Closed Bug 1382427 Opened 2 years ago Closed 2 years ago
Support running reftest in a mode than runs the test with Retained Display lists on, and references with it off
The Retained DisplayLists feature is a rendering optimization that caches the current display so that later Layout mutations are faster to paint. Our existing reftests are good at validating initial rendering, but not so good at testing Layout mutations with subsequent paints. The current reftest harness does this: 1. start test 2. wait for stable paint 3. take snapshot for bitmap comparison We need to modify the reftest harness so that it will test the cases that break when incorrect DisplayList calculations from Layout mutations end up painting the wrong pixels.
Blocks what was a qf:p1 bug.
I talked to Matt, he says it's very unlikely for this to land in 57. In line with our triage rules that makes this P2.
Whiteboard: [qf:p1] → [qf:p2]
I though the way the reftest harness works tested at least some of this codepath...
Running reftests with the pref reftest.compareRetainedDisplayLists set to true will run them with RDL turned on for the tests and off for the reference files.
Note these patches build on top of the patches from bug 1404181.
Try run with just part 1, running tests with RDL enabled, references with it off: https://treeherder.mozilla.org/#/jobs?repo=try&revision=e902775b6126f9212aa73bb8cfcbabd665d73911 Try run with just part 2, fuzzing reftests without RDL enabled: https://treeherder.mozilla.org/#/jobs?repo=try&revision=53eb85b8c1adbc744241c1a917dafd1e437f0edb Try run with both parts 1 and 2: https://treeherder.mozilla.org/#/jobs?repo=try&revision=898eb728fa6d88ab4801e6f10d3d47b0f88bde66
(In reply to Jonathan Watt [:jwatt] (needinfo? me) from comment #7) > Created attachment 8920063 [details] [diff] [review] > part 3 - WIP fuzzing patch to the reftest harness This is based on the 10 year old add-on, refdyn, from: https://bugzilla.mozilla.org/show_bug.cgi?id=373610 refdyn's code is no obsolete I didn't try and resurrect it in its previous add-on form. Instead I integrated the still useful parts into our current reftest harness and modified it with some of the current ideas from domfuzz.
Attachment #8920063 - Attachment is obsolete: true
I've pretty much rewritten the fuzzer code (e.g. to apply and then remove a stack of changes, rather than push and remove one change at a time) and moved it off to bug 1415558.
Let's make this bug just about supporting running reftest in a mode than runs the test with Retained Display lists on, and references with it off.
Summary: Reftest changes to validate retained DisplayLists → Support running reftest in a mode than runs the test with Retained Display lists on, and references with it off
Comment on attachment 8926424 [details] [diff] [review] patch Review of attachment 8926424 [details] [diff] [review]: ----------------------------------------------------------------- Looks good!
Attachment #8926424 - Flags: review?(matt.woodrow) → review+
Pushed by firstname.lastname@example.org: https://hg.mozilla.org/integration/mozilla-inbound/rev/c2bf7ca0839c Support running reftests with retained display lists on for the test files only. r=mattwoodrow
Backed out changeset c2bf7ca0839c (bug 1382427) for failing reftests r=backout on a CLOSED TREE https://hg.mozilla.org/integration/mozilla-inbound/rev/907fb4683dfef81fbb746035939ac8154b66dfe5 https://treeherder.mozilla.org/#/jobs?repo=mozilla-inbound&revision=c2bf7ca0839c7be4964557d10eab2b198004c803&filter-resultStatus=testfailed&filter-resultStatus=busted&filter-resultStatus=exception
Pushed by email@example.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/4162a11c0a7e Support running reftest in a mode that runs the test with Retained Display lists on, and references with it off. r=mattwoodrow
(In reply to Noemi Erli[:noemi_erli] from comment #15) > Backed out changeset c2bf7ca0839c (bug 1382427) for failing reftests > r=backout on a CLOSED TREE I pushed an old patch from the wrong tree with testing changes to modules/libpref/init/all.js to flip the comparison pref on.
You need to log in before you can comment on or make changes to this bug.