Skip to content

CI-CD tests: fix the package Playwright specs failing in the nightly - #4019

Merged
LesiaPavlenko merged 5 commits into
masterfrom
opavlenko/cicd-triage-0831
Aug 31, 2026
Merged

CI-CD tests: fix the package Playwright specs failing in the nightly#4019
LesiaPavlenko merged 5 commits into
masterfrom
opavlenko/cicd-triage-0831

Conversation

@LesiaPavlenko

Copy link
Copy Markdown
Collaborator

Five nightly package-Playwright failures, all measured and fixed on dev. Nightly Build-Deploy #1295#1297 reported the same 22 package failures three nights running; these are the ones that turned out to be test bugs.

  • softStep turned declared skips into failures. The BiostructureViewer specs raise Error('Test is skipped: …') for a Mol*/WebGL precondition and filter it out of realErrors at the end of the test. softStep re-threw any error with that message prefix, so it escaped the step and Playwright failed the test. Only a real TestSkipError — what test.skip() raises — is re-thrown now. All 10 files using the - PowerPack / Formula refreshing. The spec drove the Formula info panel through .add-new-column-dialog-cm-div. That class belongs to the dialog; the widget-hosted editor uses -widget-cm-div, and has since 979ac73 fixed the constructor to read its widget argument. Written before that, the spec could never find the host, and all four of its failed steps came from it — including the save-reopen persistence check, which read like a separate product bug. Now accepts either class. Green on dev in 35.8s.
  • Peptides / Info Panels. Setup waited 30s for the Context Panel's Details pane immediately after opening the table, before anything was selected. The panel renders the panes of the current object, which at that point is the platform info object. Measured: Details appears the instant grok.shell.o is the column, which step 2 already does. Green on dev in 17.0s.
  • Bio / empty-input ×3 — test.fail() removed. Team decision: a test blocked by a known issue stays a plain failure and carries the ticket number. GROK-16111 stays in the assertion message; the tests are deliberately not renamed, since tests.name is the dashboard's key and a rename leaves the old title behind as a permanently red row.
  • Bio / monomer_library. S1.4 polled 10s for the Bio:manageMonomerLibraries dialog. Measured on dev: the first dispatch takes 12.3s, a second 0.2s — the first call pays the library load, and CI is slower than dev, which is why it failed there nightly and passed locally. Poll widened to 40s. Green on dev in 1.1m.

Every spec here was run against dev before and after. Merges cleanly onto current master.

LesiaPavlenko and others added 3 commits August 31, 2026 18:00
…ng it

softStep re-threw any error whose message starts with "Test is
skipped:". That is the suites' declared-skip idiom - the step records
it and the spec's end-of-test realErrors filter drops it - so
re-throwing escaped the step and Playwright failed the test. Two
BiostructureViewer specs were reported red for a Mol*/WebGL skip that
CI cannot satisfy. Every file using the idiom filters it at the end,
so nothing else changes. Only a real TestSkipError, which test.skip()
raises, is re-thrown now.
The spec drove the Formula widget through .add-new-column-dialog-cm-div.
That class only belongs to the dialog: the widget-hosted editor labels
its CM host -widget-cm-div, and it did so from 979ac73, which fixed
the constructor to read its widget argument rather than the field it
assigns later. Written before that, the spec kept the dialog class, the
host could never be found, and all three info-panel edits failed - along
with the save-reopen persistence check downstream of them, which read as
a separate product bug.

Accept either class. Verified on dev: the spec is green in 35.8s.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QN9RdbhoZ9sEoDNeM1EdCh
The Setup step waited 30s for .grok-prop-panel [name="pane-Details"]
straight after addTableView, and step 2 waited for it again before
setting grok.shell.o. The Context Panel renders the panes of the
current object, and right after a table view opens that is the
platform info object - Client/Server/Services/Plugins, no Details.
Measured on dev, Details and Peptides both appear the moment shell.o
is the AlignedSequence column, which is what step 2 does.

Setup now checks only that the panel is open. Verified on dev, green
in 17.0s.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QN9RdbhoZ9sEoDNeM1EdCh
@LesiaPavlenko
LesiaPavlenko force-pushed the opavlenko/cicd-triage-0831 branch from b4e0849 to c29bbd6 Compare August 31, 2026 15:00
LesiaPavlenko and others added 2 commits August 31, 2026 19:30
Team decision: a test blocked by a known issue stays a plain failure
and carries the ticket number, rather than being marked test.fail().
An xfail hides a real product bug behind a green suite, and Playwright
and the CI report disagree about what it means.

The three empty-input-row-viewers specs drop test.fail(). The ticket
number stays in the assertion message, so the failure names it in the
CI report without renaming the tests - a renamed test would leave its
old name on the dashboard as a permanently red ghost row.

Verified on dev, 3 failed, all on the balloon assertion, which is what
CI already reports.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QN9RdbhoZ9sEoDNeM1EdCh
S1.4 polled 10s for the Bio:manageMonomerLibraries dialog. Measured
on dev, the first dispatch takes 12.3s and a second one 0.2s - the
first call pays the library load - and CI is slower than dev, which
is why the step failed there every night while passing locally. Poll
for 40s instead.

Verified on dev, green in 1.1m.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QN9RdbhoZ9sEoDNeM1EdCh
@LesiaPavlenko
LesiaPavlenko force-pushed the opavlenko/cicd-triage-0831 branch from c29bbd6 to 21183af Compare August 31, 2026 16:31
@LesiaPavlenko
LesiaPavlenko merged commit b12b825 into master Aug 31, 2026
12 of 14 checks passed
@LesiaPavlenko
LesiaPavlenko deleted the opavlenko/cicd-triage-0831 branch August 31, 2026 16:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant