Relax parsing restrictions around host and hostname

[pvande]

This approach allows us to match and process any potential input and still meet the contractual requirements of Request#split_authority without requiring the input to be well-formed or RFC-compliant. In turn, this eliminates unexpected failures on invalid-but-functional inputs.

If we deemed it necessary to do validation (particularly on IP addresses), the AUTHORITY regex still provides enough context on which validations should apply, though doing so would require some consideration on what should be returned upon validation failure.

If validations can be ruled unnecessary – I suspect this should be the case – the implementation could be replaced by the semantically identical:

def split_authority(authority)
  /\A(?<host>\[\g<addr>\]|(?<addr>.*?))(:(?<port>\d+))?\Z/ =~ authority
  return host, addr, port&.to_i
end

This gives up differentiation between IPv6, IPv4, and DNS addresses, but is arguably simpler.

(Resolves #1604)

[ioquatix]

This looks good to me, you took what I did and made it way better. Thanks!

[ioquatix]

I’ll backport to 2-2-stable as I consider this a bug fix.

[ioquatix]

@pvande what is the difference between \A and \Z vs ^ and $?

[jeremyevans]

\A matches the beginning of the string. ^ matches any beginning of line in the string. \Z matches the end of string, allowing possible newline at the end of the string. $ matches the end of any line in the string.

In general, you are much more likely to want \A and \Z (or \z, which matches the end of string) than ^ and $. Using ^ and $ usually leads to bugs, in my experience.