[benjamin-di]

On TCP/TLS, a client transaction can finish immediately after queuing a final response. If both Done() and Responses() are ready, sipResponse can select completion first and report an incomplete transaction despite a buffered final response. This also loses authentication challenges before a retry can be constructed.

Drain queued responses when completion is observed, retaining normal cancellation and missing-final-response handling. Regressions cover buffered 200, 401, 403 and 407 responses after a provisional response, plus completion without a final response when the response channel is either closed or still open. The open-channel cases are bounded and release their owned channel on failure; they check that an incomplete transaction returns promptly and retains the provisional-response count.

Validation on upstream 1df341e plus this patch: focused race-enabled tests repeated ten times; go test -race -coverprofile=coverage.out ./pkg/... passed. sipResponse has 100% local statement coverage, including the previously untested completion-with-empty-open-channel return. Replacing the nonblocking drain with a blocking receive makes both new cases fail within their watchdog, without a race or global test timeout. The original buffered-final regression fails on the unchanged stable implementation and passes with the fix.

The latest commit is a test-only follow-up. The runtime patch carries no transport/configuration changes and adds no dial retries.

Rebase validation (2026-10-30)

Rebased onto ffd1a64b1755; current head 7c0255e79b00. go test -race ./pkg/... passes on this head. No live SIP call was made.

Rebase validation (2026-10-31)

Rebased onto main at e73272dedf3b; current head 1eca7b475a7d.

The complete go test -race ./pkg/... suite and git diff --check pass. No live SIP call or separate integration fixture was run after this rebase.

[CLAassistant]

All committers have signed the CLA.

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 70.42%. Comparing base (0460b40) to head (1eca7b4).
⚠️ Report is 435 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #923      +/-   ##
==========================================
+ Coverage   65.25%   70.42%   +5.16%     
==========================================
  Files          51       46       -5     
  Lines        6588     9149    +2561     
==========================================
+ Hits         4299     6443    +2144     
- Misses       1915     2191     +276     
- Partials      374      515     +141     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[benjamin-di]

Added the test-only follow-up in 37a61ef for a completed transaction whose response channel is empty but still open, both without responses and after a provisional response. It verifies a prompt incomplete-transaction error with the correct response count; the watchdog releases its owned channel before reporting a stalled receive.

The focused race-enabled tests pass ten repeated runs, and go test -race -coverprofile=coverage.out ./pkg/... passes. Both new cases fail with a deliberately blocking receive. Hosted tests/integration pass, and Codecov now reports all modified lines covered. No runtime change.