mirth: second batch of checks; all Python ported to mirth-lab (Rust)

[zmaril]

Stacked on #29.

Second batch of checks (docs/checks.md)

• suggest-diff (check 18): every machine-applicable suggestion applied one at a time must keep the program compiling. Finding 29: lint fixes that break builds.
• diag-check (check 13): no compiler-internal debug output in diagnostics, spans in bounds. Finding 30.
• repro-diff (check 15): outputs depend only on inputs (repeat, other directory, threads, decoy libraries).
• gate-check (check 17): nothing unstable usable from stable. Finding 31: #[rustc_main] on a struct ICEs.
• xlink (check 5): every target builds core/alloc and links a probe with no undefined symbols. Timeouts are now their own result class.
• instr-check (check 21): PGO and coverage instrumentation round trips.
• scale-check (check 10): compile time, memory, frames and future sizes grow about linearly. Finding 32: on long iterator chains the new solver got 10× slower and uses 4× the memory, bisected to nightly-2026-08-04 (#160254 is the only solver PR in that range; not confirmed with a revert build).

The Python is now Rust: crates/mirth-lab

All 36 scripts in rustc/*.py (about 6,200 lines) are now one library plus 32 clap subcommands.

• The library has shared modules for UI test headers, compiling and running, typed JSON diagnostics, the sweep driver (rayon, --recheck/--known/--pause-on-finding), output normalization, Miri, artifact comparison, Cargo messages, coverage sites and the edit mutations.
• mirth-rewrite is split into a library so rewrite-diff can run it in-process.

Each script was compared against its Python version before it was deleted:

• Same findings: the oracle sweeps.
• Same counts and gap lists: callgraph on build-blk; only the order of tied rows differs.
• Same files, or the same lines in a different order: the other coverage tools.
• Same results on the same inputs: the flag tools and fuzz/replay.
• Exceptions:
  • Tools that make random edits use a Rust random generator, so seeds from Python runs don't replay the same edits.
  • flag-fuzz was only smoke-tested.

The shell callers build and call mirth-lab, and the docs name the subcommands. ui-solver-diff was removed because solver-diff covers it.

Python still in the repository: one-off research scripts under docs/ and ur/.

Commits

• flag-walk, flag-fuzz, coverage-flags: --rows must be lo:hi (a malformed value ran every row; Python raised)
• Finding 32: new-solver compile-time regression on long iterator chains, bisected to nightly-2026-08-04 (#160254 the only solver PR in range); facts and the 200-map test
• xlink: a timed-out build is its own result class
• rustc/coverage-flags.py removed (ported to mirth-lab coverage-flags)
• coverage, coverage-compact, coverage-generators, grammar-coverage and ui-coverage scripts removed (ported to mirth-lab); docs name the subcommands
• wip: fuzz, fuzz-replay, replay in mirth-lab (own mutations and cargo modules, to reconcile with port-ui)
• Cargo.lock: mirth-lab's libc dependency
• mirth-lab: coverage, coverage-compact, coverage-generators, grammar-coverage, ui-coverage (validated against the Python scripts); shell callers build and use mirth-lab
• wip: flag tools
• mirth-lab ui-fuzz, and the mutations and Cargo-artifact comparison in the library; ui-solver-diff.py removed (solver-diff covers it)
• instr-check: executable-no-mangle-strip is expected under IR PGO (clang -fprofile-generate fails the same link)
• mirth-lab callgraph: the reachability analysis in Rust (same counts and gap lists as callgraph.py on build-blk; ties ordered differently); coverage-report.sh uses it
• The Python oracle scripts removed (ported to mirth-lab and validated); docs name the mirth-lab subcommands; how to run them in checks.md
• mirth-lab: release-diff, xlink, scale-check (wait4 rusage through libc), abi-diff (typed differences; known and undecided classes labelled by variant); all 14 oracle tools ported
• mirth-lab: repro-diff, gate-check, instr-check (validated: same results as Python, instr-check with the harness artifacts it surfaced fixed); spawn retried on ETXTBSY
• mirth-lab: miri-diff, rewrite-diff (mirth-rewrite in process), suggest-diff (identical to the Python sweep: 7,357 suggestions, 159 tests with findings)
• mirth-lab: the checks in Rust. Core library (uitest, rustc with typed JSON diagnostics and timeouts, driver with the frontier options, artifacts, miri, normalize) and opt-diff, solver-diff, crash-di
• scale-check.py (check 10: growth exponents of compile time, memory, future size and stack frames over size-parameterized programs)
• instr-check.py (check 21: PGO and coverage round trips), xlink.py + xlink-probe (check 5: build-std and link for every target)
• gate-check.py (check 17: unstable attributes and library items from stable code); finding 31
• repro-diff.py (check 15: repeat, path, threads, decoy); nothing new beyond the known parallel-frontend issues
• diag-check.py (check 13: debug output in diagnostics, spans out of bounds, errors without a location, duplicates); finding 30
• suggest-diff.py (check 18: every machine-applicable suggestion applied and compiled); finding 29: lint fixes that break builds

🤖 Generated with Claude Code

https://claude.ai/code/session_01QXiEXbESemwqMLYKaWLDbT