http.server has support for old-style HTTP 0.9 requests ("command path" without a version string). This issue proposes to remove this support, since there are probably no clients around which still issue such requests.

See issue10711 for the client side.

LGTM

The question really is whether we want or want not to support RFC 1945 (i.e HTTP/1.0). If we want to support it, we also must comply to section 3.1, which says

 The version of an HTTP message is indicated by an HTTP-Version field
 in the first line of the message. If the protocol version is not
 specified, the recipient must assume that the message is in the
 simple HTTP/0.9 format.

If we then choose to not support HTTP/0.9 (after recognizing that the message is HTTP/0.9), we must then reject the request. According to RFC 2145, we then may send back error 505 (HTTP version not supported).

If a client sends a request without specifying the version, GET /, then the Python server should behave like other common servers are behaving, namely support rfc 1945 for HTTP 1.0 spec and return the response without headers.

Removing the HTTP 0.9 from the http.client is okay. But punishing *any client* for not sending the correct request is not so desirable, especially when there are other servers which are lenient to his behavior for whatever reasons.

So, I am not a firm +1 to this. We should arrive a conclusion and then go about with the change. Martin's point is valid here too.

How, if at all, has the requirement for HTTP 0.9 support changed since this issue was first raised?

To answer your question, Mark, RFC 7230 has removed the expectation that HTTP/1.1 servers will be able to support HTTP/0.9 requests.

Does this need any deprecation cycle, or can we just remove everything immediately in 3.6?

I suspect that the HTTP 0.9 support never properly worked, but I don’t have enough info to be sure. Wouldn’t a HTTP 0.9 request send b"GET <path>\r\n" with a single line, and then wait for a response? If so, Python’s server hangs waiting for a second blank line (or EOF), so the protocol will deadlock.

In any case, I would be in favour of removing HTTP 0.9 support. Senthil: do you know of actual clients that do HTTP 0.9 requests?

I left a review comment. The patch will also need manual merging with current code.

Also I guess documentation should be added somewhere, even if there is no deprecation step.

FWIW I doubt HTTP 1.1 ever required support for other HTTP versions; see for instance <https://tools.ietf.org/html/rfc2068#section-19.7>.

This v2 patch adds another test that includes what I understand is a true HTTP 0.9 request (although the mock connection infrastructure closes the connection before receiving the response, so you don’t see any deadlock). This is based on <https://tools.ietf.org/html/rfc1945#page-22>, which includes only one CRLF newline. I also added an entry to the What’s New page.

There is so very little code in support of HTTP/0.9 that I don't think we get any benefit from removing it.  So why bother with the churn?  It is cleaner to just leave it place.

There are a few small benefits from removing it, but I am not too fussed if we decide to leave it.

* If we keep it, should we fix it so that it doesn’t deadlock? Otherwise, we carry around buggy and untested code that claims to be a HTTP 0.9 server but isn’t. Keeping it as it is doesn’t feel “clean” to me.

* Easy way to resolve bug reports like Issue 26578.

* Potentially slightly easier to maintain the rest of the code in the future.

If we do keep it, I would like to add comments clarifying that it does not implement the real HTTP 0.9 protocol.

Here is a slightly updated patch I have been sitting on. I think the main difference was tweaks to the documentation.

Issue 28548 would also benefit if the HTTP 0.9 code was removed. The presence of the HTTP 0.9 code means a suboptimal response is sent in some error cases. IMO it would be better if the server responded using HTTP 1.0, where we can use error codes like “400 Bad request”. The current code will not use HTTP 1.0 unless it sees that the client supports it.

V4 patch adjusting to recent code change.

Raymond: Given the problems caused by the current code, would you reconsider your opposition?

Otherwise, what do you think should be the future of the code? Should we fix it so that it handles real HTTP 0.9 requests, or just admit that it doesn’t implement the protocol properly.