1592046 - Debian 10 - writing-mode/tables/table-caption-top-1.html, table-caption-bottom-1.html | image comparison, max difference: 64, number of differing pixels: 61

Status: RESOLVED INVALID

(Core :: Layout: Block and Inline, defect, P5)

Description

[Treeherder Bug Filer]

Filed by: egao [at] mozilla.com
Parsed log: https://treeherder.mozilla.org/logviewer.html#?job_id=271937755&repo=try
Full log: https://queue.taskcluster.net/v1/task/aY1k6kDNT9ShzHqXSqG_YQ/runs/0/artifacts/public/logs/live_backing.log
Reftest URL: https://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=https://queue.taskcluster.net/v1/task/aY1k6kDNT9ShzHqXSqG_YQ/runs/0/artifacts/public/logs/live_backing.log&only_show_unexpected=1

---

suite: reftest-5

context:

two tests fail in the manifest file layout/reftests/writing-mode/tables/reftest.list with same pixel difference.

[task 2019-10-18T17:24:17.057Z] 17:24:17 INFO - REFTEST TEST-START | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-top-1.html == file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-top-1-ref.html
[task 2019-10-18T17:24:17.057Z] 17:24:17 INFO - REFTEST TEST-LOAD | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-top-1.html | 72 / 91 (79%)
[task 2019-10-18T17:24:17.096Z] 17:24:17 INFO - REFTEST TEST-LOAD | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-top-1-ref.html | 72 / 91 (79%)
[task 2019-10-18T17:24:17.275Z] 17:24:17 INFO - REFTEST TEST-UNEXPECTED-FAIL | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-top-1.html == file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-top-1-ref.html | image comparison, max difference: 64, number of differing pixels: 61

[task 2019-10-18T17:24:17.296Z] 17:24:17 INFO - REFTEST TEST-START | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-bottom-1.html == file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-bottom-1-ref.html
[task 2019-10-18T17:24:17.297Z] 17:24:17 INFO - REFTEST TEST-LOAD | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-bottom-1.html | 73 / 91 (80%)
[task 2019-10-18T17:24:17.302Z] 17:24:17 INFO - REFTEST TEST-LOAD | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-bottom-1-ref.html | 73 / 91 (80%)
[task 2019-10-18T17:24:17.460Z] 17:24:17 INFO - REFTEST TEST-UNEXPECTED-FAIL | file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-bottom-1.html == file:///builds/worker/workspace/build/tests/reftest/tests/layout/reftests/writing-mode/tables/table-caption-bottom-1-ref.html | image comparison, max difference: 64, number of differing pixels: 61

Comment 1

[Edwin Takahashi (:egao | infrequent contributor)]

:jkew - I'm observing deviation from reference image on debian10 test image for these two tests. To me this looks like a font rendering difference. I wonder if installing the Ubuntu Light font would address this issue?

Comment 2

[Jonathan Kew [:jfkthame]]

It looks like pixel-snapping of glyph positions is working out slightly differently in left-to-right vs right-to-left directions.

It's possible it would work out differently with another font, though there's no particular reason to expect that; it's equally likely similar issues would arise, perhaps on different glyphs.

The discrepancy would probably be much less, perhaps nothing, if subpixel glyph positioning were enabled; I think that now depends on platform hinting settings. If the system is configured with "light" (rather than "full") hinting, you'd get subpixel positioning and so the pixel-snapping that's happening (differently) here would not occur.

But it'd also be OK to just mark this as fuzzy, I think. The test is not actually failing what it aims to test, and there'll always be a risk of tiny discrepancies as a result of floating-point errors accumulating differently when we compare LTR vs RTL versions of what should be "the same" text.

Comment 3

[Edwin Takahashi (:egao | infrequent contributor)]

Closing this bug due to redirection of efforts from debian10 to ubunu1804 image, and this bug is not reproducible on the ubuntu1804 image.