Skip to content

e2e: recordings from Release 98; replay keys a query's parameters in order - #775

Merged
adamjohnwright merged 1 commit into
mainfrom
e2e/release-98
Nov 2, 2026
Merged

adamjohnwright merged 1 commit into
mainfrom
e2e/release-98

Conversation

@adamjohnwright

Copy link
Copy Markdown
Contributor

No spec: test fixtures and the replay harness. Moves the e2e recordings from Release 97 to Release 98, as the beta backend now serves.

What changed

  • The whole suite re-recorded against Release 98. Release 97's saved release files are removed, and so are two orphan recordings (tests renamed long ago, the only ones still saying 97).
  • Replay keys a request's query with its parameters in order. The site asks for one pathway's details with the same parameters in two orders, from two services, and which one a test makes depends on timing; replay aborted the order it had not recorded. The probes' .api.json keys are normalised on load, and check-har uses the same key. A unit test covers it, seen failing without the sort.

Tests the move needed

  • A new analysis result for the seven tests that read one: the Release 97 result was removed with the release.
  • Circadian clock's illustration lost its <foreignObject> in Release 98, so the "a figure that cannot be made says so" test puts one back.
  • The dragged-tissue test holds its submission unanswered: it is about the stepper.
  • The legacy-link tests back out mid-load; the reads they abandon join check-har's ABANDONED list.
  • Stubbed answers say release 98.

Checks moved, with reasons

  • known-problems.json: six totals up. Compared kind by kind with main, all but one are kinds main already counts. The routes have more recorded answers (3–4 more on three, 1 more on the reaction exporter), and Release 98's objects are larger. The one new kind is a releaseStatus field in a pathways/low/entity answer, which the site's type does not list (1 instance).
  • schema.spec: no recorded search answer holds a deleted object's entry any more (Release 97 offered two for TP53; Release 98 offers none for anything the tests ask), so that entry kind is no longer covered by recorded data.

Also

Local results: 424 passed, 0 failed, 7 skipped. One of those skips was caused by the key change; it is fixed and the test passes. Release 97's last skipped count was 6.

🤖 Generated with Claude Code

…order

The whole suite re-recorded against the Release 98 backend, and Release
97's saved release files and two orphan recordings (tests renamed long
ago, the only ones still saying 97) removed.

What the move needed:
- A new analysis result for the seven tests that read one: the Release 97
  result was removed with the release. Made on the local analysis service
  from eight checkpoint genes; it finds 8 entities in R-HSA-69620.
- Circadian clock's illustration lost its <foreignObject> in Release 98, so
  the test that an unexportable figure says so puts one back.
- The dragged-tissue test holds its submission unanswered: it is about the
  stepper, and an answered submission moves the form past the step.
- The legacy-link tests back out of the pathway browser mid-load; the reads
  they abandon are listed in check-har's ABANDONED, as two already were.
- Stubbed answers say release 98.

Replay keys a request's query with its parameters in order. The site asks
for one pathway's details with the same parameters in two orders from two
services, and which a test makes depends on timing: a recording held one,
replay was asked for the other and aborted it. The probes' .api.json files
are keyed again on load, so their older keys still match. check-har uses
the same key.

Checks moved, with reasons:
- known-problems.json: six totals up. Compared by kind with main, every
  kind but one is one main already counted; there are more recorded answers
  on those routes (3-4 more on three, 1 on the reaction exporter) and
  Release 98's objects are larger. The one new kind: a releaseStatus field
  in a pathways/low/entity answer, which the site's type does not list.
- schema.spec: no recorded search answer has a deleted object's entry any
  more (Release 97 offered two for TP53; Release 98 offers none for what
  the tests ask), so that kind is no longer one the recordings cover.

Also: the reaction view's render state carries no analysis (tested; set
aside from #772 for this move), and the failure message drops a host
written in capitals.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@adamjohnwright
adamjohnwright merged commit b9ae173 into main Nov 2, 2026
8 checks passed
@adamjohnwright
adamjohnwright deleted the e2e/release-98 branch November 2, 2026 16:24
adamjohnwright added a commit that referenced this pull request Nov 2, 2026
From #775's review: the popup's type lists releaseStatus; ABANDONED pinned; comments
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