Two changes that go together, because the first is untestable without the second.
`max_size` goes 20 → 50. The comment justifying 20 was out of date: it cited the
256 open-file soft limit macOS gives GUI apps, but `run()` has since called
`rlimit::increase_nofile_limit(10240)` before any pool is opened, so the ceiling
it was sized against is gone. 50 WAL connections is roughly 150 descriptors
against 10240, and only `min_idle` is opened up front — r2d2 creates the rest on
demand — so the idle cost is unchanged. The comment now says what actually
bounds the number: commands nest a couple of connections each, and since #542
every frontend command dispatches as its own task, so they arrive together.
On its own that is a guess with no way to check itself. A pool that cannot hand
out a connection fails with `Error(None)` and nothing else, whether every
connection is checked out or none could be opened — unrelated problems with
identical symptoms, and raising the ceiling only helps the first. So both pools
now acquire through one helper that times the wait, warns past a second, and
folds the pool's state into `Error::PoolTimeout`. The error a command returns
now says which failure it was, which is what the next report needs to be worth
anything.
Blob pool sizing is untouched; nothing points at it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The event task spawned per send still reached `connect().unwrap()`. A pool
timeout there panicked the task and took every later timeline event for that
send with it, straight past the `warn!` branch already sitting underneath. The
acquisition now joins the upsert in the same `if let Err`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two users on 2026.7.x reported the app going half-dead: local editing kept
working, but Send did nothing and every encrypted secret rendered blank until
they restarted. The webview console carried a Rust task panic —
Delete task failed: task 1314 panicked with message
"Failed to get new DB connection from the pool: Error(None)"
`Error(None)` is r2d2's acquire timeout. `QueryManager::connect` and `with_tx`
turned that timeout into `expect`, so a pool that was momentarily full took down
whatever task asked for a connection instead of reporting a failure the caller
could show, retry, or ignore. Every DB-backed path shares those two functions,
which is why one stalled pool reads as "Send is dead and the secrets are gone".
Acquisition is now fallible end to end: `connect()` and `BlobManager::connect()`
return `Result`, and so do the `db()`/`blobs()` accessors on every host. Call
sites propagate with `?` where they already returned a result; the fire-and-forget
writers (streaming events, the model-change poller) log and carry on. `with_conn`
had no callers and only added a third panic, so it is gone.
Two places held a pooled connection across work that has nothing to do with the
database and can block for a long time — the OS keyring in
`EncryptionManager::get_workspace_key`, and the sync directory walk in
`cmd_sync_calculate`. Both now hand the connection back first. Those are the
paths behind the two visible symptoms.
Not fixed here: the gRPC streaming block in yaak-app-client still `.unwrap()`s
its database writes, as it did before, so acquisition failures panic there too.
`ProxyQueryManager` in the separate proxy app keeps its own `expect`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A "Write with Siri" pill appears over empty editors on the macOS 27 beta,
because Codemirror's contentDOM is a contenteditable like any other. The
`writingsuggestions` attribute opts an editable element out, and setting
it here covers every editor at once.
If the pill turns out to be separate from writing suggestions, the
remaining lever is WKWebView's writingToolsBehavior, which would disable
Writing Tools for the whole window.
It has never had an onSelect. Introduced inert in #193 and carried
through the codebase split, so clicking it only ever closed the dropdown.
Reload Schema in the message editor's schema dropdown does the job it
looked like it was for.
`latest` could only ever come from a release tag, and tagging also fires the
app and CLI releases. A dispatch that names a version now publishes it and
latest, which is how the first image gets out before a release exists.
The web build bound Cmd/Ctrl +, - and 0 and then did nothing with them, so zoom appeared broken. Those keys are the browser's own page zoom, which scales the whole page and remembers it per site.
Adds an interfaceZoom capability: true on desktop, where the host zooms the webview, false in a browser. When false the app binds nothing and the hotkeys screen drops the three rows it can't configure.