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
Chain::resync_realtime resynchronizes a running realtime endpoint without stopping it: the
callbacks play the fallback while every plugin is reset as by reset, which ends the notes the
plugins hold and applies an announced latency, and the helper's worker goes on with a new
generation. It returns the endpoint's whole latency, as start_realtime does, and changes
nothing on a stopped endpoint. Stopping the endpoint, resetting the chain and starting it again
still works.
Chain::snapshot returns a ChainSnapshot with each slot's project state and native bypass,
also while the endpoint runs. Chain::recover_realtime(&snapshot) recovers the endpoint in a
new helper: it prepares a chain from the snapshot, applies the bypass, and moves the same
endpoint to it while the audio callback keeps the RealtimeEndpoint it has. The callback plays
its fallback during the recovery, and after latency priming callbacks the new chain's output.
Switch to the returned chain and drop the old one. A failed recovery leaves the endpoint with
the old chain and retries nothing. recover_realtime loads with the load timeout and prepares
with the control timeout.
RealtimeStatus::stalled reports how long the helper's worker has gone without completing a
quantum the callback submitted, which grows while a plugin hangs in its processing; a crashed
helper shows through helper_monitor. Together they tell when to call recover_realtime.
Chain::reprepare_slot prepares one slot again on a fresh instance holding the active one's
project state, which drops a pending restart request, as reprepare does for the chain. It
uses the load timeout and works while the realtime endpoint runs.
Changed
restore_state, import_preset, load_discovered_preset and select_factory_preset work
while the realtime endpoint runs: the helper stages the fresh instance as it does offline and
the worker swaps it in between quanta, and the stream goes on. A replacement that changes the
chain's latency resynchronizes the endpoint right after, as resync_realtime does. Only reset, prepare_audio, reprepare, realtime_endpoint, processing through the chain and
rendering still fail with Error::RealtimeRunning; use resync_realtime instead of reset,
and reprepare_slot instead of reprepare for a restart request.
The first callback of every new generation (a start, a resync or a recovery) ends the notes
that earlier outputs delivered through produced and did not end, with note offs at offset 0;
the endpoint keeps track of up to 256 notes, and events that do not fit count as dropped_events. Notes a replaced instance sent to later slots are still ended only by a
reset or resync.
restore_state, presets and factory presets on a slot with a pending VST3 or CLAP restart
request now succeed, also offline, and drop the request with the replaced instance; they
failed with RestartRequired for that slot before.
The internal helper protocol is now 29. Applications and helpers with different protocol
versions fail the handshake; rebuild and distribute the helper with the application.
Fixed
VST3 reset, the VST3 parameter flush before prepare_audio and Audio Unit reset also run
with subnormal numbers flushed to zero, as 0.1.3 documented for every format.
Discovery reads a VST3 bundle's moduleinfo.json in time and memory proportional to its size.
The JSON5 parser copied a string or identifier again for each escape sequence in it, so a 256 KB
file of escapes allocated 17 GB and took a third of a second; plughost now requires json5 1.3.3,
which fixes it.