Commit Graph
2 Commits
Author SHA1 Message Date
Alby96andClaude Opus 5.5 a368dc8bf0 fix(projects): phase 5a fix round 2 of the projects engine
The sidecars stay the truth of a sequence's projects in the cases round 1's
checkers found still open, and no version or setting is lost on the way:

- An index-only Main meeting a Main another index made is written back as
  another project, and Main passes to theirs in one transaction through a
  free temporary name, so ux_projects_name and ux_projects_main never fail
  (the SqliteException used to escape ListAsync and leave the index and the
  sidecars disagreeing). A refusal is logged; the next reconcile settles it.
- The reconcile takes Main from the records, in step with the index, when
  another index passed it on: exactly one Main per sequence, names from the
  records where free.
- A project folder with versions and no record, of a project no index row
  names (a creation cut before its record), no longer holds the sequence
  back forever: it is left as it is and indexed once its record arrives.
- A record behind the index (a projects folder put back from an older
  backup) is made whole from the index: its missing versions and the record
  are written again, at every catch-up and before any append, rename or
  delete. Only a record ahead of the index or naming another version is
  refused, with a message that says so.
- An index meeting Main walks the chain down to the version the record names
  as mirrored (a bound, since that record write may lag project.json), so a
  lagging record never makes an External that reverts a newer push. When the
  record names a mirrored ancestor, settings written since merge with the
  unmirrored versions, as for an index that knows its mirror.
- The catch-up merge's External + Merge pair is resumable: an External head
  whose parent is a push no mirror wrote is finished into its Merge version
  next time, by this index or one rebuilt from the sidecars.
- A project deleted at the home has no head there, so a workspace copy of it
  reads Gone, never In sync; Remove local copy without discard says the
  project was deleted in the library.
- DatabaseExceptions are contained per sequence in the archive scan's project
  reconcile and in ProjectService reads.

Tests: SQLite (PostgreSQL-ready) for each scenario, including a cut between
the two catch-up versions, a 40,000-exclusion version through save, push and
rebuild, and the App's Remove local copy of a project deleted at home.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 11:24:15 +01:00
Alby96andClaude Opus 5.5 eac4a36a7b fix(projects): one photo merges as one, newer settings are never merged
A workspace copied before the home hashed or converted a photo named it
differently from the versions the home made since, so a pull merged the
same photo as two: frame edits were silently undone and Keep mine placed
the library's value. Before replaying, the workspace media now learn the
home's hashes (same frame id, size and capture time) and conversion
aliases, so every side names each photo alike. Rebind also gives a
renamed entry the occurrence a placement looks for, so positions on
burst photos keep applying once hashed, and folds entries of one photo
(in Rebind and in the merge's keyed form) instead of keeping one.

Settings of a newer Chronos are no longer merged member by member with
this one's: ProjectMerge refuses when both sides changed. A workspace
pull keeps its local versions and working state untouched and records a
conflict of format that only Save both settles (into a new project of
this format); with nothing local it takes the newer head read-only.
Main's catch-up leaves a newer head and settings written meanwhile as
they are.

When Main's catch-up merge overrides values another writer stored, that
writer's settings are kept as an External version before the Merge one,
so nothing is lost from the history.

Unknown members inside Auto values, camera keyframes and curve points,
and on pending entries once placed, are kept like ProjectSettings' Extra.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-03 10:05:27 +01:00