@hienlh/ppm 0.13.4 → 0.13.6

package/dist/web/assets/keybindings-store-V12kZZHO.js

@@ -1 +0,0 @@
-import"./vendor-markdown-0Mxgxy0L.js";import"./api-client-r4nyVy7H.js";import{E as e}from"./index-DJOjXTcq.js";export{e as useKeybindingsStore};

package/docs/journals/2026-04-22-compare-files-feature-ship.md

@@ -1,53 +0,0 @@
-# Compare-Files Feature Ship
-
-**Date**: 2026-04-22 14:35
-**Severity**: Low
-**Component**: Web IDE / Diff Viewer
-**Status**: Resolved
-
-## What Happened
-
-Shipped compare-files feature (file diff picker) for PPM web IDE. Four trigger paths — tab context menu, file-tree context menu, command palette, keyboard shortcut `Mod+Alt+D`. Reused existing DiffViewer + git-diff tab type; zero backend changes. Added zustand store with persist middleware, ComparePicker singleton at App root, and in-flight guard to prevent double-invoke race.
-
-## Technical Details
-
-- **370 new lines of code** across 3 new files + 6 edited files
-- **Zustand store** (`useCompareStore`) with `persist` middleware that strips dirty content >500KB
-- **Dirty-buffer semantic**: snapshot at select-time, not compare-time (closing the tab doesn't invalidate selection)
-- **Custom event** (`open-compare-picker`) dispatched from 4 trigger points, singleton listener mounted at App root
-- **Keyboard shortcut**: `Mod+Alt+D` single-stroke (chord parser doesn't support `Cmd+K D` yet)
-- **Module-scope subscription**: `useProjectStore.subscribe` auto-clears selection on project switch
-- **In-flight guard**: ref-based tracking in `handlePick` to block concurrent invocations (reviewer comment M5)
-- **Test coverage**: 1698 pass, 48 pre-existing unrelated failures, 0 new regressions
-- **Code review**: 7.5/10, 0 critical, 6 major (1 addressed: M5; 3 deferred: M3/M4/M6)
-
-## Why This Matters
-
-Snapshot-at-select semantics felt risky initially but aligns with user mental model: "I'm picking a version to diff." Closing the tab doesn't invalidate the selection — this is more intuitive than invalidating on every interaction. Deferring byte-accurate persist limit (M4), clear-on-error (M3), and legacy multi-select-compare unification (M6) kept scope tight; all are low-risk YAGNI.
-
-## Key Decisions
-
-1. **Same-project only**: Cross-project selection auto-cleared rather than supported. Reduces state complexity; users can switch projects and re-pick.
-2. **Single-stroke `Mod+Alt+D`**: Chord parser limitation means `Cmd+K D` unsupported. Single-stroke is good enough; avoiding yak-shave on chord parsing.
-3. **Snapshot timing**: Capturing dirty buffer at select-time (not compare-time) prevents stale diffs after file edits. User closes tab; selection still valid.
-4. **Defer judgment calls**: M3/M4/M6 are correctness edge-cases, not showstoppers. Shipping first; addressing in follow-up if telemetry justifies.
-
-## Lessons Learned
-
-- **Persist middleware footprint**: 500KB strip limit is conservative but safe; no observed bloat in dev.
-- **Custom events + singleton**: Avoids prop-drilling 4 trigger points through component tree. Event-driven pattern cleaner than callback props.
-- **In-flight guard as ref**: Simpler than promise-based debounce; prevents race without async complexity.
-- **Deferred ≠ broken**: 6 major review comments sound high, but only 1 was blocking (M5). Others are judgment-calls; shipping with known trade-offs is valid.
-
-## Next Steps
-
-- Monitor telemetry for M3 (clear-on-error) if users report stale diffs
-- Consider byte-accurate persist limit (M4) if DB bloat observed
-- Unify legacy multi-select-compare path (M6) in v0.14 if feature gains adoption
-
-Commit: `cc09bb5b9e652a2feefde4e533934a6f53301895`
-
----
-
-**Status:** Resolved
-**Summary:** Shipped compare-files feature with 4 trigger paths, zustand persist store, and event-driven ComparePicker. 370 new LOC, 0 new test regressions, 1 review blocker addressed (M5), 3 judgment-calls deferred (M3/M4/M6). Snapshot-at-select semantics and same-project-only scope kept complexity tight.