1341077 - Import CSS reftests from upstream

Status: RESOLVED FIXED

(Testing :: web-platform-tests, defect)

Description

[James Graham [:jgraham]]

Once CSS reftests land in wpt upstream we need to import them into m-c as part of the normal import process. This is expected to cause ~30,000 additional files to be added, so will need to be checked for perf. consequences.

Comment 1

[Sam Sneddon [:gsnedders]]

There's also the duplication that we'll have from <layout/reftests/w3c-css/submitted>, which is currently one-way synced to csswg-test.

Comment 2

[Sam Sneddon [:gsnedders]]

Given as far as I'm aware the script to sync <layout/reftests/w3c-css/submitted> to csswg-test <vendor-import/mozilla> are currently tied to Mercurial, it seems like on the face of it there isn't any point in rewriting the script and having a second syncing system for wpt. If that's the case, I presume we can then move the tests out of vendor-import and into their respective spec directories?

Comment 3

[Gregory Szorc [:gps]]

When I looked at the WPT files in the Servo repo, I saw that there were a number of very small files defining what are essentially per-test settings. Given that there is a non-trivial cost to a file in the checkout (slows down several version control operations both locally and in automation), would there be a way to consolidate these settings files so there were fewer files? Is that even a problem for these CSS reftests?

FWIW, I last brain dumped about the consequence of excessive files in version control in bug 977849 comment 29 and below. As a person tasked with scaling version control at Mozilla, I'm committed to supporting whatever developers say we need to do, which includes running a monorepo for Firefox. I just want to encourage a reasonable amount of due diligence so we avoid unwanted strain on version control.

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

So I'm very uncomfortable with this, given my concerns about the way we run wpt reftests (multiple bugs filed, probably not complete) and my opinion that we should stop using a separate harness for them.  The existing tests we have in w3c-css/submitted/ definitely need to continue to be run with our own reftest harness given that we'd otherwise lose substantial test coverage.