Autosave only ran after 4s of idle, and every keystroke reset that timer, so a
writer who never stopped had zero saves (measured: 15s of typing = 0 requests).
Saving also never rebound the URL, so a new draft could not be found again after
a reload, the local backup was cleared on every autosave, and a failed save had
no retry and stayed failed once the network came back.
Client
- autosave: keep the idle debounce but add a 30s max-wait window that is not
reset by further keystrokes, and retry failures at 5s / 15s / 60s
- bind `?edit=<id>` into the URL once the server returns a post id, so reloads,
new tabs and history all reopen the same draft
- only clear the local backup when the save settled and nothing changed since
- key the local backup by draft id instead of the constant `jio-studio-0`
- restore snapshots through hydrateBlocks so code blocks come back editable
- never disable publish while an autosave is in flight, and mark autosave
states separately (自动保存中… / 已自动保存 / 自动保存失败，将重试)
- on 409, adopt the server rev and tell the author instead of failing silently
- status bar reports the real save state instead of a bogus Markdown hint
Server
- draft saves no longer rewrite the lifecycle: pending / future / publish /
private posts keep their status, so an autosave cannot pull a submitted post
out of the review queue
- drafts and autosaves stop consuming the publish flood quota, which is meant to
cap actual publishing (measured: 4 drafts used to burn 4 slots)
- autosave skips wp_save_post_revision for both plain drafts and submission
versions; manual saves and submits still create a revision to roll back to
- media-only drafts (image / table / code / hr) count as content
- the response now carries the resulting status so the client can tell
"just submitted" from "still under review"
Tests cover the max-wait window, retry schedule, URL binding, backup clearing,
status labelling and every server-side draft rule.
Mark new regular and chunked uploads as pending and claim persisted article references, including partial saves and large reference sets. Protect associated media from cleanup and add upload lifecycle regression coverage.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
kate522
