1.10.0: reorder-noise fold, ARXML/A2L consistency check, JSON/SARIF output - #33
Merged
Conversation
No production caller -- only a test's skipUnless used it, and the git button gates on repo_root() being None instead. Inline the availability check into the test so the helper does not linger as dead API.
Embedded Coder routinely re-emits the same independent assignments in a different order (output ports, temporaries). The text reads it as a change, but the block computes identical values -- the last residual codegen churn the text-based rules cannot see past, because it is a reschedule, not a rename, comment or whole-block move. Decide it on meaning, not spelling: when the whole surviving change set is a dependence-preserving permutation of a straight-line block of side-effect-free scalar assignments, it is proven noise (kind 'reorder', ignorable). The scalar- LHS + no-call restriction makes distinct names denote distinct objects, so name-keyed read/write sets are the exact data dependence; two schedules that agree on the order of every dependent pair are linear extensions of the same DAG and compute the same result. Any unsafe line, multiset mismatch or flipped dependent pair leaves every hunk real -- fail-safe, all or nothing.
A model's ARXML (its contract) and its generated C (its behaviour) come from one regenerate and are expected to change together. When one is Modified and the other Identical, each file is individually fine and the inconsistency lives strictly between them -- the one failure a file-by-file view structurally cannot point at, and the everyday sign of a partial or stale regenerate. New consistency.py computes it from verdicts the tool already stands behind, so it invents nothing. Surfaced in the report (a 'Consistency check' section) and the terminal summary. Advisory ONLY: absence of a partner change can be legitimate (a hand-written file, an ARXML-only edit), so it never folds a file, moves a count or changes the exit code. A2L is not paired -- a recal touches no code. Only the C<->ARXML pair, only when both are present in the compare.
The HTML report is for a human and the exit code is for a gate; neither lets a build server read WHAT changed. New serialize.py writes the scan as data under two shapes: --json is the whole record (per-file verdict, hunks, renames, AUTOSAR extras, summary, consistency advisories) under a versioned schema, so a consumer pins 'schema' and is insulated from an internal refactor; --sarif is a SARIF 2.1.0 log of only the files needing action (modified/added/deleted/error), for GitHub or Azure DevOps code scanning to annotate inline. Both additive to the HTML report and independently usable. The JSON carries the same exit_code the process returns (one _exit_code seam), so file and $? agree. A failed machine-output write exits 2, same as a missing report -- a pipeline that asked for the file must not proceed as if it got one.
…tures fixtures/demo is a before/after pair a human runs to see the three new features at once, and test_demo.py asserts each claim so the demo cannot drift: - SpeedCtrl -- three independent gains re-emitted in a different order fold to Unimportant (reorder), proving the values are unchanged. - TorqueLimiter -- C regenerated (gain 1.25 -> 1.45) while the ARXML stayed identical, so the consistency check flags it; nothing else does. - PedalMap -- C and ARXML changed together (a new Scaled port) so no flag, and a new A2L characteristic is summarised without ever triggering one. README.md gives the run command and what to look for; the SARIF omits the Unimportant SpeedCtrl.c, so only the four actionable files are findings.
The direction was backwards. Flagging 'C changed but ARXML did not' fired on the ordinary case -- an internal logic or gain edit touches no interface, so the code legitimately moves alone. The real desync is the reverse: an ARXML (interface) or A2L (calibration) change with the generated C left identical -- a new port or characteristic the code never picked up, the sign of a stale regenerate. Now flag a model only when a surface (ARXML or A2L) really changed while its C did not; a code-only change is never flagged. A2L is included, and both surfaces collapse into one message. Moved the report section below the AUTOSAR changes and dropped the long caption. Demo gains a StaleGen model for the flag and keeps TorqueLimiter as the code-only case that must stay quiet.
The AUTOSAR rollup showed a generic '+1 A2L', which hid whether a characteristic or a measurement moved. Split it by kind (+1 Characteristic / +1 Measurement) through one view_model seam so the report chips and the viewer header spell it the same way. Ports, interfaces, runnables and events already carried their noun. In the viewer the chips now ride on the file header line, right after the name and before 'Change k of N', instead of a second 'AUTOSAR / A2L:' line below -- one line, no prefix, less vertical space spent restating the category.
One compare instead of three: fixtures/demo now also carries copies of the
noise-rule fixtures (rules/ -- comment, uuid, timestamp, rename, added,
deleted) and the model-grouping fixture (models/Ctrl) beside the four feature
models, so a reviewer sees every ignorable kind, an add/delete, and more than
one model in the Overview table in the SAME run that shows the reorder fold,
the consistency check and the A2L-by-kind chips.
Copies, not moves: tests/fixtures/{old,new,model_old,model_new} stay put
since other tests pin exact counts and paths against them; duplicating a few
small fixture files costs nothing next to that.
test_demo.py gains coverage for the folded-in paths (every ignorable kind
present, added/deleted present, the four models still their own Overview
rows) and the SARIF assertion moves from an exact set to inclusion checks now
that the merged tree has more actionable files than the four models alone.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
reorderhunk kind) — Embedded Coder rescheduling independent assignments no longer reads as a change.+1 Characteristic/+1 Measurement) instead of a generic+1 A2L, in both the report and the viewer.--jsonand--sarifoutput for pipelines to read the scan as data instead of screen-scraping the HTML report.gitsource.git_available().tests/fixtures/demo) that exercises every feature above plus every noise rule in a single compare, with tests locking its claims.Test plan
python -m unittest discover -s tests— 696 tests passpython -m ruff check .— cleanpython packaging/release_check.py 1.10.0— readytests/fixtures/demoend-to-end (report, JSON, SARIF) and eyeballed the output