LEX App v2.1.1
Main changes
- Timestamps no longer display shifted by your timezone. On some deployment targets every timestamp in the app — edit times, creation times, history change times, calculation log times — was displayed shifted by the viewer's UTC offset, so a Berlin user editing at 13:43 saw "11:43". The stored values were correct but were served without any timezone marker, so browsers read them as local time. Datetimes now always carry an explicit marker and render in the viewer's correct local time. (#645)
Section headings in the calculation log. A calculation can now group its output into titled sections — a table of contents for a run — by passing a plain string to
model_logging_context, with no backing record required:with model_logging_context("Data preparation"): LexLogger().add_text("Loading input files…").log() with model_logging_context("Validation"): LexLogger().add_list(["schema ok", "types ok"]).log()Headings nest freely inside each other and inside model contexts, appear as titled nodes in the execution tree, and a section that never logs anything is left out automatically. Log streaming, caching and audit behaviour are unchanged — headings only shape the tree. (#644)
Optimizations
- Faster showcase test runs. The business-view showcase suite now fans out across a CI matrix instead of running end to end, cutting wall-clock time substantially. (#643)
Internal
- Test plan restructured. The test plan is now sharded into per-cluster directories with machine-readable allocation files and a generated dashboard, with CI gates enforcing consistency on every pull request. (#642)
- New regression gates covering timezone-unambiguous serialization, calculation-log heading behaviour, and heading-transparent root-calculation detection.
No frontend changes in this release. The interface is unchanged from v2.0.0rc221.
Upgrade note: run database migrations on upgrade. This release adds one new nullable column for the calculation-log sections — instant, with no data rewrite. No configuration changes are needed.
Correction (added retrospectively). This note originally said that no data migration was needed because "stored values were always correct; only the display was wrong". That held for timestamps the application stamps itself, but not for datetimes typed in by users on instances running v2.0.0rc212 or later: those had been written against a mismatched database session timezone since that release, so making them display honestly also made the earlier corruption visible. v2.1.4 addresses both the cause and the affected data — see its upgrade note if you are running a PostgreSQL instance that passed through rc212.
