Repository navigation
chore(abi): move NeSyAssistTesting to a non-proof fixture (A5, 3/3) - #513
hyperpolymath wants to merge 6 commits into
Conversation
Owner ruling on open question A5: delete the stale copies, salvage the rest.
Each file sat at the src/abi/ root outside echidnaabi.ipkg, so nothing
type-checked it. Each one was diffed against the packaged module it shadows:
BojForeign.idr = Boj/Foreign.idr (body identical)
OverlayForeign.idr = Overlay/Foreign.idr (body identical)
TentaclesForeign.idr = Tentacles/Foreign.idr (body identical)
TypeLLForeign.idr = TypeLL/Foreign.idr (body identical)
Types.idr = EchidnaABI/Types.idr (body identical)
Layout.idr = EchidnaABI/Layout.idr (body identical)
Foreign.idr older EchidnaABI/Foreign.idr: stale module name
Echidna.ABI.Foreign, no Result type, AnyPtr callback
Comments that named a shadow now name the packaged module: the four
generated/abi headers (no generator for them exists in the tree), four Zig
FFI sources, the L1 Cap'n Proto handover note and issue echidna-04. Nothing
else imported these files. src/rust/corpus/profile.rs matches only on
fixture paths under a temporary root, and reports/ are historical
snapshots, so neither changes. PROOF-STATUS R2 is updated.
CoprocessorForeign and NeSyAssistTesting follow in the next two PRs of this
stack.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019a95kk88TZ1iegsBK3kox3
src/abi/CoprocessorForeign.idr declared `module Coprocessor.Foreign`, so its module name did not match its path and the package could not list it. Nothing type-checked it. Owner ruling on A5: salvage it into the package if it type-checks. - git mv to src/abi/Coprocessor/Foreign.idr (contents unchanged; the module declares %default total and uses no believe_me, assert_total or postulate). - echidnaabi.ipkg lists Coprocessor.Foreign, so idris2-abi-ci type-checks it. This PR's ABI job is the type-check. If it fails, the module comes out again rather than being forced through. - w0.sh check 07b now looks for Coprocessor/Foreign.idr. At the old path the check would skip the moved file and pass vacuously. The selftest is updated (68 cases, 0 failed). - The SPARK spec comments and PROOF-STATUS R2 name the new path. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019a95kk88TZ1iegsBK3kox3
src/abi/NeSyAssistTesting.idr does not parse. It states an empirical property of the neural ranker, which no proof can establish. Owner ruling on A5: move it to tests/ as a non-proof fixture. - git mv to tests/fixtures/non-proof/, with a README that says nothing there is a proof or is type-checked. - After this stack, every .idr file under src/abi/ is listed in echidnaabi.ipkg (checked by script). PROOF-STATUS R2 and the wiki Home table say so; the wiki row no longer names CapnSchemas, which #472 packaged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019a95kk88TZ1iegsBK3kox3
) ## Why `src/abi/CoprocessorForeign.idr` declared `module Coprocessor.Foreign`. Its module name did not match its path, so the package could not list it, and nothing type-checked it. Owner ruling on A5: put it in the package if it type-checks. ## What - `git mv` to `src/abi/Coprocessor/Foreign.idr`. The contents are unchanged. The module declares `%default total` and uses no `believe_me`, `assert_total` or `postulate`. - `echidnaabi.ipkg` lists `Coprocessor.Foreign`. **This PR's ABI type-check job is the test.** If it fails, the module comes out again; it is not forced through. - `scripts/gates/w0.sh` check 07b now looks for `Coprocessor/Foreign.idr`. At the old path the check would skip the moved file and pass vacuously. The self-test is updated: `bash scripts/gates/w0.selftest.sh` gives 68 cases, 0 failed. - The SPARK spec comments (`coprocessor_safety.ads`) and PROOF-STATUS R2 name the new path. `idris2` is not available in this session, so the type-check is left to CI. Stacked on the A5 PR 1/3 branch (`claude/new-session-th4e9w-a5-shadows`). --- _Generated by [Claude Code](https://claude.ai/code/session_019a95kk88TZ1iegsBK3kox3)_ Co-authored-by: Claude <noreply@anthropic.com>
