From 6e4170790f7385151f4db4ae536f470436eabb2c Mon Sep 17 00:00:00 2001
From: Brennan Benson <79079362+brennanb2025@users.noreply.github.com>
Date: Thu, 20 Aug 2026 00:45:01 -0700
Subject: [PATCH] test(codex): correct what the denyExistence rig actually
 models
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

MEASURED on both platforms: a file-permission denial leaves existsSync TRUE
and fails only the content read — chmod 000 gives EACCES on macOS, icacls
/deny (R) gives EPERM errno -4048 on Windows, with stat/lstat succeeding in
both. The rig's docblock claimed this mode modelled that denial. It does not.

What it models is the UNC / \\wsl$ transport, where an unreachable distro
reports errno UNKNOWN at every level and existsSync folds it to false.

The distinction decides what the D29 guard is worth: under a permission denial
the pre-fix code already failed safe, because existsSync was true so it took
the read branch and threw. Only a transport that lies about existence reaches
the rebuild-from-empty path. No behaviour change; the comment was wrong, not
the code.
---
 ...hared-state-survives-a-failed-read.test.ts | 20 +++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/src/main/codex/sta-4823-shared-state-survives-a-failed-read.test.ts b/src/main/codex/sta-4823-shared-state-survives-a-failed-read.test.ts
index 100a04c6b29..445a994ca73 100644
--- a/src/main/codex/sta-4823-shared-state-survives-a-failed-read.test.ts
+++ b/src/main/codex/sta-4823-shared-state-survives-a-failed-read.test.ts
@@ -26,12 +26,20 @@ const denials = vi.hoisted(() => {
       state.readOnlyPaths.add(path)
     },
     /**
-     * `existsSync` reports false and content reads throw, but `statSync` and the
-     * atomic replace still work. This is the shape that actually loses data: the
-     * caller concludes "no file here", builds fresh content, and the rename
-     * lands because rename needs permission on the DIRECTORY, not the target.
-     * Guarding the write as well would let a test pass because the write failed
-     * rather than because the fix refused.
+     * `existsSync` reports false while the path is still there, and the write
+     * still lands.
+     *
+     * MEASURED, and not what I first assumed: a file-permission denial does NOT
+     * produce this. `chmod 000` on macOS and `icacls /deny (R)` on Windows both
+     * leave `existsSync` TRUE, leave `stat`/`lstat` succeeding, and fail only
+     * the content read (EACCES / EPERM errno -4048). The shape modelled here is
+     * the UNC / `\\wsl$` transport, where an unreachable distro reports errno
+     * `UNKNOWN` at every level and `existsSync` folds that to false.
+     *
+     * The distinction decides what the guard is worth: under a permission denial
+     * the pre-fix code already failed safe, because `existsSync` was true so it
+     * took the read branch and threw. Only a transport that lies about existence
+     * reaches the rebuild-from-empty path.
      */
     denyExistence(path: string): void {
       state.existenceDeniedPaths.add(path)