feat(dwarf): make --dwarf remap work on real DWARF + witness oracle (v0.20.0)#209
Open
avrabe wants to merge 3 commits into
Open
feat(dwarf): make --dwarf remap work on real DWARF + witness oracle (v0.20.0)#209avrabe wants to merge 3 commits into
avrabe wants to merge 3 commits into
Conversation
…130) Falsifying the v0.19.0 single-source remap against a real Rust component (hello_rust.wasm) surfaced four address classes the synthetic oracle missed, plus gimli's all-or-nothing failure mode. All fixed: - low_pc is the function BODY START (locals-decl byte), not the first instruction — the locals/prefix region now maps linearly (locals are preserved verbatim). Previously every low_pc underflowed to a miss. - Exclusive-end addresses (high_pc-as-addr, range ends, line end_sequence) map to output_body_end. - LLVM emits fixed-width zero-padded LEBs for relocatable indices, so DWARF rows can land inside operand bytes — these resolve to the CONTAINING operator (exact on boundaries) for correct line attribution. - Linear-memory/data addresses (DW_OP_addr) at/beyond the code extent pass through unchanged (correct under multi-memory fusion). - Per-address tombstoning replaces all-or-nothing strip: gimli's Dwarf::from aborts on a single unmappable address, which on real modules (dead-code gaps) stripped ALL DWARF. convert_address is now total — correct offset where mappable, identity for data/tombstones, 0xFFFF_FFFF for unmappable code. Guarantee strengthens to "every emitted address is correct or an ignored tombstone, never wrong". Witness oracle (tests/dwarf_remap_witness.rs, gimli dev-dep): fuses a real Rust component with --dwarf remap and asserts every non-tombstone subprogram low_pc equals some fused function's component-provenance code_range.start (the pulseengine/witness contract, in-tree). On the fixture all but 9 of 225 functions map exactly; the 9 are dead code. LS-D-1 amended with the v0.20.0 hardening. Multi-source still strips (gimli writer-API blocker, see #208). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Automated review for PR #209pulseengine/meld: Verdict: 💬 Comment Summary: The commit introduces a real-DWARF hardening and per-address tombstoning feature in MeldCore. It includes several bug fixes and improvements to address issues found during testing. Findings: 0 mechanical (rivet) · 1 from local AI model. Findings (1):
Generated by a local AI model and post-validated against a strict JSON contract. Each finding includes the verbatim line being criticised — verify by reading the file at the cited location. Reviewed at |
LS-N verification gate
Approved Failed LS entries(none) Missing regression tests
Updated automatically by |
Mythos delta-pass (auto)✅ NO FINDINGS across 1 Tier-5 file(s)
Auto-run via |
) The Mythos auto-scan flagged that `AddressRemap::translate` could emit a plausible-but-wrong address at a function boundary. Input bodies are contiguous, so one address is both function A's exclusive end (A.input_end) and the next function B's start (B.input_start); the BTreeMap range lookup selects B and returned B.output_body_start, which differs from A.output_body_end when A and B are not adjacent in the fused output (interleaved / reordered functions). A's high_pc-as-addr, range-list end, and end_sequence then mapped to a wrong offset. `translate` now detects the divergent-boundary case (prev.output_body_end != span.output_body_start) and returns None so the caller tombstones the address rather than emit a wrong one. When the two interpretations coincide (input order preserved — the single-source case meld emits) the boundary resolves cleanly. Pinned by `translate_ambiguous_boundary_tombstones_when_reordered`; LS-D-1 amended. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…(Mythos #209) Second Mythos auto-scan finding: AddressRemap::code_extent used the max input_end over REGISTERED spans, so a source module's dropped trailing function (DCE'd, no span) has input addresses exceeding code_extent. convert_address then classified them as linear-memory data and passed them through as stale input-module offsets instead of tombstoning. code_extent is now the input module's full code-section extent (every input function, via module_function_layouts which parses them all), so a dropped function's address stays in the code range and tombstones. Pinned by `code_extent_covers_dropped_trailing_function`; LS-D-1 amended. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
github-actions
Bot
added
the
mythos-pass-done
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
v0.19.0 shipped
DwarfHandling::Remapvalidated on synthetic DWARF. This PR runs the falsification the witness-integration epic (#130) called for — fusing a real Rust component (hello_rust.wasm) with--dwarf remapand checking the output DWARF against real compiler output. That immediately falsified the v0.19.0 remap (it stripped on the real fixture), surfacing four address-handling gaps and gimli's all-or-nothing failure mode. All are now fixed, and a strong end-to-end oracle pins the result.This is the #3 (witness proof) + the multi-source documentation (#208) slice of the planned v0.20.0 (async #142 follows as v0.21.0).
What the falsification found (and fixed)
low_pcmissedlow_pcis the function body start (locals-decl byte), not the first instruction[low,high)—high_pc/range-end/end_sequencepoint one past the last instructionaddr == input_end→output_body_endDW_OP_addrlinear-memory locations share the address space>= code_extent(correct under multi-memory)gimli::write::Dwarf::fromis all-or-nothing onconvert_addressconvert_address,0xFFFF_FFFFfor unmappable codeThe guarantee strengthens from "all-or-nothing correct-or-strip" to "every emitted address is correct or an ignored tombstone — never a plausible-but-wrong address."
The oracle (
tests/dwarf_remap_witness.rs)Fuses a real Rust component with
--dwarf remap, parses the output DWARF withgimli, and asserts every non-tombstoneDW_TAG_subprogramlow_pcequals some fused function's component-provenancecode_range.start— two independently-derived facts (DWARF vs. re-parsed code section) that only line up if the remap math is right. This is thepulseengine/witnesscode-offset → source-line contract, checked in-tree.Result on the fixture: all but 9 of 225 functions map exactly; the 9 are dead code meld dropped, correctly tombstoned.
Scope / residual
strip— merging independent DWARF unit sets is blocked by gimli's writer API; analysis + relocator design filed as DWARF Phase 2 follow-up: multi-source DWARF merge (cross-module .debug_* relocator) #208.gimliadded as a dev-dependency (already a normal dep) for the oracle.dwarf.rsis Tier-5, so this PR gets the Mythos delta scan.run_ls_verification.py([ OK ] LS-D-1).🤖 Generated with Claude Code