Theme reliability review: 9-10 October 2026¶
The follow-up audit reproduced and fixed additional theme bugs in the SuperQode 2.10.1 checkout. The earlier passing tests did not cover these cases. Each fix has a regression test; this review describes the behavior developers can expect from the current implementation.
Bugs fixed¶
| Developer action | Failure found | Current behavior |
|---|---|---|
| Select an edited catalog theme | The picker attempted installation again and rejected the edit. | Selecting an installed theme applies the developer's version. |
| Reinstall a palette after editing its description | Metadata changes were reported as conflicting colors. | Description changes are retained; conflicting colors still prevent overwriting. |
| Launch or save with deeply nested preferences | JSON recursion could crash theme loading or saving. | Loading falls back with a warning; saving reports the error and retains the file. Preference reads also reject non-regular files and files exceeding 1 MiB. |
| Use a middle-gray terminal background | System mode rejected some gray backgrounds; text on hover/active surfaces could be unreadable. | Surface shades preserve one readable foreground, and hover/active backgrounds participate in contrast validation. |
| Open the picker with an appearance pair | The cursor started on SuperQode rather than the resolved palette. | The current light/dark palette is highlighted. |
| Change terminal appearance with a preview open | The preview stayed stale; repaint hooks omitted other screens. | Repaint hooks run across the screen stack and refresh the preview. |
Launch --theme light in a project containing a file named light | Startup treated the project file as JSON before checking the theme name. | An installed name takes priority; ./light explicitly selects the file. |
| Edit a theme linked from dotfiles | Resolving the link lost its user-directory provenance and disabled hot reload. | File/directory links keep their watch path, including repointed links and atomic replacements retaining size and timestamp. |
| Receive delayed or interrupted terminal color responses | A 100 ms timeout leaked RGB fragments into the composer; unfinished reports could swallow typing and control keys. | Recognized RGB framing survives transport delays, while interrupted headers/payloads recover keys, navigation and paste. Standalone Escape retains its timeout. ANSI reply indices are normalized. |
| Import JSON containing an exceptionally long integer | Python's JSON integer limit raised an uncaught exception. | The CLI reports a theme error and the import dialog stays open with an inline message. |
| Type immediately after startup or use a picker before the delayed focus callback | The sidebar view watcher could focus the hidden file tree, dropping the first composer character or picker key. | The primary screen autofocuses the composer, and hidden sidebar views no longer request focus. |
| Save a palette loaded through a project-file startup override | Only its name was saved, so the next launch could not rediscover it. | Explicit Save installs a native user copy, including palettes in appearance pairs; collisions retain the existing file and preference. The original project file is retained. |
The audit also corrected test isolation. The theme preference path was captured before tests changed HOME, so mounted tests could load the developer's own installed palettes and alter later gallery expectations. Tests now redirect that path into their temporary home.
The broader regression run exposed an unreliable connection-return scroll test. Markdown soft line breaks had collapsed its supposed long transcript into one paragraph, putting the saved position near a changing viewport boundary. The test now creates a genuinely scrollable transcript and waits for the draft layout before saving its position. Both terminal sizes retain the transcript, draft, runtime and exact reading position in that check.
A model-picker check failed intermittently in broad runs while isolated runs passed. Additional diagnostics exposed the startup focus race above: typing could become raft instead of draft, and a deterministic test found the file tree focused when the delayed callback was disabled. Primary-screen autofocus and the hidden-sidebar focus guard remove that initial dependency on the timer. The picker test also now runs in a temporary project, disables unrelated startup/catalog jobs, and waits for its model-list worker. It verifies typing during the pending RPC and immediate Down/Enter selection afterward. Live provider startup remains a separate validation requirement.
Validation¶
The final full regression run passed 6,256 tests, with 30 skipped, 24 integration tests deselected and 139 warnings. The focused theme, navigation, model-picker and developer-workflow run passed all 330 tests. No failures remain in these verified runs; the exclusions and terminal/platform limits below remain part of the release assessment.
The installed core wheel was checked against the source: all 765 packaged Python modules matched, and the wheel contained the 71 catalog palettes, JSON schema and upstream license. Its gallery and import dialog were exercised at 80ร24 and 120ร40; install/apply/save, cancellation, draft retention and repeated installation of the complete collection passed.
The wheel also passed 556 system-background cases (all 256 grays plus 300 seeded RGB samples), 5,419 delayed RGB fragmentation cases, 20 interrupted-report input cases, and native export/import round trips for all 71 catalog themes. The final watch guard also verifies that launching from the themes directory does not try to reload built-in palettes as files. Real macOS PTYs exercised delayed fragments and interrupted headers in addition to typing, paste, resizing and command routing: 62 checks passed for source, and the same 62 passed for the installed package. Source and installed-package results, full regression logs, responsiveness measurements and the strict docs build are recorded under validation/pi-theme-implementation/theme-audit-*.
These checks establish the tested paths, rather than a guarantee that no bug can occur. Real Windows/Linux terminal appearance reporting, SSH transports and live provider workflows still need participant testing. See Themes for supported behavior and Hackathon readiness for the broader developer trial.