There should also be a mechanism to get the actual headers of 101 responses (maybe 1xx in general?) I'm not sure which mechanism, though.

I wanted to fetch the headers from the 101 response, because that's what I'm interested in. Everything after the 101 response + headers is protocol specific and shouldn't be handled by PHP imho. With this patch it was easily possible to implement a WebSocket handler on top of Guzzle.

It seems to me the current code for handling 101 is a bug, so a PR may target PHP-7.4.

Actually that was my intention but I somehow messed up.

Thanks for the PR, but the solution doesn't look right.

Imho this condition should solely check for HTTP 100-continue but that might break things

Thanks for the clarification!

Imho this condition should solely check for HTTP 100-continue but that might break things

Yes, that's why I suggested to implement some mechanism to be able to get the headers of such requests. Maybe something like follow_location; if that was enabled by default, no BC break would be introduced.

Anyhow, I don't think that we actually break anything in practice with your patch (in theory maybe upgrade requests from HTTP/1.0 to HTTP/1.1, but 1xx responses are not supposed to be sent to HTTP/1.0 clients anyway). But I'm not sure.

Could you please file a respective bug ticket for easier reference. And, if possible, adding a test case would be great as well.

Yes, that's why I suggested to implement some mechanism to be able to get the headers of such requests. Maybe something like follow_location; if that was enabled by default, no BC break would be introduced.

Adding a new stream context option should be do-able. Something like return_informational (or return_http1xx)?

Could you please file a respective bug ticket for easier reference. And, if possible, adding a test case would be great as well.

Sure

Adding a new stream context option should be do-able. Something like return_informational (or return_http1xx)?

Something like this, or maybe the other way round, e.g. ignore_informational. Might be worth a discussion (attempt) on the internals mailing list.

I've added the requested bug report and a unit test.

Thank you!