node does not process HTTP informational responses correctly

[daguej]

There are currently two places in the HTTP client that look for a HTTP 1xx response. Both are to implement 100 Continue:

• L405: Don't free the parser when the 100 message is complete because a final response will come later
• L449: Emit the continue event and reset the parser

There is also some code to handle upgrades, which covers HTTP 101 Switching Protocols.

This is good, but is not fully in line with the spec:

A client MUST be able to parse one or more 1xx responses received prior to a final response, even if the client does not expect one. A agent MAY ignore unexpected 1xx responses.

node should deal with responses other than Continue and Switching Protocols -- ≥ 102 ≤ 199. The current behavior is for node to emit a response with the 1xx status code and headers, ending the response. This is incorrect: we MUST be able to cope with one or more 1xx informational responses before the final response arrives.

A good example of this is 102 Processing¹. A server MAY send this informational response to inform the client that it's still working on the request:

GET /long-running-operation HTTP/1.1
Host: some-server

HTTP/1.1 102 Processing
X-Maybe-Extra-Header: foo

HTTP/1.1 102 Processing

HTTP/1.1 200 OK
...

^{1 - I know that RFC 4918 removed a definition for HTTP 102 "due to lack of implementation," but it remains an IANA-registered HTTP response code as originally defined in the otherwise obsolete RFC 2518. In other words, HTTP 102 is still a perfectly legal response code that should be supported.}

So when a HTTP response is received with a status code ≥ 102 and ≤ 199, node should not emit a request event; instead, it should reset the HTTP parser in preparation for the final response message.

Additionally, I propose adding an info event to the ClientRequest object that provides a IncomingMessage representing the informational response. This will allow an application interested in handling the 1xx update an opportunity to do so.

[Trott]

@nodejs/http

[miles-po]

I opened PR #18033 to address this. Will need help with tests.

[apapirovski]

Fixed in #18033