Skip to content

test(completion): trace a missing snippet selection - #149

Draft
BartvHelvert wants to merge 1 commit into
masterfrom
test/snippet-selection
Draft

BartvHelvert wants to merge 1 commit into
masterfrom
test/snippet-selection

Conversation

@BartvHelvert

@BartvHelvert BartvHelvert commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner

What this changes

Investigation of #146, rebased onto master and reduced to the test it belongs to.

SplitModeSnippetsTest records the client's text and selection together with the backend editor's
text and selection on each observation, traces the platform's template state in both IDEs, and
keeps a UI hierarchy if the scenario fails. The existing checks for typing and Tab navigation stay,
and insertion is still performed once.

Refs #146

What changed since it was opened

The repeat_ui_test workflow commit is gone from this branch. It was duplicated with #150,
differing only in the hard-coded test class, so it is being landed once on its own, generalised to
take the class as a workflow input. Repeats of this class will be dispatched with that instead.

Its batch found a failure, but not this one. The twenty jobs on 1dd8913d produced one, in
windows-2025 repeat 8/10: [first, PROBE Widgetbody {, , }] — the Tab between the
placeholder and the body did nothing, so the body text landed on the end of the placeholder. That
is a different step from this PR's, it was diagnosed by the pre-existing assertion rather than by
the diagnostics added here, and it has its own cause: a key press discarded while the IDE waits on
a lock. Filed as #153 and fixed in #152.

#146 itself did not reproduce in twenty runs and is still unexplained. Its distinguishing fact
stands: insertion provably succeeded — CompletionFinished(reason=SUCCEEDED), client insertion
logged — and the placeholder still never appeared selected. Nothing found since accounts for that,
and the lost-keystroke mechanism does not: a discarded Enter would have meant no insertion at all.

So the sampling added here is still aimed at the open question, which is whether the text failed to
expand or the selection was lost after it did.

What to do with it

Draft, and still worth keeping drafted. Next step is another twenty or thirty repeats through the
extracted harness, once that lands. If #146 will not reproduce, close this and leave the issue open
with what is known; the diagnostics can come back when it next fires.

How it was tested

Checklist

  • Nothing is special-cased for one extension.
  • A new feature has a UI test. Diagnostics in an existing UI test.
  • No internal, override-only, obsolete or deprecated platform API. Template implementation
    names are log categories only; remote reads use public Editor and SelectionModel API.
  • docs/ is updated where behaviour changed, docs/status.md especially. Investigation only.
  • Third-party material keeps its notice and is listed in NOTICE.md. None added.

A single completion request can succeed without leaving the first
placeholder selected. Capture text and selection in both processes,
trace template state, and retain a failure hierarchy to distinguish
a missing insert from a lost template selection.

Refs #146
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