Skip to content

fix(hypatia): triage all 15 findings — 8 at source, 7 baselined (closes the red gate) - #146

Merged
hyperpolymath merged 1 commit into
mainfrom
fix/hypatia-baseline-triage
Jul 27, 2026
Merged

fix(hypatia): triage all 15 findings — 8 at source, 7 baselined (closes the red gate)#146
hyperpolymath merged 1 commit into
mainfrom
fix/hypatia-baseline-triage

Conversation

@hyperpolymath

Copy link
Copy Markdown
Owner

Closes the last red gate on main (governance / Validate Hypatia Baseline, failing since ≥07-22 when the scanner started actually working).

Fixed at source (8):

  • The critical.hypatia-ignore:86 "Generic API key": self-referential FP; the comment explaining the API-key exemption itself contained the KEY=\"...\" assignment shape. Reworded.
  • my-hir guarded unwrap → match; my-lint 2× genuine empty-name panics → is_some_and; my-llvm SAFETY-comment adjacency.
  • ENSAID abi-verify workflow: src/abi/*.idr (nonexistent) → real Idris tree via ipkg build (per-file --check is a known fake gate).
  • Roadmap doc Planned: src/{hir,mir,runtime} → the shipped crates.

Baselined (7 entries, expire 2026-10-27, tracked in #145): parser self.expect() name-collision FP (157×), my-fmt infallible Write-to-String unwraps (26×), my-llvm internal-invariant unwraps (3×), my-mir/string.rs deliberate invariant expects, measure_depth example, modules.md illustrative user-project paths.

Verified: ran the reusable's own apply-baseline.sh locally against the post-fix finding set → 7 suppressed, 0 kept, exit 0. cargo test -p my-hir -p my-lint green.

Hygiene (owner-approved): .gitignore secrets entries salvaged from the 07-26 sweep stash; mise.toml boilerplate → pinned rust 1.97.0.

🤖 Generated with Claude Code

Source fixes:
- .hypatia-ignore: reword the API-key comment so the secret matcher's
  assignment-shape no longer fires on its own exemption text (the critical)
- my-hir: replace guarded unwrap with a match on into_iter().next()
- my-lint: is_some_and guards kill 2 genuine empty-name panics
- my-llvm: SAFETY marker directly adjacent to the unsafe block (the
  16-line invariant argument sat too far above for the adjacency check)
- ENSAID_CONFIG.a2ml: abi-verify workflow now targets the real Idris tree
  via ipkg --build (per-file --check is a fake gate, exit 0 on missing module)
- docs/wiki/roadmap/compiler.md: 'Planned: src/{hir,mir,runtime}' paths
  now point at the shipped crates

Baseline (.hypatia-baseline.json, 7 entries, expires 2026-10-27, #145):
parser.rs self.expect() name-collision FP; my-fmt infallible Write-to-String
unwraps; my-llvm internal-invariant unwraps; my-mir/string.rs deliberate
invariant expects; measure_depth example; modules.md illustrative paths.
Validated against apply-baseline.sh locally: 7 suppressed, 0 kept, exit 0.

Hygiene (owner-approved salvage from the 07-26 sweep stash):
- .gitignore: secrets/env entries only (.env, *.pem, *.key, .venv/)
- mise.toml: replace 30-tool all-'latest' boilerplate with pinned rust 1.97.0

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
@hyperpolymath
hyperpolymath merged commit 1ff3bf5 into main Jul 27, 2026
1 check passed
@hyperpolymath
hyperpolymath deleted the fix/hypatia-baseline-triage branch July 27, 2026 18:12
hyperpolymath added a commit that referenced this pull request Jul 27, 2026
#147)

Follow-up to #146 — the one finding that survived. hypatia's
`unsafe_block` rule is a bare `unsafe\s*\{` regex
(`lib/rules/code_safety.ex`): it flags every unsafe block regardless of
documentation, so no SAFETY comment can ever satisfy it (the adjacency
fix in #146 was a no-op against a rule that never looks). The single
unsafe block in my-llvm is mandatory (inkwell declares `build_gep`
unsafe) and fully documented, so this uses the same scoped
`.hypatia-ignore` mechanism as the existing `coq_axiom`/`transmute`
keyword-match exemptions — a permanent fact should not be expiring
baseline debt. Upstream rule fix is a candidate for hypatia itself.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
hyperpolymath added a commit that referenced this pull request Jul 29, 2026
…al (#148)

Closes out the my-lang session by recording current state in-repo (not
only in dev-notes) and fixing the documentation defects found while
doing it.

