Stop setting HTTP_VERSION to SERVER_PROTOCOL

[mattrobenolt]

I'm opening this to discuss the merits and reasonings for setting HTTP_VERSION inside of the different rack handlers.

See my "Notes" under #969 for a little bit of context.

tl;dr is in all of the handlers, rack does this little bit:

env[HTTP_VERSION] ||= env[SERVER_PROTOCOL]

This causes an environment variable to get set to for HTTP_VERSION, when according to the CGI spec, HTTP_* is explicitly reserved for client headers. This makes it confusing if something downstream is attempting to collect all client headers, this becomes ambiguous. Did the client sent a Version header? Or is this the Version header that Rack is declaring?

This behavior is wrong, but I'm not sure of the ramification of simply... stopping doing this. I can't seem to find things that reference this, except the internals that were fixed up in #969, but no idea how common this is downstream to actually use this value.

[ioquatix]

but no idea how common this is downstream to actually use this value.

I don't do this in falcon and I haven't had any complaints.

[jeremyevans]

Rack is still setting HTTP_VERSION in the environment in all handlers. Unicorn sets it and ignores any Version request header submitted. I'm guessing there are some middleware that are relying on HTTP_VERSION instead of using SERVER_PROTOCOL. So this becomes a case of whether the backwards compatibility breakage is worth adding support for the Version request header. I don't have a strong opinion on that, but I would lean toward keeping backwards compatibility.

[ioquatix]

Moving this to v3.0.0 milestone.

[tenderlove]

IMO it's fine to remove this for v3.0.0