Race Condition in File Creation

Race Condition in File Creation | Generated by AI

Home 2026.04

Looking at the code path, a lock isn’t the root cause. The duplicate happens because create_filename silently falls back to a counter suffix when the slug already exists.

The race path the user hypothesizes (concurrent ww note):

Two invocations on the same machine can race past check_duplicate_notes():

Both pass check_uncommitted_changes() and git pull --rebase
Both call check_duplicate_notes() — neither sees the other’s file yet

Both call create_filename() — one gets ...-en.md, the other hits the fallback at create_note_utils.py:77-79:

while os.path.exists(file_path):
    file_path = os.path.join(notes_dir, f"{date_str}-{short_title}-{counter}-en.md")

Both commit & push — end result: two near-identical notes.

A lock would prevent this race. But it wouldn’t prevent the more common path.

The likelier path (and what the test output actually shows):

The pair 2026-04-20-iphone-daylight-timelapse-en.md and ...-timelapse-1-en.md is the signature of the counter-suffix branch. That branch can be reached even sequentially, not just concurrently, when:

check_duplicate_notes misses because _are_notes_quick_similar requires the first 250 chars byte-identical + ≥450/500 char matches — any whitespace/clean-up diff dodges it (check_duplicate_notes.py:12-30)
LLM regenerates the same slug for similar-but-not-identical clipboard content
create_filename treats an existing slug as a signal to disambiguate, not as a red flag

What actually fixes it:

Make slug collision fatal in create_filename — if {date}-{slug}-en.md exists, abort (or require a flag). The counter-suffix fallback is the real hole.
Strengthen check_duplicate_notes to also flag same-date same-slug before writing.
Optionally add a flock on the notes dir for belt-and-suspenders against the in-process race.

Fix (1) is the highest-leverage — it closes both the race and the sequential repeat. Want me to make that change?

Back Donate