You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
The service-worker cache name is a constant, but migrations.md claims it is release-keyed #48
docs/migrations.md states that the service-worker shell cache is release-keyed and bumped on every
deploy, so that activate deletes the previous release's cache wholesale. The code does not do
that.
docs/migrations.md — "public/sw.js names its cache for the release (bumped on every deploy), so activate deletes the previous release's cache wholesale."
public/sw.js:23 — const CACHE_NAME = "obelus-shell-v1";, unchanged across every commit that
has touched the file (git log -S"obelus-shell" -- public/sw.js shows one commit).
public/sw.js:22 — the comment // Keyed to the release: bump this on every deploy. is therefore
false.
pnpm run deploy has no bump step, and no gate checks it.
Impact
activate's keys.filter((key) => key !== CACHE_NAME).map((key) => caches.delete(key)) loop is
inert, so the documented rollback mechanism is not the one doing the work. Rollback correctness then
rests entirely on the navigation-time syncShellAssets prune, which only runs on a successful
online navigation. The behaviour is probably acceptable; the documentation is wrong about which
mechanism is load-bearing, which is the part that matters when someone reasons about a rollback.
Fix
Either make the claim true — derive CACHE_NAME from a build-injected release id and add a gate
that it differs from the deployed one — or correct docs/migrations.md and the sw.js comment to
describe the prune as the actual mechanism.
Do not change the migration policy itself; this is only about the claim.
Origin
Found by the Thermos review (thermo-nuclear, storage/UI + code-review Standards axis).
Decision (Writer, 2026-09-26): correct the docs. Rewrite the docs/migrations.md passage and the public/sw.js comment to name the navigation-time syncShellAssets prune as the mechanism that does the work. CACHE_NAME stays a constant. No behaviour change, no deploy step, no migration-policy change.
Fixed in #70 (merged into v1.3), following the 2026-09-26 decision: docs and the sw.js comment now name the navigation-time prune; CACHE_NAME unchanged.
Activity
kiran-brahma commentedon Sep 26, 2026
Decision (Writer, 2026-09-26): correct the docs. Rewrite the
docs/migrations.mdpassage and thepublic/sw.jscomment to name the navigation-timesyncShellAssetsprune as the mechanism that does the work.CACHE_NAMEstays a constant. No behaviour change, no deploy step, no migration-policy change.Name the navigation-time prune as the shell cache's rollback mechanism (
kiran-brahma commentedon Sep 26, 2026
Fixed in #70 (merged into
v1.3), following the 2026-09-26 decision: docs and thesw.jscomment now name the navigation-time prune;CACHE_NAMEunchanged.Generated by Claude Code