e2e: scroll-chapters can fail on a loaded CI runner (a step just over its 15-verse limit) #157

New issue

Open

opened 2026-11-03 14:00:25 +00:00 by claude-bot · 0 comments

claude-bot commented

2026-11-03 14:00:25 +00:00

Collaborator

scroll-chapters failed once in CI on PR #154 (run 320, head 033b7b5), with the other 45 cases passing:

jump 37 -> 55
down: 39 top-verse reports, steps of at most 18 verses
scroll-chapters: FAIL: a chapter slid on or off made the text jump (down)

Run 320 made 39 top-verse reports going down. Ten local runs of the case on the same commit all passed, with 61 or 62 reports and steps of at most 9 verses. So reports were further apart on the runner, and the steps between them grew. The case's own comment expects this ("more when a loaded one delays the reports").

The limit is 15 verses and a real chapter jump is 22 or more in Gen 1-11, so 18 is between the two. This has not been reproduced locally, so load on the runner is the likely cause, not a proven one. PR #154 doesn't touch the reading pane or scrolling.

Possible fix: judge the step per wheel click or per elapsed time rather than per report, so that a delayed report isn't read as a jump, while keeping a chapter's jump detectable. Simply raising the limit leaves little margin below 22.

🤖 Generated with Claude Code

`scroll-chapters` failed once in CI on PR #154 (run 320, head 033b7b5), with the other 45 cases passing: ``` jump 37 -> 55 down: 39 top-verse reports, steps of at most 18 verses scroll-chapters: FAIL: a chapter slid on or off made the text jump (down) ``` Run 320 made 39 top-verse reports going down. Ten local runs of the case on the same commit all passed, with 61 or 62 reports and steps of at most 9 verses. So reports were further apart on the runner, and the steps between them grew. The case's own comment expects this ("more when a loaded one delays the reports"). The limit is 15 verses and a real chapter jump is 22 or more in Gen 1-11, so 18 is between the two. This has not been reproduced locally, so load on the runner is the likely cause, not a proven one. PR #154 doesn't touch the reading pane or scrolling. Possible fix: judge the step per wheel click or per elapsed time rather than per report, so that a delayed report isn't read as a jump, while keeping a chapter's jump detectable. Simply raising the limit leaves little margin below 22. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

claude-bot referenced this issue

2026-11-03 14:31:03 +00:00

CI: cut the e2e job's 17 minutes on one runner (after #148) #158

claude-bot referenced this issue

2026-11-03 15:27:25 +00:00

Modules redesign (#148 step 4) #154