[rack] use the Rack env instead of the Rack request object

[palazzem]

What it does

Uses directly the env object (we have one env per request) instead of the Rack Request. This is required because in some frameworks (like Rails) the request.path_info could be changed at runtime, especially during exception handling. This makes results not reliable for our http.url tag. For instance:

# suppose we call a Rails endpoint like /error
request = ::Rack::Request.new(env)
assert_equal(request.path_info, '/error')
status, headers, response = @app.call(env)
assert_equal(request.path_info, '/error')  # this fails!

The stored value now is /500.

[ufoot]

Left a comment on REQUEST_URI vs PATH_INFO but can ship as is, no worry.

[ufoot]

I'm OK with this, globally, but on a later release we might dig into it and figure out why a framework would do that, because maybe the PATH_INFO, when it's overridden, is overridden to a value that is technically less related to the "physical" http request but more meaningful in the app context.

[palazzem]

actually PATH_INFO must include the request URL as defined in Rack docs: http://www.rubydoc.info/github/rack/rack/file/SPEC#The_Environment

We may double check on Rails why it's happening. This is just a reasonable default because we know about this exception, but definitely something to take in consideration if it happens also for other frameworks.

[ufoot]

👍 good to go, and anyway, this is orthogonal to / independent from the short-term goal we have: have rack support to avoid loosing traces.