCA3MjkwMmUxLi43M2MyOTVlIDEwMDY0NAot LS0gYS9Ub29scy9XZWJLaXRUZXN0UnVubmVyL0luamVjdGVkQnVuZGxlL3F0L0xheW91dFRlc3RD b250cm9sbGVyUXQuY3BwCisrKyBiL1Rvb2xzL1dlYktpdFRlc3RSdW5uZXIvSW5qZWN0ZWRCdW5k bGUvcXQvTGF5b3V0VGVzdENvbnRyb2xsZXJRdC5jcHAKQEAgLTI3LDkgKzI3LDEyIEBACiAjaW5j bHVkZSAiY29uZmlnLmgiCiAjaW5jbHVkZSAiTGF5b3V0VGVzdENvbnRyb2xsZXIuaCIKIAorI2lu Y2x1ZGUgIkFjdGl2YXRlRm9udHMuaCIKICNpbmNsdWRlICJJbmplY3RlZEJ1bmRsZS5oIgogI2lu Y2x1ZGUgPFFEaXI+CisjaW5jbHVkZSA8UUZvbnREYXRhYmFzZT4KICNpbmNsdWRlIDxRT2JqZWN0 PgorI2luY2x1ZGUgPHF3ZWJzZXR0aW5ncy5oPgogCiBuYW1lc3BhY2UgV1RSIHsKIApAQCAtNTUs NiArNTgsMTUgQEAgcHJpdmF0ZToKIAogdm9pZCBMYXlvdXRUZXN0Q29udHJvbGxlcjo6cGxhdGZv cm1Jbml0aWFsaXplKCkKIHsKKyAgICAvLyBNYWtlIFdlYktpdDIgbWltaWMgdGhlIGJlaGF2aW91 ciBvZiBEdW1wUmVuZGVyVHJlZSwgd2hpY2ggaXMgaW5jb3JyZWN0LAorICAgIC8vIGJ1dCB0ZXN0 cyBhcmUgc3VjY2Vzc2Z1bGx5IHBhc3NlZC4gT24gdGhlIGxvbmcgcnVuLCBRdCB3aWxsIG1vdmUg dG8gUVJhd0ZvbnQsCisgICAgLy8gd2hpY2ggbWFrZXMgdGhlIHVzZSBvZiBRRm9udERhdGFiYXNl IHVubmVjZXNzYXJ5LgorICAgIC8vIFNlZSBodHRwczovL2J1Z3Mud2Via2l0Lm9yZy9zaG93X2J1 Zy5jZ2k/aWQ9NTM0MjcKKyAgICBRV2ViU2V0dGluZ3M6OmNsZWFyTWVtb3J5Q2FjaGVzKCk7Cisj aWYgIShkZWZpbmVkKFFfT1NfU1lNQklBTikgJiYgUVRfVkVSU0lPTiA8PSBRVF9WRVJTSU9OX0NI RUNLKDQsIDYsIDIpKQorICAgIFFGb250RGF0YWJhc2U6OnJlbW92ZUFsbEFwcGxpY2F0aW9uRm9u dHMoKTsKKyNlbmRpZgorICAgIGFjdGl2YXRlRm9udHMoKTsKICAgICBRT2JqZWN0Ojpjb25uZWN0 KCZtX3dhaXRUb0R1bXBXYXRjaGRvZ1RpbWVyLCBTSUdOQUwodGltZW91dCgpKSwgV2F0Y2hkb2dU aW1lckhlbHBlcjo6aW5zdGFuY2UoKSwgU0xPVCh0aW1lckZpcmVkKCkpKTsKIH0KIAotLSAKMS43 LjIuMwoK