test(desktop): the unreachable-server test keeps its port

[mbeisser1]

Bug Description

Expected: an_unreachable_server_is_reported_without_the_media_link in src-tauri/src/commands/download.rs (renamed a_server_that_hangs_up_is_reported_without_the_media_link here) passes on every run.

Actual: it failed once in four full src-tauri runs, when the request reached a live server.

Root Cause

The test bound 127.0.0.1:0, read the port, and released the listener before the request. Seven other tests in the crate start httpmock servers on ephemeral ports, so one of them could take the released port in between, and the request then reached that server instead of failing.

Fix Description

The test keeps its listener for its whole life. A thread accepts each connection and drops it at once, so the request fails immediately with "Could not reach the server", the port is never free for another test, and no timeout is involved. The assertions are unchanged. Test-only.

How to Reproduce (Before Fix)

1. Run cargo test --manifest-path src-tauri/Cargo.toml several times.
2. Occasionally another test's httpmock server takes the released port.
3. The test fails because the error does not start with "Could not reach the server".

How to Verify (After Fix)

1. cargo test --manifest-path src-tauri/Cargo.toml a_server_that_hangs_up passes in about 0.03 s.
2. The full src-tauri suite passes (206 tests).
3. Read the test: the listener is moved into a thread and never dropped before the request.

Impact Assessment

• Severity: Low
• Users affected: None; developers and CI see a flaky test.
• Duration: Since the test was written; it predates PR feat(desktop): download ffmpeg, ffprobe and wtsexporter into the Tools Directory at start (#1053) #2001.

Regression Risk

None for the product: only a test changes. The accepting thread lives until the test process ends, which holds one port and one idle thread.

Checklist

• Root cause identified and documented above
• Fix addresses the root cause (not just symptoms)
• Added test to prevent regression (the fix is the test itself)
• Existing tests pass locally
• Tested the specific reproduction steps

Related Issues

Closes #2002.

🤖 Generated with Claude Code

[mbeisser1]

Review of #2020: Standards 2 findings (one has no line in the diff and follows in a top-level comment), Spec none, Correctness none.

[mbeisser1]

Finding with no line in the diff.

Standards · judgement (PR description): the checklist ticks "Added test to prevent regression", but the PR adds no test; it changes how an existing test holds its port. A ticked box that is false misleads a later reader. Fix: untick it and say the fix is the test itself.

[mbeisser1]

Standards · judgement (PR description): the checklist ticks "Added test to prevent regression", but the PR adds no test.

Fixed in the description: the box is unticked with "(the fix is the test itself)", and the description names the renamed test.

Review of #2020, closed out.

• Standards: 2 findings (2 Fixed); re-review found nothing.
• Spec: no findings.
• Correctness: no findings (the reviewer ran the test binary alone 300 times with no failure); re-review found nothing.
• Deferred: none. Declined: none. No commits for CI failures; no merges of the base. No user threads.