Don't send client CONNECTION_CLOSE on 1-RTT before handshake completion.

[rzikm]

Description

Fixes #4132.

Testing

Fixed tests so that they would detect the bug before the fix.

Documentation

No.

[rzikm]

I don't see how the server would get deleted. The new connection instance is held by the unique_ptr above and neither the AutoDelete bit nor ShutdownEventCallback is set, so the connection should be valid until the unique_ptr goes out of scope.

[nibanks]

I believe this was meant to handle a case where the listener could reject the connection, and then this pointer would never be set. That being said, I cannot see a path where that would happen in this test case (or the one below). And since the tests seem to pass just fine, I'm ok with these changes.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 65.02%. Comparing base (86e08e5) to head (de3f2ef).
Report is 36 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #4145       +/-   ##
===========================================
- Coverage   87.38%   65.02%   -22.36%     
===========================================
  Files          56       56               
  Lines       16984    15189     -1795     
===========================================
- Hits        14841     9877     -4964     
- Misses       2143     5312     +3169     

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

[nibanks]

@rzikim would it be possible to add (or modify) a test to validate #4132 is actually fixed?

[rzikm]

@rzikim would it be possible to add (or modify) a test to validate #4132 is actually fixed?

That is handled by the tests modified in this PR, without the fix in packet_builder.c, the async/reject case would've failed. This was not detected before because the test did not wait for server shutdown notification which would've timed out.

[nibanks]

@rzikim would it be possible to add (or modify) a test to validate #4132 is actually fixed?

That is handled by the tests modified in this PR, without the fix in packet_builder.c, the async/reject case would've failed. This was not detected before because the test did not wait for server shutdown notification which would've timed out.

Ah! I see. Ok, then let's ship it!