1550231 - [wpt-sync] Sync PR 16393 - [css-text] Follow up on #15309

Status: RESOLVED FIXED

(Core :: CSS Parsing and Computation, defect, P4)

Description

[Web Platform Test Sync Bot [:wpt-sync] (Matrix: #interop:mozilla.org)]

Sync web-platform-tests PR 16393 into mozilla-central (this bug is closed when the sync is complete).

PR: https://github.com/web-platform-tests/wpt/pull/16393
Details from upstream follow.

Florian Rivoal <[e-mail]> wrote:

[css-text] Follow up on #15309

This cleans up and addresses a number of issues in the tests added
in #15309. The main intent of the original tests is unchanged, but:

• Remove unnecessary code, such as properties set to their initial value

• Remove title and comments about the test from its reference:
  references can be used from multiple tests, so they should not describe
  any particular test.

• Reuse identical references (and delete redundant ones)

• Delete shaping-015.html, which was identical to -014

• Rephrase test assertions to say MUST rather than SHOULD, as that's
  what we mean.

• Rephrase the user-visible test instructions, so that they can more
  easily be followed by someone who is not familiar with the language

• Fully automate all tests: many of them contained a statement along the
  lines of "skip if ...", but this isn't checked by the automated
  runner, which would have reported false positives. Instead, use
  mismatch references (in some cases several of them) in addition to the
  main reference to make the test fail if the precondition is not met.
  Similarly, automate the Mongolian tests by using a mismatch reference
  with text-orientation:upright to catch the cases of browsers not
  properly supporting Mongolian.

• Delete the shaping_cchar-* tests. While they definitely outline
  desirable behavior, that behavior is neither required by the
  specification nor fully testable.

  It is not required by the specification, because the spec explicitly
  says that typographic character units divided by an element boundary
  result in undefined behavior. These are very interesting exploratory
  tests, and if we can in a later level narrow down the specification to
  clarify (at least for a subset of the general case) how this is
  supposed to work, maybe they can make a come back.

  Arguably, it could be interesting to keep them as "should" tests,
  as they are desirable even if unspecified behavior, but I don't think
  they can be made to be reliable as automated tests. Since both the
  test and of the reference depend on undefined behavior, they might
  both misbehave the same way, but as that way is unpredictable, even
  with combination of multiple match and mismatch references, I don't
  think it is possible to make an automated reftest that reliably fails
  when the browser gets it wrong. If we can make the more robust,
  I don't have an issue with them coming back flagged as "should" tests,
  but I don't think they're ready to go in their current form.

Comment 1

[Web Platform Test Sync Bot [:wpt-sync] (Matrix: #interop:mozilla.org)]

Pushed to try https://treeherder.mozilla.org/#/jobs?repo=try&revision=b81e83366a3b1be9247076cf46b7d2181ced94eb

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/097ed6793063
[wpt PR 16393] - [css-text] Follow up on #15309, a=testonly

Comment 3

[Stefan Hindli [:stefan_hindli]]

https://hg.mozilla.org/mozilla-central/rev/097ed6793063