[kinyoklion]

Summary

A rewrite of a real file is not atomic. Since an empty file became a document with no entries, a retry that reads a test's file after it is emptied and before it is written applies an empty document, which is correct, and stops retrying. ReportsIdenticalFailureOnlyOnce relied on a retry to see the finished write and timed out on Linux CI (#363). Three other tests have the same exposure: RetriesAfterFailureWithoutFurtherTriggers, SuccessfulReloadNowDisarmsThePendingRetry, and RecoveryAppliesEvenWhenContentIsUnchanged.

The four tests now read their content from a scripted reader, so every read sees one complete content. With the tests' own writes made non-atomic (truncate, pause, write), all four failed before this change, and every reloader test passes after it. Each rewritten test fails against a mutant of the behavior it covers. Test-only; no SDK code changes.

_{Opened from termag}

SDK-3248

---

Note

Overview
Fixes flaky FileDataReloader tests on Linux CI where automatic retries could read a real file mid-rewrite (empty or partial JSON), apply a valid-but-wrong empty document, and stop retrying before the test’s intended content was visible.

Adds a ScriptedContentReader test double that implements IFileReader and returns atomically updatable in-memory content (or a simulated missing file when set to null). Four retry-focused tests now drive content through that reader instead of Write() / File.Delete: ReportsIdenticalFailureOnlyOnce, RetriesAfterFailureWithoutFurtherTriggers, SuccessfulReloadNowDisarmsThePendingRetry, and RecoveryAppliesEvenWhenContentIsUnchanged. SuccessfulReloadNowDisarmsThePendingRetry also drops a DrainErrors() call that is no longer needed with the scripted reads.

Test-only; no production SDK behavior changes.

^{Reviewed by Cursor Bugbot for commit bec2b6b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[kinyoklion]

bugbot review

[kinyoklion]

bugbot review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit bec2b6b. Configure here.}

[kinyoklion]

This reply was written by an AI agent (Claude Code) working on behalf of @kinyoklion.

Closing this in favor of #394, which fixes the cause in the SDK instead of in the tests. With #394, the reloader holds back a file that a load finds empty while its content is in effect, until it has stayed empty for the retry delay (1 second), and an empty JSON file no longer parses. So a retry that reads a file in the middle of a rewrite keeps the data and reads it again. The reloader tests keep real in-place writes (truncate, pause, write) instead of a scripted reader, including the four tests this PR changed.