This is happening because the legacy implementation didn't even attempt to read the HTTP response body. I added that functionality in the standard case but not in the chunked encoding case.

I believe that #1549 will wrap that error in an InvalidMessage, which is closer to what you want. However, it still doesn't give you the HTTP error code, because websockets parses the entire request first and only then starts the opening handshake logic.

I could make an argument for skipping the body rather than crashing in this case. Let me think about it.

I forgot to mention — as a short term fix, you can use websockets.legacy.client.connect instead of websockets.connect and you buy five years. See https://websockets.readthedocs.io/en/latest/project/changelog.html and https://websockets.readthedocs.io/en/latest/howto/upgrade.html for context.

Thanks! I guess I'll go with the legacy code for now. For the record, though, I don't really care about the message body in this case, and I don't mind altering my code, I just need some way to figure out the error code.

Yes, it's a legitimate request for improvement. The HTTP code is right there and it's reasonable to expect websockets to give it to you.

I took a stab at fixing it in #1568.

I wanted to test against a real life server rather than just my test suite but I couldn't find one in 15 minutes and I'm not really excited with the idea of installing Erlang and Cowboy locally (-:

Any chance you can test the branch before I merge the PR?

Thanks for looking into it, I can confirm that the issue-1550 branch fixes the problem on my end!

Great, thank you for reporting this issue and testing the fix!