[leeguooooo]

Closes #6792.

ELF images with PT_DYNAMIC but no SHT_DYNSYM section incorrectly reported a complete, empty export inventory. The shared ELF directory reader now owns the dynamic-segment observation; CLI/MCP report unsupported with a reason. Existing symbol-table results remain intact. Read-only format correction; no new execution authority or public schema.

Validation on the current main integration:

• npm run check, test discovery, and 166 selected Linux tests pass, including actual CLI/stdio MCP cases and PE/IL2CPP/GPU/Rust reader consumers. Original author validation: 113 tests on macOS, including regressions that failed before the fix; test discovery passed.
• Real Debian libc (1,926,232 bytes) and libLLVM 14 (109,967,296 bytes), with only section-header-count/name-index bytes changed, reproduce the old false absence in both public adapters. The reviewed implementation returns unsupported in both adapters. Independent Python maps unchanged PT_DYNAMIC declarations and selected symbol bytes; original readelf agrees. The reduced files are not loaded or executed; readelf -D itself cannot enumerate these reduced layouts, and that failed oracle attempt is retained.
• Unmodified libc still returns all 3,025 exports, byte-identical complete Evidence and the same Evidence ID as the baseline. Resource-guarded public probes completed without stops or payload leftovers. No speed, memory or model-token improvement is claimed for this coverage fix.

Codex review completed with no findings. CI failed: all 61 observed failures (including 47 omitted from the aggregate report by canceled shards) reproduce with identical complete failure diagnostics on the exact CI base and merge trees. Baseline-only bypass is explicitly authorized for this maintenance session; canceled/skipped coverage remains unverified. See the final CI attribution comment. The original author's readiness declaration below is preserved.

Author readiness

Human owner: @leeguooooo

• I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work.
• I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps.
• I declare this PR ready for initial maintainer review.

[gensokyo-dev]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-23T19:02:48.971066Z	8dfa6a6	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable issue was found in the changed code.

Summary

The shared ELF reader now records whether the file declares PT_DYNAMIC. When that segment exists without SHT_DYNSYM, export inspection returns unsupported instead of claiming a complete, empty result.

• ELF export checks mark unlisted dynamic symbols as unsupported.

_{Reviews (1) · Last reviewed commit: "Merge remote-tracking branch 'origin/mai..." · Reviewed by Greptile}

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Delightful!

Reviewed commit: 8dfa6a6f88

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[gensokyo-dev]

Latest head 8dfa6a6f880837baac8d61cac5f19e08c718d83d: Codex review completed with no findings. The Linux validation now includes 166 selected tests (47 exports/PE/IL2CPP/public-adapter cases plus 119 GPU/Rust consumer cases), npm run check, and discovery of all 2,177 test files exactly once. All 220 actual SDK tool advertisements are byte-identical to the unchanged-main control.

The Source tests job in CI 40007458412 reports 11 failures: five Map-operation cases, one shared callable-alias case, four Frida cleanup cases, and one bundle-resource schema case. I reproduced those exact 11 cases against CI base 34563fcc9bd92af30668820dcac5635e27181094 and the exact CI merge tree 2ecbbb993dc673cc18df583fb47fd09e0d9a4206. Each control has 11 failures, 0 passes, 155 unselected cases; the complete failure case set and failure messages match after replacing the worktree path. These source failures are baseline failures, not introduced by this export-coverage change. No passing full-suite claim is made.

The eight built test shards remain active. Their results and any partial reports/canceled logs will be audited before merge; this source attribution does not establish that every remaining CI failure is unrelated.

[gensokyo-dev]

Final audit of CI 40007458412 on exact head 8dfa6a6f880837baac8d61cac5f19e08c718d83d:

• CI's merged report contains 14 failed assertions. Auditing all seven canceled shard logs found another 47 already-failed assertions that were absent from that report: 61 observed failures total (11 source + 50 built).
• Executed the exact observed cases on CI base 34563fcc9bd92af30668820dcac5635e27181094 and exact CI merge 2ecbbb993dc673cc18df583fb47fd09e0d9a4206, using the same pinned toolchain, lockfile, single worker, serial files and resource guard. Source cohorts each: 11 failures, 0 passes, 155 unselected. Built cohorts each: 50 failures, 0 passes, 347 unselected. Every case and complete failure message matches after replacing only the checkout path; per-file counts and case titles match the CI/canceled logs. No source fix is needed in this PR for those baseline failures.
• The PR-specific and shared-consumer validation remains 166 passing selected tests, npm run check, test discovery, and real libc/libLLVM CLI/stdio observations. The complete 220-tool SDK directory and unmodified libc's complete Evidence are byte-identical to the unchanged control. Codex review completed at this head with no findings.
• CI is still failed, and canceled/skipped lanes remain unverified. This is not a full-suite or real Windows/macOS provider qualification. Per the maintainer's explicit authorization in this maintenance session, baseline-only failures may be bypassed after attribution; merging this exact head under that authorization.

[gensokyo-dev]

Post-merge verification passed on actual main 9fd78a74f84a8e924bd58e7deaf5934bfb24f53c: rebuilt the production runtime, then repeated real public CLI/stdio MCP probes on unmodified libc plus section-table-reduced libc/libLLVM. All six observations agree, reduced copies remain unsupported, and unmodified libc keeps the identical complete report and Evidence ID. Post-merge catalog changes from intervening main commits are recorded separately; the earlier byte-identical catalog result applies to tested PR head 8dfa6a6 and its validated unchanged-main control, not the newer merged main. No target library was loaded/executed; no resource guard stops or payload leftovers in the public probe. #6792 is closed.