Changelog

Copy Markdown View Source

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

[0.3.0] - 2026-09-01

Added

  • CI Recipes guide on hexdocs: GitLab CI and CircleCI persistence recipes (cache semantics per provider, parallel-job lineages), the artifact download-and-merge flow for inspecting history locally, and keeping a cached history bounded with the retention flags. The README keeps the GitHub Actions recipe and links the guide.

  • History Schema guide on hexdocs: the JSONL line format (schema v1) documented as a public contract — envelope, every field with type and nullability, failure signatures, suite lines, consumer and producer rules, and the versioning policy (additive within v1, breaking changes bump the schema number).

  • mix temper.clean --older-than DAYS and --keep-shas N prune history instead of deleting it: files are rewritten in place keeping only lines inside the retention window, emptied files are removed, and corrupt lines are dropped with a count. A criterion only drops lines it can positively match — lines without a SHA (suite summaries, null-SHA records) are untouched by --keep-shas and age out via --older-than. Keeps a history persisted across CI runs bounded.

  • mix temper.merge --output FILE [GLOBS...] merges history files into one deduplicated file — the aggregation step for CI artifact flows (one .temper/ artifact per job, download, merge, report) and for compacting overlapping cache restores. Input globs match into dot-directories (artifact downloads keep history inside .temper/). Dedupe is byte-exact on whole lines; lines this version does not interpret (suite summaries, future schema versions) are preserved verbatim, and only corrupt lines are skipped, with counts reported.

Changed

  • mix temper.doctor dates a partial SHA gap: the recorded commit SHAs warning now reports the newest null-SHA record's timestamp and whether the gap is still occurring or every record since carries a SHA — legacy noise and an active recording-blind path are distinguishable at a glance.

[0.2.1] - 2026-08-25

Added

  • {partition} placeholder in history_path: the formatter expands it with the MIX_TEST_PARTITION value ("0" when unset) and mix temper.report / mix temper.clean widen it to * — custom paths are partition-safe by construction instead of by recipe.
  • README FAQ answering the first real-world questions: whether Temper re-runs tests, cross-commit failures, what "Suspects" means, and why a first run reports nothing.
  • mix temper.doctor — setup preflight diagnosing the silent failure modes: formatter not registered for the test env, history_path invisible outside the test env, umbrella child apps whose app-dir runs record nothing, empty history, and records without a usable commit SHA. Exits non-zero on problems so CI can gate setup on it.

[0.2.0] - 2026-08-23

Added

  • TEMPER_SHA, TEMPER_DIRTY and TEMPER_BRANCH environment variables override git context detection — for containers and other environments where git cannot answer. A non-empty TEMPER_SHA activates manual mode and takes priority over CI variables and local git.
  • mix temper.report --by-app groups findings by umbrella child app (derived from file paths), with per-app counts and involved files. JSON findings always carry the derived "app" field.

Changed

  • Recommended dependency line is now only: [:dev, :test], runtime: false, keeping mix temper.report available in the default env.
  • Documented umbrella setup: root-only dependency, formatter registration via config :ex_unit, and a partition-safe shared history path — verified end-to-end on Elixir 1.15 through 1.20.

[0.1.0] - 2026-08-22

Added

  • Temper.Formatter — ExUnit formatter recording every test outcome to .temper/history-{partition}.jsonl, with run context (git SHA, dirty flag, CI provider, seed, partition) and failure signatures. Crash-safe: on any internal error it warns once and goes inert.
  • mix temper.report — flaky tests (divergent outcomes on the same clean git SHA) sorted by flake rate, plus lower-confidence suspects from dirty-tree runs. Options: --json, --min-runs, --history.
  • mix temper.clean — deletes recorded history.
  • History schema v1: append-only JSONL, one denormalized line per test outcome, partition-safe for parallel CI jobs.