Repository navigation
Windows: path_identity atomic replace (ReplaceFileW) fails beyond MAX_PATH, host request marker replacement returns store_transient #2926
Description
Activity
Technical score (task-technical-scoring): Scope 3 (one helper + test) · Diagnosis 2 (cause stated, confirmed in #2923) · Domain 8 (Win32 path/UTF-16 rules) · Ambiguity 4 (pattern already decided in #2923: canonical parent + original leaf) · Blast 8 (shared atomic-replace helper) · Dependencies 2 · Verification 9 (forced-long fixture, negative control required) · State 2 · Concurrency 3 · Environment 7 (Windows-only, TEMP length). Raw 48 → final 21. Band 1-25 → Lite (tests change, so not Express). Veto: Verification >= 8 → Grinch also reviews the proof. Author: ac-dev-rust-v4; reviewer: ac-dev-rust-grinch-v4; no digest. Excluded as ceremony: issue, branch, PR, CI, reviews.
Step 7.5: user approved plan D:/0_repos/AgentsCommander_iac/.ac/plans/2926-path-identity-replacefile-long-path/plan-v3-path-identity-replacefile-long-path.md, Plan-Revision r1 (Grinch PLAN_APPROVED r1). No build requested.
Delivered in PR #2934, merged to main as afbffb8. Change: src-tauri/src/path_identity.rs only (+82): atomic_replace_existing canonicalizes the parent of both paths and keeps the original leaf before ReplaceFileW; new test atomic_regular_replacement_supports_long_windows_paths (negative control failed on main, passes with fix). Reviews: Grinch PLAN_APPROVED r1 and PASS on implementation + proof; user approved plan. CI green after one rerun of an unrelated timing test. No build was requested.
Situation
On Windows,
path_identity::replace_regular_file_atomic/atomic_replace_existing(src-tauri/src/path_identity.rs:1361-1394) passes both raw paths toReplaceFileW, which is limited to MAX_PATH. The temp file is created next to the marker (:1426-1431), so a long outbox root pushes it past the limit.Impact
Replacing a host request with its marker (
src-tauri/src/phone/mailbox.rs:4773) fails beyond MAX_PATH, and the error is mapped tostore_transient. Host PTY requests in long outbox roots never get marker-replaced.Evidence
Found while planning #2923: the forced-long fixture failed at this step (
mailbox.rs:16198) before reaching the #2923 defect. Confirmed from source by the adversarial reviewer. #2923 deliberately leaves this out of scope.Why / how
It is the same class of defect as #2923, in a second helper. It needs long-path handling for both
ReplaceFileWarguments inpath_identity.rs, plus a regression test. The design is decided in that issue's plan.