1752639 - WPTs block-002-wm-vlr-print.html and block-002-wm-vrl-print.html fail

Status: NEW

(Core :: Printing: Output, defect, P3)

Description

[Daniel Holbert [:dholbert]]

These two WPT tests fail with a slight mismatch on the 3rd page:

https://searchfox.org/mozilla-central/source/testing/web-platform/tests/css/css-break/block-002-wm-vlr-print.html
https://searchfox.org/mozilla-central/source/testing/web-platform/tests/css/css-break/block-002-wm-vrl-print.html

This was papered over by the fact that we didn't print any output on the 3rd page at all, due to bug 1740304. But with the fix for bug 1740304, we now render the actual content on the 3rd page (via the fragmentation-fallback) and we can detect that it doesn't quite match.

Try run showing the failures:
https://treeherder.mozilla.org/jobs?repo=try&revision=1492176bdc04ea3731affd1d9b7fc672b4976473&searchStr=tests-print-reftest

Comment 1

[Daniel Holbert [:dholbert]]

The failures say this for the "vlr" test, on Windows and Linux:

Found 388 pixels different, maximum difference per channel 74 on page 3

And for vrl:

Found 302 pixels different, maximum difference per channel 74 on page 3

On macOS, only the first test ("vlr") fails, with 383/74 px different. vrl passes for some reason.

Link: https://treeherder.mozilla.org/logviewer?job_id=365956541&repo=try&lineNumber=1025

Comment 2

[Daniel Holbert [:dholbert]]

Attachment: block-002-wm-vlr-print.html: screenshot of testcase rendering on Linux

Comment 3

[Daniel Holbert [:dholbert]]

Attachment: block-002-wm-vlr-print.html: screenshot of reference rendering on Linux

Comment 4

[Daniel Holbert [:dholbert]]

Attachment: block-002-wm-vrl-print.html: screenshot of testcase rendering on Linux

Comment 5

[Daniel Holbert [:dholbert]]

Attachment: block-002-wm-vrl-print.html: screenshot of reference rendering on Linux

Comment 6

[Daniel Holbert [:dholbert]]

Attachment: block-002-wm-vlr-print.html: screenshot of testcase rendering on macOS

Comment 7

[Daniel Holbert [:dholbert]]

Attachment: block-002-wm-vlr-print.html: screenshot of reference rendering on macOS

Comment 8

[Daniel Holbert [:dholbert]]

In all of the mismatches here, the failure is a subtle positioning (or antialiasing?) difference in the gray vertical border-segment.

Comment 9

[Daniel Holbert [:dholbert]]

(Adding dependency on bug 1669333 which is where we added these tests.)

Comment 10

[Daniel Holbert [:dholbert]]

note, it looks like maybe the 3x5 cards are rendered to a portrait-oriented canvas on Mac (and only on Mac). I filed bug 1752643 on that; that's likely an oversight, and it's probably why Mac (and only mac) passes the vrl testcase.

Comment 11

[Daniel Holbert [:dholbert]]

(In reply to Daniel Holbert [:dholbert] from comment #1)

On macOS, only the first test ("vlr") fails, with 383/74 px different. vrl passes for some reason.

Update: now we've got the same fuzzy-failure on mac that we have on other platforms; see bug 1828182.