Adds Storage_Server next to upstream's Storage_Local and swaps which one
is instantiated; the rest of nullboard.html is untouched so upstream
changes still merge.
The Storage contract is synchronous — setItem has to return true or false
there and then — which no round trip can satisfy. So the whole keyspace
is held in memory and the server is somewhere to push it to:
- The initial copy is embedded in the page by the server, already
parsed and in hand before the app boots. In the page rather than a
.js file so another site can't <script src> it and read the boards.
- Writes are collected by key, so a note edited repeatedly is one
write, and flushed shortly after as a single transaction. A failed
batch goes back on the queue and retries with a backoff rather than
being dropped.
- The logo shows saving/not saved, and closing the tab with writes
still queued asks first.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UU1vyTHj3uE9PJYSxRxwkU
Nullboard's storage layer bottoms out in getItem/setItem/delItem over
string keys, with boards, revisions and undo history all modelled on top
of them client-side. So the server needs to be nothing more than a
key/value store, and it doesn't parse a board anywhere.
One password, bcrypt-hashed, in the same database as the boards. The
session secret lives there too, so restarting the service doesn't sign
you out and the secret never has to exist in the unit file or in git.
Until a password is set every route returns 503 pointing at
bin/validboard-passwd. A first-run setup page would be friendlier, but it
would also hand the board to whoever found the URL first.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UU1vyTHj3uE9PJYSxRxwkU
