LEX App v2.1.4
Read the upgrade note before deploying this release. It corrects a data problem that
affects PostgreSQL instances which ranv2.0.0rc212or later, and the repair is a command you
run yourself, with a date only you know.
Main changes
- The redesigned interface from
v2.1.3has been rolled back. The appearance returns to thev2.1.2design — the sidebar, theme, settings panel, column formatting, record chips, quick-edit panel and the redesigned calculation log view are no longer present. Two data-correctness fixes from that work are kept: a cleared file field stays cleared, and the fixes below. - Timezones are now handled correctly throughout. The framework has moved to timezone-aware timestamps stored in UTC. In practice: a time you type is interpreted in your deployment's timezone, stored as an unambiguous instant, and shown back to you as the same wall-clock time you entered. The display timezone defaults to
Europe/Berlinand can be set per instance with theLEX_TIME_ZONEenvironment variable.
Bug fixes
- Times entered in a form are stored as the time you meant. On affected instances a datetime typed as 11:00 could be read back as 13:00. Both halves are fixed: the app now records an explicit instant when you submit a form, and the database session runs in the deployment's timezone so values read back as local rather than UTC. Existing incorrect values are not touched automatically — see the upgrade note.
- Spreadsheet exports show the times you see on screen. Spreadsheets have no concept of a timezone, so a time has to be written as a plain wall-clock. Exports previously wrote UTC, so a row showing 11:00 in the grid arrived as 09:00 in the file; on one export path the file failed to generate at all. Exports now render every time in the timezone of the browser that asked for them.
- Point-in-time views land on the right moment. The "as of" anchor was being mangled in transit, so a request for a specific past moment could silently fall back to the present.
Upgrade note
No schema migration. The database columns were already capable of storing timezones, so nothing is altered and no downtime is needed. Set LEX_TIME_ZONE if your deployment is not in Europe/Berlin.
Whether your data needs repair
You are affected if all of these are true:
- your instance runs PostgreSQL, and
- it ran
v2.0.0rc212or later at some point before this release, and - users typed datetimes into forms during that period.
What happened. From v2.0.0rc212 the database connection's timezone changed while the framework was still treating datetimes as having no timezone at all. Times that users typed were quietly re-anchored as they were written. v2.1.1 then began serving datetimes with an explicit timezone marker, which is correct — and which is why the earlier corruption became visible as a shift of an hour or two.
Timestamps the application sets itself — creation and edit times — were never affected. Only datetimes entered by people are at risk, and only those written during the window above.
Repairing it
A command is provided. It is a dry run unless you pass --apply.
lex rebase_incident_datetimes --cutoff 2026-07-02T09:00:00Z--cutoff is required and has deliberately no default, because the window is per instance: an instance became affected on the day it deployed rc212 or later, not on any release date. Use your own deployment timestamp.
Getting --cutoff wrong is not symmetric, so err on the late side:
Consequence | |
|---|---|
Too early | Correct pre-upgrade rows get re-anchored — this corrupts good data |
Too late | Some bad rows are left uncorrected — recoverable, just run it again |
Other options: --until bounds the end of the window and defaults to now, so re-running after this upgrade never re-shifts rows that are already correct. --source-zone is the timezone your users' wall-clock times were meant in, defaulting to Europe/Berlin. --models limits the run to named models; by default it covers all of your models.
Daylight saving is handled per value, using the offset in force on each value's own date rather than one fixed offset. Two hours a year are genuinely ambiguous — the hour skipped in spring and the hour that occurs twice in autumn. Values in those windows are corrected on a best-effort basis and reported separately in the dry run so you can check them by hand.
Read the dry-run output before applying it.
