Nicolas Grekas

>the meaning of ReturnTypeWillChange in PHP is that we acknowledge that we will change our own return type later, before migrating to a version when the parent forces us to...

Would it be possible to build this on the ReturnTypeWillChange attribute? I had a quick look at the code and we don't do anything at the moment, but it would...

What's controversial is the annotation. I pushed a commit where the existing attribute is used instead. > I understand ReturnTypeWillChange to be for the implementing class rather than the parent...

What understand about the attribute: it means what it names says, aka "this method is going to have a native return type in the future". And because the author of...

OK, let me close. I think this approach with the attribute would be useful because it works in more situations. In addition to the phpstan rule, that'd give better visibility...

~This HTTP transaction is not valid per HTTP/1.1~ `Connection: close` is how HTTP/1.0 behaved, but since 1.1, either Content-Length should be sent, or Transfer-Encoding should (to chunked typically). Are you...

HTTP/1.1 actually allows the server to close the connection to tell about the end of the response: https://www.rfc-editor.org/rfc/rfc2616#section-4.4 (item 5.) We now need to be sure the code you proposed...

(and yes, a test case would be great, to confirm the behavior of all implementations, not only curl)

See https://github.com/curl/curl/issues/4624 It looks like curl ignores this error unless it's built with the debug flag. But we need to make this specific for the SSL_ERROR_SYSCALL error. Not sure if...

> curl_error($ch) === 'OpenSSL SSL_read: SSL_ERROR_SYSCALL, errno 0' yes > I'd still keep the restriction on 0.0 === curl_getinfo($ch, \CURLINFO_SIZE_DOWNLOAD) nope, that'd be wrong > -1.0 === curl_getinfo($ch, \CURLINFO_CONTENT_LENGTH_DOWNLOAD) here...