[mathbunnyru]

Fixes flaky SubscriptionSourceTests on macOS CI, seen as ReadError_Reconnect failures and a segfault in GotResultWithValidatedLedgersWrongValue_Reconnect.

• The serverConnection() test helper assumed the server always gets exactly one failed SSL attempt, then the plain connection.
• With the fixture's 5 ms connection timeout, a loaded runner can time out the SSL attempt before it reaches the server. The helper then took the plain connection as the "failed" one and waited for another. That either hung until the 30 s timeout or picked up the next reconnect's SSL attempt ("Read error: bad method"). The helper then read from the failed connection, which caused the segfault.
• Reproduced locally by running at background priority (taskpolicy -b) with all cores busy, and with a temporary patch that drops the SSL attempt.
• Added TestWsServer::acceptConnectionSkippingSslAttempt(). It skips at most one failed connection, so the SSL attempt may or may not arrive, and returns errors instead of hiding them.
• SubscriptionSourceTests and ForwardingSourceTests, which used the same pattern, now use the helper. The SSL-then-plain fallback is still covered by WsConnectionTests.TrySslUsePlain.
• With the SSL attempt dropped half the time, the old helper failed on the first iteration and the new one passed 200 of 200.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[godexsoft]

Strange workaround but makes sense i guess.