When we see test runner in "commands" section of tox, we substitute it with our test runners so user may benefit from test trees.
We also pass "offset" because test runners must use it as "parent node".
Set "_jb_do_not_patch_test_runners" to disable this substitution
(cherry picked from commit c7d469e1fed52f2d7fe3aec6d96b665c4aa80a4e)
GitOrigin-RevId: b9eec43a6d26fecb72270b733266ef31989fec2c
- changed behaviour in safe way by introducing new method to check that particular expression does not depend
on committed PSI
GitOrigin-RevId: f5ec732613cdafdcef8a1d48eb8d04135c634047
- call `commitDocument` explicitly in cases when `calculateQuickResult` can redirect it's calculation to heavy `calculateResult`
GitOrigin-RevId: 41f0c47c57f4528d7d7a25208024ec28bb8c2e12
- removed `commitDocument` call when `isQuick` argument is passed
- pass `element` or `null` to `equalsToText` call (since there is not real need to take it for text comparison)
GitOrigin-RevId: 27c5e1a2f484fc56d41dfd210832ae94c9a9b7ab
With xdist, setupClass may be reported several times (one for each test). Next report leads to error since it is already
reported as closed.
We ignore setup methods: there is no sane place to display them in the test tree, anyway.
GitOrigin-RevId: 3c7bcf2f3942e78fcbfba5fb0cae1cbdb2c92730
Documented in ``TestDurationStrategy``.
``MANUAL`` is used for multiprocess tests where root must use "wall time" (calculated on Java side) and suites must not have duration at all.
In parallel mode tests are still reported sequentially, but out of order and testFinish is never reported for suites, because we do not know when suite is completed, so they must be closed on Java side.
Strategy is also passed to Java side to change duration calculation logic: root duration is wall time, not sum of all children
