canvas-pilot/docs/development/handoff.md

8.0 KiB

Conversation handoff — 2026-11-16

Read this when resuming work in the standalone Canvas Pilot repository. This is a snapshot of conversation context, not a replacement for the current code, Git status, plan, contracts or AGENTS.md. Recheck status before acting on dated information.

Where work stands

The migration and README redesign are complete. The user is reopening the conversation in this project; no additional feature implementation is currently underway. Application behavior was preserved during extraction.

Git and publication state at handoff

The migration commit 4eef1bc was pushed and verified on Gitea. Follow-up commits a595838 (audit/agent guidance) and b2fcab4 (README redesign) were committed locally but not pushed. This handoff is another local commit. Check git log origin/main..HEAD and fetch when appropriate rather than assuming this snapshot is still current. Do not push without a user request.

The original repository's extraction/link commit is 7b91ede, committed locally and not pushed. Its working tree was clean at the last check.

The Gitea repository description and avatar were updated through the API and verified. The avatar is crates/canvas-pilot/assets/app-icon.png. The description explains desktop layout editing, offline work and Midra/Aquilon geometry transfers. These settings are server-side; they do not require a Git commit.

The new Jenkins job hookup has not been performed in this conversation. Setup is in CI. The three new badge URLs returned HTTP 404 at the last check; the README initially kept them in a collapsed Build status section pending the first publication. The user subsequently requested moving them back to the top, which has been done. See release readiness for the later no-build audit. No new book deployment was performed during migration. Canvas Pilot publishes only its own book and badges; beffroi-tools retains ownership of the shared manuals landing page. Never synchronize/delete that shared root.

A user-provided Gitea token remains at ../gitea-token.txt on this workstation, outside both repositories. It was successfully used for repository settings and avatar upload. Do not copy or display it. Repository API operations worked; querying /api/v1/user returned 403, so account-profile access must not be assumed. Gitea's own /swagger.v1.json documents the API, including base64 repository-avatar upload.

Decisions and preferences to retain

  • The name is Canvas Pilot. It is independent software, not affiliated with Analog Way. Keep existing settings paths and serialized Midra identifiers compatible.
  • Adapters should be first-class and practical, with a shared intermediate geometry representation and isolated family-specific behavior. Avoid abstraction for its own sake. Preserve targeted reads and partial writes for performance.
  • Aquilon support is completed for the documented standalone AQL Cmax 6.2.x profile; do not reopen the completed task or extend its claim to untested configurations.
  • Multiple editor screens are a distant stretch goal: a dropdown selects one screen at a time. Other adapters need protocol documentation and a usable emulator before development. Their ranking is research, not a promised release schedule.
  • Export cleanup was explicitly marked solved. Saving a captured input as a separate image was removed from the backlog at the user's request; do not restore that task.
  • The operator cannot perform exact pixel/subpixel measurements. Keep precision and mask-rasterization questions in the engineer-facing stretch goal; do not make those measurements a prerequisite for ordinary operator acceptance.
  • Explicit layer aspect modes take precedence over shared input fill. Crop is applied first; 1:1 uses centered native pixels, Fullscreen stretches, Centered fits inside, and Cropped fills/clips. Advanced fit stays behind an opt-in control. Existing helpers support the intended crop/move interactions, and masking uses a separate magenta box. Consult the code and guide before altering their behavior.

Pending input-ratio scope

The user described correcting a stretched/deformed input, usually left on Native, with preset or custom ratios. Manual research established the distinction between Content ratio and Transform to; the result is already recorded in Formats and crop. Advanced device controls already exist. A follow-up question about adding a local editor correction/preview model was not answered before the migration. Do not treat this as an implemented feature or a finalized new schema. Resume from the documented distinction and clarify the desired local behavior when that work is taken up; preserve the input-agnostic editing workflow and opt-in shared writes.

Video plans

The user plans a showcase video and a video manual, and asked whether videos also belong in the book. No videos, hosting choice or embed implementation exist yet. The feature demonstration checklist already supplies an ordered recording script.

The user approved the following approach and asked to put it in the stretch goals:

  • Showcase near the top of the README and book introduction.
  • A walkthrough linked from quick start, with chapter timestamps.
  • Optional short demonstrations beside difficult interactions: crop, 1:1 movement, masks and device mapping/transfers. Start with the two main videos first.
  • Written instructions remain the searchable reference; captions, no autoplay, and the application version help viewers. Host large video files outside Git and link or embed them once a host is chosen.

The existing screenshots are real captures from before the Canvas Pilot rename; the redesigned README explicitly labels them. Their recipes and provenance are in the screenshot guide. Do not edit old screenshots to imply a new capture; recapture the application if updating them.

Device testing context

The user explicitly authorized testing and writes on the emulator at 192.168.10.146:3000 and later said it was online again. At the last recorded work it was an Aquilon emulator; AWJ TCP 10606 was also needed. This is historical scope, not evidence that the endpoint is online or unchanged today. Discover identity before writing, preserve/restore the baseline, and keep emulator results distinct from physical-output acceptance.

The physical Midra at 192.168.10.143:80 was used read-only in the recent work; the emulator authorization does not extend to that device. No device connection is needed for repository, documentation or routine automated tests.

Last validation evidence

The extraction passed formatting, both Clippy feature configurations, rustdoc, Python tests and Rust tests. Canvas Pilot had 395 passing default-feature tests and 390 in each restricted configuration; the remaining workspace had 149 Rust tests. Measured line coverage was 91.03% for Canvas Pilot and 89.83% for beffroi-tools against a 75% floor. These are historical measurements, not live CI badges.

The expanded book built with pinned mdBook 0.5.4 and validated 41 pages and 1,826 local links/assets. The README was rendered locally and visually inspected; local links and badge tests passed. Subsequent documentation edits should rebuild the book; code changes need the relevant checks in CI.