Skip to content

Release 4.5.2 (18.11.2026) — migration of large databases, all-or-nothing migration (schema v80)

Choose a tag to compare

@github-actions github-actions released this 18 Nov 01:52
· 165 commits to main since this release
b9b2703

Release 4.5.2 (18.11.2026) — migration of large databases, all-or-nothing migration (schema v80)

Migration only: no change to the model, results or data version. A failed migration now leaves the original database in place.

Fixes

  • A large database migrates. A migration step that deletes many rows (values, flow-group memberships, entity
    alternatives) stopped with "Expression tree is too large (maximum depth 1000)" when the rows were scattered over
    the database, as in a large database edited over time. Such a step now deletes them in several commits.
  • A failed migration leaves the database as it was, also from the command line. flextool-migrate-database,
    flextool-run --migrate and the other migrations outside the FlexTool window now copy the database before the
    first step (as <database>.premigration.bak, which needs free space as large as the database) and move it back
    in one rename when a step fails or the migration is cancelled, as the window already did; before, a migration
    that failed part way left the steps before the failure in the database. When the copy cannot be put back, it is
    kept, the console says how to recover it, and a new migration of the database refuses to start until it is dealt
    with.
  • A migration that is killed outright can be finished. A killed migration (the process stopped, the console
    closed) can leave the database half migrated, with its original in <database>.premigration.bak. The next
    migration of the database then refuses to start, names the copy and the time it was made, and says how to go on:
    put the copy back (close every program that uses the database, delete the database and rename the copy to the
    database's name) or, when the copy is out of date, delete it; then run the migration again. Every step that
    deletes in several commits also finishes on a rerun: before, a rerun of the version 56 step could lose is_DC /
    is_non_synchronous values, and a rerun of the version 70 step could keep an output capacity_min_coeff left
    over from version 36.
  • flextool-migrate-database, flextool-run --migrate and flextool-update print the reason when a database
    is not migrated
    (a kept pre-migration copy, no room for the copy, or a restore that failed) and exit with code
    1, instead of showing a traceback. flextool-update goes on with the other databases and lists those it could
    not migrate at the end.
  • Running the version 56 step again no longer switches off reserves active in the base alternative. A rerun of
    the step, after an interruption or after the whole step had been applied and a later step failed, set
    is_enabled to no in the base alternative of every reserve, overwriting the yes the first run had written
    for the reserves active there.