[PR #2967] [CLOSED] fix(apple): ignore inert XML declarations in plist preflight #5524

New issue

Closed

opened 2026-10-31 01:07:18 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 01:07:18 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/2967
Author: @rudycelekli
Created: 10/13/2026
Status: ❌ Closed

Base: main ← Head: candidate/rea-next10-xml-inert-20261013

📝 Commits (2)

0769dd5 fix(apple): ignore inert XML declarations in plist preflight
bd02383 fix(apple): scan plist XML preflight tokens once

📊 Changes

3 files changed (+190 additions, -17 deletions)

View changed files

➕ src/artifacts/apple/InterfaceBuilderDecodeBudget.test.ts (+106 -0)
📝 src/artifacts/apple/InterfaceBuilderDecodeBudget.ts (+56 -17)
📝 tests/boundary/filesystem/appleXmlDeclarations.test.ts (+28 -0)

📄 Description

An ordinary Interface Builder plist containing a DOCTYPE example inside a comment or CDATA is currently omitted as an entity-expansion budget failure. Consume opaque comment/CDATA/processing-instruction text before inspecting declarations. Use one forward-only, quote-aware token scan so malformed unterminated declarations cannot repeatedly rescan suffixes. Real internal entity subsets remain refused before decoding; the pinned decoder still owns syntax and values. The public workflow retains the original archive digest, title and complete archive coverage. AI-assisted implementation and verification with OpenAI Codex; no linked issue because this came from a reproduced source audit.

Verified against base 8fef2aa6: npm run check, test discovery and 80 distinct focused/boundary tests pass, including the existing Interface Builder bundle and compiled CLI cases. The committed bounded-child regression fails on the previous regex after its five-second deadline and passes with the forward-only scanner; additional controls preserve quoted external identifiers containing both > and [, opaque processing instructions, and internal-subset refusal. Fresh compiled CLI and SDK stdio MCP outputs agree on Evidence ID and normalized result, with subsequent ping and client cleanup. The owned app contains a benign clang-built Mach-O that was inspected, never executed.

_{🔄 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/2967 **Author:** [@rudycelekli](https://github.com/rudycelekli) **Created:** 10/13/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `candidate/rea-next10-xml-inert-20261013` --- ### 📝 Commits (2) - [`0769dd5`](https://github.com/morluto/rea/commit/0769dd52c6274931b40675cad6a7f444a97a94ae) fix(apple): ignore inert XML declarations in plist preflight - [`bd02383`](https://github.com/morluto/rea/commit/bd023834aa66f9dc8a546d1d24ce3705eb2127c4) fix(apple): scan plist XML preflight tokens once ### 📊 Changes **3 files changed** (+190 additions, -17 deletions) <details> <summary>View changed files</summary> ➕ `src/artifacts/apple/InterfaceBuilderDecodeBudget.test.ts` (+106 -0) 📝 `src/artifacts/apple/InterfaceBuilderDecodeBudget.ts` (+56 -17) 📝 `tests/boundary/filesystem/appleXmlDeclarations.test.ts` (+28 -0) </details> ### 📄 Description An ordinary Interface Builder plist containing a DOCTYPE example inside a comment or CDATA is currently omitted as an entity-expansion budget failure. Consume opaque comment/CDATA/processing-instruction text before inspecting declarations. Use one forward-only, quote-aware token scan so malformed unterminated declarations cannot repeatedly rescan suffixes. Real internal entity subsets remain refused before decoding; the pinned decoder still owns syntax and values. The public workflow retains the original archive digest, title and complete archive coverage. AI-assisted implementation and verification with OpenAI Codex; no linked issue because this came from a reproduced source audit. Verified against base `8fef2aa6`: `npm run check`, test discovery and 80 distinct focused/boundary tests pass, including the existing Interface Builder bundle and compiled CLI cases. The committed bounded-child regression fails on the previous regex after its five-second deadline and passes with the forward-only scanner; additional controls preserve quoted external identifiers containing both `>` and `[`, opaque processing instructions, and internal-subset refusal. Fresh compiled CLI and SDK stdio MCP outputs agree on Evidence ID and normalized result, with subsequent ping and client cleanup. The owned app contains a benign clang-built Mach-O that was inspected, never executed. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-31 01:07:18 +00:00