## The Justfile was parse-dead
`README.md` documented `just build` / `just test` as the quickstart.
Neither worked: a `//` C-style comment on line 2 made `just` reject the
**entire file** (`error: unknown start of token '.'`), so no recipe was
reachable — and every recipe was an `@echo` stub regardless. `just
verify` checked `.machine_read/` (a path that has never existed) and
swallowed every failure with `||`, so it always reported success.

Now: real cargo recipes, `-all` variants for the LLVM-dependent crate,
and a `verify` that exits non-zero. **Verified by running them** — `just
build` exits 0; `just test` reports **221 passing, 0 failing**.

*(Estate note: 143 of 751 Justfiles across the estate are parse-dead by
the same class of defect — 76× unindented shell blocks, 35× `//`
comments, 32× `Justfile`/`justfile` case collisions. Out of scope here;
recorded in dev-notes.)*

## For machines — `.machine_readable/6a2/STATE.a2ml`
- version `0.1.0` → `0.2.0` (was drifting from `Cargo.toml`)
- new `[security-posture]`: distinguishes **expiring baseline debt** (7
entries → 2026-10-27, #145) from **permanent scoped exemptions** (3,
each a rule-precision problem), plus upstream rule-fix candidates
- new `[toolchain]`; `known-local-limitation` records that `my-llvm`
needs system LLVM 21, so a bare workspace build failure is not misread
as a regression
- dialect model corrected: **solo ⊂ duet ⊂ ensemble**, with `me` an
outside projector — not four co-equal surfaces
- `superseded-branch` records the evidence that
`feat/stdlib-fs-env-format` is safe to delete

## For humans — README + CHANGELOG
Quickstart matches reality (incl. the my-llvm caveat); dialect
containment; measured test count (137+ → 221); governance/security
posture; `contractiles/` path corrected (it is under
`.machine_readable/`, not the repo root). CHANGELOG gains
Security/Fixed/Changed entries for #143, #146, #147.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
hyperpolymath added a commit that referenced this pull request Aug 12, 2026
… path real

The README documented `just build` / `just test` as the quickstart, but the
Justfile was PARSE-DEAD — a `//` C-style comment on line 2 made `just` reject
the whole file (`error: unknown start of token '.'`), so no recipe was
reachable; and every recipe was an `@echo` stub anyway. `just verify` checked
a `.machine_read/` path that has never existed and swallowed failures with
`||`, so it always claimed success.

Justfile: real cargo recipes (build/test/lint/fmt/check/proofs/pipeline),
split `-all` variants for the LLVM-dependent crate, and a `verify` that exits
non-zero when a path is missing. Parse- and run-verified: `just build` exits 0,
`just test` reports 221 passing / 0 failing.

STATE.a2ml (machines): version 0.1.0 -> 0.2.0 to match Cargo.toml; current
session + maintenance status; a new [security-posture] section distinguishing
EXPIRING baseline debt from PERMANENT scoped exemptions, with the three active
exemptions and their rule-precision rationale, plus upstream rule-fix
candidates; a [toolchain] section; the my-llvm/LLVM-21 local limitation
recorded so a bare workspace build failure is not misread as a regression; and
the dialect model corrected from four co-equal surfaces to solo ⊂ duet ⊂
ensemble with me as an outside projector.

README (humans): quickstart now matches reality incl. the my-llvm caveat;
dialect containment; measured test count (137+ -> 221); governance/security
posture; `.machine_readable/` and `_exploratory/` documented, and the
`contractiles/` path corrected (it is not at the repo root).

CHANGELOG: Security / Fixed / Changed entries for #143, #146, #147.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant