[dxbjavid]

Calling open on a NativeDB that is already open makes _open_utf8 throw "DB already open" and then close the existing sqlite3 handle, but the freed pointer is left behind in NativeDB.pointer, so later calls on that connection touch freed memory; in practice the next connection opened tends to reuse the same allocation, and closing the first connection then closes the second one's database and leaves it on a dangling handle. The fix only throws on that path and leaves the open connection as it was, so it stays usable and is closed normally later, and the new test in ConnectionTest fails with SQLITE_MISUSE without it.

[gotson]

Hi, CI is failing on the newly added test.

[dxbjavid]

That failure is the bug itself rather than a problem with the test. The regular jobs run against the prebuilt binaries checked into src/main/resources, which still have the sqlite3_close on the already-open path, so the statement after the rejected open lands on the freed handle and comes back with SQLITE_MISUSE, which is exactly what the test is there to catch. The one job that compiles the native library from source (test external amalgamation) passes the full suite including this test, and the change in NativeDB.c only takes effect once the native libs are rebuilt.

I tried to run the Build Native workflow on my fork to push rebuilt libraries to the branch, but Actions can't be switched on for the fork from my side. If you can run Build Native against this branch the test should go green; otherwise I can rebuild and push the libraries myself, or drop the test if you'd rather keep the PR to the C change.

[gotson]

thanks for the clarification