darling-durability

name

darling-durability

description

Make in-progress Darling work durable so it survives checkout/reset/handoff: triage dirty & uncommitted work, rescue at-risk commits onto clean fix/* branches, commit the manifest repo, run west dw handoff, and verify. Use before ending a session, before any west update / git checkout / west patch apply, or whenever `git status` shows uncommitted work in the Darling workspace. Prevents silent loss of perf work (e.g. perf#21b nearly lost off a generated integration branch; the whole manifest repo was once uncommitted).

darling-durability: make workspace work durable

The Darling workspace has three durability traps, all of which have bitten us:

1. Uncommitted worktree changes — lost on any git checkout/west update. west dw handoff CANNOT carry dirty worktrees (it only prints them).
2. Commits only on a GENERATED integration/<profile> branch — west patch apply/clean regenerates that branch and DISCARDS anything committed only there.
3. An uncommitted manifest repo (darling-workspace) — the single most fragile spot; it holds the exported patch profile (patches.yml + patch files), tooling, and the handoff bundles themselves.

Canonical rules (workspace CLAUDE.md): clean fix/* branch = editable source; patch files + patches.yml = portable; integration/* + profile west.lock.yml = generated (never edit). Never push unless explicitly asked.

Procedure

1. Triage — find every at-risk piece

cd ~/work/darling-dev
west forall -c 'test -n "$(git status --porcelain)" && echo "DIRTY: $(pwd)"' 2>/dev/null | grep -i dirty

Also inspect the manifest repo itself (it is NOT in ):