Before doing a new release of pg8000 I generally run the full suite of SQLAlchemy tests (as well as pg8000's own test suite). This time with pg8000 1.10.2 I found that the tests hung at:

test/dialect/test_oracle.py::CompatFlagsTest::test_ora8_flags PASSED

doing CTRL-C let the tests carry on, not missing out any. Since pg8000 passed its own tests, I went ahead and released it anyway. Publish and be damned! Not a good attitude.

So here I am at the Bristol Hack Night, and determined to fix this bug. I reached out to the SQLAlchemy mailing list, and got some pointers from the ever helpful and hugely talented Mike Bayer.

I've made no progress. Even printing stuff out to stderr made the bug go away. What will I do?

Skip forward many days...

I'm now at the Meanbee Hack Night in Bath. After the lack of progress last time, I've got some new leads. The aforementioned Mike Bayer made a second post saying:

44a9820b4e02f65b3884fa2c016efc has a fix which adds explicit cleanup to a few connection objects that are checked out and not closed, leading to the pool._refs collection to not be empty when that particular test starts.

So I'll fetch that and see how the tests run...

Everything runs okay now, so case closed.