URI::Generic.use_proxy? cannot match IPv6 literal addresses in no_proxy

URI::Generic.use_proxy? から始め、issue に示されている no_proxy のパース用正規表現を調べます。提示された Ruby の例で動作を再現し、その後、ベア形式、角括弧で囲まれた形式、ポート付きで角括弧で囲まれた形式、および CIDR 形式の IPv6 エントリが認識されることを確認するとともに、既存のプロキシマッチングが引き続き機能することを確認します。

Summary

URI::Generic.use_proxy? silently ignores IPv6 literal addresses in no_proxy / NO_PROXY, causing IPv6 targets to always be proxied even when they should be excluded.

Root cause

The parsing regex in use_proxy? splits no_proxy entries on colons to extract an optional port:

no_proxy.scan(/([^:,\s]+)(?::(\d+))?/) {|p_host, p_port|

IPv6 addresses contain colons as part of their syntax (e.g. 2001:db8::1), so this regex mangles them; the address is split into fragments rather than parsed as a host. As a result:

• Bare entries like 2001:db8::1 are never correctly captured into p_host
• Bracketed entries like [2001:db8::1] are captured literally (with brackets) and never reach the IPAddr.new(p_host).include?(addr) check since IPAddr.new("[2001:db8::1]") raises InvalidAddressError
• Bracketed entries with port like [2001:db8::1]:8080 are similarly broken

Reproduction

require 'uri'
require 'resolv'

# IPv6 target that should bypass the proxy
no_proxy = "2001:db8::1"
hostname  = "2001:db8::1"
addr      = Resolv.getaddress(hostname) rescue hostname
port      = 80

puts URI::Generic.use_proxy?(hostname, addr, port, no_proxy)
# Expected: false (should bypass proxy)
# Actual:   true  (incorrectly routes through proxy)

Expected behavior

All three real-world no_proxy formats for IPv6 should be recognized: