Changes

The two most recent releases in full, then the version history. The distribution's Changes file is the authoritative record.

0.47 — 7 October 2026

Enhancements

• extract-schemas --function NAME (-F) — extracts everything, then deletes every schema except NAME.yml and limits the fuzz run and the summary table to that one routine. Built for isolating a single failure in the self-fuzz suite; if the name is not found, the available routine names are printed and the command exits 1.
• A clearer self-fuzz summary. The end-of-run table now lists every processed routine, not only the ones that fuzzed successfully, with the module name, the routine and either a test count or a plain reason for skipping: internal helper, excluded from the fuzz list, instance method needing an object, mandatory object parameter without a can hint, or FAILED.

Documentation

• Expanded the MUTANT_SKIP examples.

Bug fixes

• $pkg was extracted as a parameter. The invocant filter only knew $self and $class, so $pkg in import(), $proto in older constructors and $klass were emitted as position-0 string parameters — giving modules that use use Module key => value a spurious pkg: {type: string} entry. All extraction loops, the POD/code agreement check and the inline body guard now share the full self|class|pkg|proto|klass set. Three regression tests cover it.

0.46 — 18 August 2026

Dashboard

• CPD drill-down pages. A per-file page shows the source with duplicated lines highlighted, and ◆ markers linking to the other copy of each block. Dup% badges link to it, and the mutation page's legend does too.
• One visual style for per-file pages. Duplication pages now use the same header, footer, CSS variables and dark-mode toggle as mutation pages.
• Sortable mutation and duplication tables. Any column header sorts; arrows show the active column and direction. Sorting detects numbers versus text, honours data-sort-type="date", and respects a data-value attribute where the displayed text is not the sort key.
• Fixed a column-alignment bug. The TER column appeared in data rows but only in the header when LCSAJ data existed, shifting the Dup% index. TER is now always in the header.
• Drag to reorder columns, with the active sort column following its column to its new position.
• Persistent sort and column order for all four tables, stored in localStorage and restored on load. Columns are identified by their original index, so a saved order survives a new column being added.

Tests

• t/domain.t — equivalence partitioning and boundary value analysis across six modules: confidence thresholds at 19/20/39/40, the signal-to-bucket mappings and their alphabetical tie-breaks, numeric min/max boundaries where the code uses strict comparisons, the exists-versus-defined distinction in required attributes, cyclomatic and early-return thresholds, nesting depth, purity classes and mutation-field deduplication.
• t/data-flow.t — define-use chains: corpus staleness after minimisation, evidence isolation between objects, weight => 0 stored as zero rather than defaulted to one, YAML export/import round trips, file handles closed inside export(), and string literals not mistaken for IO.
• Plus the path, transaction and path-coverage suites described in the repository's CLAUDE.md.

Earlier releases

Full history: MetaCPAN's changelog, or the Changes file in the repository.

Upgrading

cpanm App::Test::Generator

Schemas are compatible across releases: a schema that works with 0.47 has worked since the transform syntax settled. Generated harnesses are regenerated by tooling, so a stale one is replaced rather than migrated. If you keep generated .t files in version control, regenerate them after upgrading — an older harness may not know about a type or a config key that the new version accepts.

Back to the overview · Quick start · Schema reference