Skip to content

Latest commit

 

History

History
78 lines (62 loc) · 4.95 KB

1.2.0.md

File metadata and controls

78 lines (62 loc) · 4.95 KB

Lesscode 1.2.0

Prepared 2026-11-11. The tag v1.2.0 is pushed by the owner; the release workflow then builds, signs and publishes the archives (release signing).

Proposed tag message:

Lesscode 1.2.0: refactoring in checked steps (step plan, pin, check, record, guard); budgets count tokens; working copies follow their source

What is new

  • Refactoring in checked steps. A new command group, step, judges one small change on a Git branch. You declare the step in a file (base commit, files in scope, your test command, a limit for growth) and step check answers by exit code: the change stays in scope, the tests pass before and after, and in no function do more planted faults go unnoticed than before. step plan shows per function where your tests notice faults, step pin says whether a new test pins a fault they missed, step record keeps an append-only ledger, and step guard lets a branch through a pre-push hook or a pipeline only with a verdict for exactly the committed code, tests and declaration. Guide: steps.md; the reasoning: testability.md.
  • budgets counts tokens. Size ceilings are in tokens outside comments and test code: 2,800 per file and 560 per function by default, with 300 syntax units per file as before. Reformatting no longer moves a file over or under its ceiling (R90).
  • deconvolute run can be aimed. functions in the run configuration limits a run to named functions of a file, and --only runs single mutants by number.
  • Stored outcomes depend on the tests. The files under test_paths are part of the identity of a stored outcome, so a changed or added test makes the affected mutants run again without --no-reuse.
  • An agent skill for steps, .agents/skills/refactor-in-steps/. It walks the phases and calls the commands. It cannot accept a step; the exit codes do that.

Fixed

  • A working copy that is used again now follows its source. Releases 1.1.0 and 1.1.1 made each copy under the work directory once and used it as it was in later runs. After you changed a test, or a source file that is not mutated, a second deconvolute run with the same work directory ran the mutants against the old files, with or without --no-reuse. From 1.2.0 the files that changed are copied again, removed files are removed, and the setup commands run again when anything changed. If you keep receipts from a second run on a reused work directory made with 1.1.x, run them again (R93).
  • The top-level help described the binary as it was at 1.0 and named only some of its commands.

What changed for existing users

  • A budgets file written for 1.1.1 or earlier keeps working: an entry with max_physical_lines or max_fn_lines is still held to those ceilings in lines. An entry without them is now held to the token defaults, which are close to the earlier 400 and 80 lines but not the same; check before you rely on it.
  • Outcome stores written by 1.1.x no longer match and are left alone. The first run with 1.2.0 runs every mutant again.
  • step check and step record write three files into your repository, at the paths your declaration names: a verdict, an evidence file and a ledger. Nothing else in the binary writes into a source tree.
  • The evidence file is marked step-evidence-provisional-v0. It records which command ran for which acceptance criterion and how it exited, not which tests ran. Its format will follow openidea-kiro's verification-evidence design and will change.
  • Everything else is as in 1.1.1.

What it was tried on

Four real steps, each from declaration to guard (R92, R93): one in this repository (Rust), one in commons-csv (Java), two in OpenUltraCode (TypeScript). Three were accepted as written. The fourth was refused because the change left six more faults unnoticed in one function, was pinned with nine tests, and was then accepted. That fourth step is also how the working-copy fault above was found.

Not tried on a real step: an override, a release entry on the ledger, a frame specification, a function marked out of reach, and a push of a step branch through the installed hook. These are covered by tests on fixture repositories only.

What is not checked

An accepted step is not shown to behave the same as before. It passed your tests and lost no fault sensitivity in the functions that were run. The list of what no command checks is in steps.md.

For the owner, before tagging

  1. Tag a commit for which CI is green on all five platforms.
  2. Set date-released in CITATION.cff to the day of the tag if it is not 2026-11-11.
  3. git tag -a v1.2.0 -m "<the tag message above>" and git push origin v1.2.0.
  4. Approve the release environment when the workflow asks.
  5. After publication: check signatures, checksums and the attestation, and that Zenodo archived it.