[PR #1375] [CLOSED] fix(browser): reject damaged PNG chunks before screenshot comparison #2837

New issue

Closed

opened 2026-10-17 21:10:51 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-17 21:10:51 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/1375
Author: @rudycelekli
Created: 10/9/2026
Status: ❌ Closed

Base: main ← Head: fix/png-chunk-validation-current-20261009

📝 Commits (2)

b25c616 fix(browser): reject damaged PNG chunks before screenshot comparison
06ff9d5 test(browser): correct simulated capture PNG checksum

📊 Changes

4 files changed (+179 additions, -6 deletions)

View changed files

📝 src/browser/CdpBrowserProvider.ts (+7 -0)
📝 src/browser/PngVisualDiff.test.ts (+137 -0)
📝 src/browser/PngVisualDiff.ts (+34 -5)
📝 tests/fixtures/fakeCdpBrowserResults.ts (+1 -1)

📄 Description

Problem

Screenshot comparison trusted PNG chunks after checking the artifact digest. A real Chrome screenshot with damaged CRC, duplicate header or an unknown critical chunk could be reported as identical even when Chrome could not decode it. Related palette/transparency/chunk-order errors could also produce false pixel comparisons.

Change

Validate chunk CRCs and supported PNG structure before pixel comparison: first/unique IHDR, valid optional PLTE, consecutive IDAT, correctly placed unique tRNS and empty IEND. Keep valid RGB/RGBA, optional palette and ordinary ancillary chunks. Preserve actionable input-format reasons through CLI and MCP instead of suggesting provider repair.

Verification

Frozen baseline: five new test scenarios failed; the ten existing controls passed. After: all 15 tests pass. The ordering scenario exercises eight malformed structures and three valid controls.
Real Chrome 154 produced the original screenshot and rejected three corrupted variants that baseline REA accepted. Final compiled public CLI and stdio MCP reject all three with invalid_input/invalid_format, while the valid original succeeds.
Independent reviewer reran the source suite and native Chrome pixel/decoder oracle: three valid controls remain identical and seven malformed neighbors are rejected. Chrome tolerates some malformed orderings; this does not claim exhaustive PNG conformance or identical decoder policy for every format.
npm run check passed; source/test typecheck, formatting and lint passed. Built-in CRC32 availability also checked on supported Node 22.19.0; main tests use pinned Node 24.18.0.
Refreshed on current main a9f5dad5: unchanged frozen tests reproduce 5 failures/10 passing controls before, 15 pass after; exact locked dependencies, runtime builds and npm run check pass. Fresh Chrome 154 public CLI/stdio MCP repeat the three rejected-image regressions and valid control. Subsequent main changes through 39dabd06 do not touch either production owner.
Hosted verification exposed an invalid IDAT CRC in the simulated CDP screenshot fixture. Corrected only the four checksum bytes, preserving size and pixels; 44 affected PNG/MCP/CLI/provider owners and required check pass. Independent CRC inspection and actual Chrome pixel equality verified. Hosted rerun remains pending.
Other PRs/current main exhibit a separate short-circuit alias graph expectation failure, under native-oracle investigation; no production assertion has been weakened to force green.
No Hopper/Ghidra execution required/performed. Hosted checks may still be pending.

AI assistance was used for investigation, implementation and verification.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/morluto/rea/pull/1375 **Author:** [@rudycelekli](https://github.com/rudycelekli) **Created:** 10/9/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `fix/png-chunk-validation-current-20261009` --- ### 📝 Commits (2) - [`b25c616`](https://github.com/morluto/rea/commit/b25c616af951fcce41ef1d161561b3615327b773) fix(browser): reject damaged PNG chunks before screenshot comparison - [`06ff9d5`](https://github.com/morluto/rea/commit/06ff9d5187d05cb4fb4238b6194a9b0bd3e60a3b) test(browser): correct simulated capture PNG checksum ### 📊 Changes **4 files changed** (+179 additions, -6 deletions) <details> <summary>View changed files</summary> 📝 `src/browser/CdpBrowserProvider.ts` (+7 -0) 📝 `src/browser/PngVisualDiff.test.ts` (+137 -0) 📝 `src/browser/PngVisualDiff.ts` (+34 -5) 📝 `tests/fixtures/fakeCdpBrowserResults.ts` (+1 -1) </details> ### 📄 Description ## Problem Screenshot comparison trusted PNG chunks after checking the artifact digest. A real Chrome screenshot with damaged CRC, duplicate header or an unknown critical chunk could be reported as identical even when Chrome could not decode it. Related palette/transparency/chunk-order errors could also produce false pixel comparisons. ## Change Validate chunk CRCs and supported PNG structure before pixel comparison: first/unique IHDR, valid optional PLTE, consecutive IDAT, correctly placed unique tRNS and empty IEND. Keep valid RGB/RGBA, optional palette and ordinary ancillary chunks. Preserve actionable input-format reasons through CLI and MCP instead of suggesting provider repair. ## Verification - Frozen baseline: five new test scenarios failed; the ten existing controls passed. After: all 15 tests pass. The ordering scenario exercises eight malformed structures and three valid controls. - Real Chrome 154 produced the original screenshot and rejected three corrupted variants that baseline REA accepted. Final compiled public CLI and stdio MCP reject all three with invalid_input/invalid_format, while the valid original succeeds. - Independent reviewer reran the source suite and native Chrome pixel/decoder oracle: three valid controls remain identical and seven malformed neighbors are rejected. Chrome tolerates some malformed orderings; this does not claim exhaustive PNG conformance or identical decoder policy for every format. - `npm run check` passed; source/test typecheck, formatting and lint passed. Built-in CRC32 availability also checked on supported Node 22.19.0; main tests use pinned Node 24.18.0. - Refreshed on current main a9f5dad5: unchanged frozen tests reproduce 5 failures/10 passing controls before, 15 pass after; exact locked dependencies, runtime builds and `npm run check` pass. Fresh Chrome 154 public CLI/stdio MCP repeat the three rejected-image regressions and valid control. Subsequent main changes through 39dabd06 do not touch either production owner. - Hosted verification exposed an invalid IDAT CRC in the simulated CDP screenshot fixture. Corrected only the four checksum bytes, preserving size and pixels; 44 affected PNG/MCP/CLI/provider owners and required check pass. Independent CRC inspection and actual Chrome pixel equality verified. Hosted rerun remains pending. - Other PRs/current main exhibit a separate short-circuit alias graph expectation failure, under native-oracle investigation; no production assertion has been weakened to force green. - No Hopper/Ghidra execution required/performed. Hosted checks may still be pending. AI assistance was used for investigation, implementation and verification. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-17 21:10:51 +00:00

closed this issue
added the
pull-request
label

bsakaguchi referenced this issue

2026-10-31 01:06:33 +00:00

[PR #2837] [CLOSED] fix(targets): recognize WebAssembly before inferring a .hop database #5428

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

morluto/rea#2837

No description provided.

Rows
Columns