[Python] mrhttp: a project with many problems about frameworkbenchmarks

[errantmind]

No comment on standards compliance (haha) but the primary issue I can see is mrhttp is using a cached implementation which not allowed by the tfb rules for plaintext.

[uNetworkingAB]

Forget about mrhttp. Look at gnet. Its entire HTTP parser literally is:

requests[] = split(data, '\r\n\r\n')

and was the official 1st place in Round 22. And it has been tagged as "implementation approach: realistic".

[uNetworkingAB]

I don't think mrhttp does anything spectacularly wrong. It clearly does parse the headers and it does apply some kind of higher level handling above TCP. It uses AVX2 intrinsics to get a bitmap of the ; or \r which is an okay way. It's most likely not standards compliant or secure in any way, but it does look like an HTTP server to me. I'll give it a pass.

[remittor]

@uNetworkingAB

It's most likely not standards compliant or secure in any way

So this is not just a matter of compliance with the HTTP standard! I have already described above that the basic principle of working with the TCP data stream has not been observed!
If the HTTP-request is split into 2 parts (which TCP allows), then server mrhttp will not be able to understand that it was sent 1 HTTP-request.

but it does look like an HTTP server to me. I'll give it a pass.

First, mrhttp must become a simple TCP-server, and only then - an HTTP-server.