[chrjohn]

No description provided.

[chrjohn]

@copilot
BogusTrustManagerFactory is missing from MINA now. Can we copy it from an older version into our code base?

[Copilot]

@copilot
BogusTrustManagerFactory is missing from MINA now. Can we copy it from an older version into our code base?

Done. I copied BogusTrustManagerFactory (Apache 2.0 header kept) into quickfix.mina.ssl and removed the MINA import from SSLContextFactory. The class compiles standalone with javac. I couldn't run the full Maven build here because it fails on missing generated resources, so the quickfixj-core module build and its tests haven't been run.

[chrjohn]

@copilot
Build is failing. Seems like there is a recursion.

[Copilot]

@copilot
Build is failing. Seems like there is a recursion.

I couldn't find a recursion, and I haven't fixed the build. I made no new commit. The CI logs have no StackOverflowError, and the failures reproduce locally.

• Where it fails: quickfixj-core tests fail on MINA 2.2.10: SSLCertificateTest, SocketInitiatorTest, HttpProxyTest, SocksProxyTest, SSLAndNonSSLTest, and the acceptance tests (heartbeat timeouts). The compile error from the missing BogusTrustManagerFactory is gone.
• What the failures look like: In SSLCertificateTest only the TLSv1.3 [1] runs fail, and only after earlier tests in the same JVM. Run alone, SSLCertificateTest#shouldAuthenticateServerCertificate* and SocketInitiatorTest#testLogonAfterServerDisconnect pass on 2.2.10. Failing runs show the TLS handshake finishing, the initiator sending Logon, and the acceptor never receiving it.
• Thread dumps: A dump from the full run showed QFJ Timer and ssl-exec-* threads BLOCKED inside MINA's SSLHandlerG1. That is MINA's new non-blocking SSL pipeline, which 2.2.10 enables by default.
• What I tried: Calling setUseNonBlockingPipeline(false) in AbstractSslFilter switches to the older handler, but SSLCertificateTest still had 20 failures. I reverted that change.
• Version check: The full SSLCertificateTest class passes 48/48 on MINA 2.2.4. It fails 22 of 48 on 2.2.5, with the setUseNonBlockingPipeline(false) change still in place for that run. So the regression came in with a MINA release after 2.2.4.

I think this is a MINA 2.2.5+ regression, not something in our code, but I haven't confirmed it with MINA. Staying on 2.2.4 is the safe option, or the bump could wait for a MINA fix. Which do you want?

[chrjohn]

Superseded by #1348