…ession-th4e9w-a5-nesy Signed-off-by: Jonathan D.A. Jewell <6759885+hyperpolymath@users.noreply.github.com>
|
No actionable comments were generated in the recent review. Summary
Merge Risk:
|
…4e9w-a5-nesy # Conflicts: # docs/PROOF-STATUS.adoc
|
This security review couldn’t start because the paying account has exhausted its Codex allowance and has insufficient available credits. The payer can check their balance and add credits in the Codex usage dashboard, then try again. If you do not manage the paying account, contact this repository's admins. |
|
…the Agda receipt (#520) ## Summary This follows up on #509 and #510, both already merged. An independent review of that batch raised two gaps, and this PR closes them. ## Changes - **`dogfood-proofs-ci.yml` (verif-idris gate from #509).** The gate only checked that `%default total` is present. A later `partial` or `covering` annotation, or a `%default partial|covering`, switches the totality check off again, and the package still builds. Idris 2 defaults to covering, and under covering `bad : Void; bad = bad` type-checks. The gate now rejects both forms. - **`idris2-abi-ci.yml`.** A new step applies the same ban to `src/abi/`. It runs on every event, because it is cheap. - **`scripts/ci/check-agda.sh` (from #510).** `git rev-parse HEAD` ran inside `echo`'s arguments, where a failure escapes `set -e`. The receipt would then name an empty commit and the script would still pass. The commit is now read on its own line. **Not in this PR:** requiring `%default total` in every `src/abi/` module. That waits for #513, which moves the one module that lacks it (`NeSyAssistTesting.idr`) out of `src/abi/`. ## RSR Quality Checklist ### Required - [x] Tests pass: both new gates were run against the current tree and against a broken copy (see Testing). - [x] Code is formatted: no Rust changed. - [ ] Linter is clean: actionlint was not available in this session. The repo's Workflow Security Linter runs on this PR. - [x] No banned language patterns - [x] No `unsafe` blocks without `// SAFETY:` comments: not applicable. - [x] No banned functions - [x] SPDX license headers present: no new files. - [x] No secrets, credentials, or `.env` files included ## Testing - **verif-idris step:** extracted with `yq` and run on the current tree, exit 0. On a scratch corpus where `B.idr` has a `covering` line, it prints `::error file=…B.idr::module declares a partial or covering definition or default` and exits 1. - **src/abi step:** run on the current tree, exit 0. On a scratch `src/abi/X.idr` with a `partial` line, it prints the error and exits 1. - **`check-agda.sh`:** `bash -n` passes. The changed line was not exercised in this session: Agda is not installed here. --- _Generated by [Claude Code](https://claude.ai/code/session_019a95kk88TZ1iegsBK3kox3)_ --------- Co-authored-by: Claude <noreply@anthropic.com>



Why
src/abi/NeSyAssistTesting.idrdoes not parse. It also states an empirical property of the neural ranker, which a proof cannot establish. Owner ruling on A5: move it totests/as a non-proof fixture.What
git mvtotests/fixtures/non-proof/NeSyAssistTesting.idr.tests/fixtures/non-proof/README.adoc(CC-BY-SA-4.0). It says the file is not a proof, is not type-checked, and is cited by no row..idrundersrc/abi/is listed inechidnaabi.ipkg. I checked this with a script overgit ls-files. PROOF-STATUS R2 and the wiki Home table now say so. The wiki row no longer namesCapnSchemas, which fix(abi): CapnSchemas proved its claims about implicit variables; qualify, gate it in the ipkg #472 packaged.Stacked on the A5 PR 2/3 branch.
Generated by Claude Code