Feature #8090: resolv.rb checks platform based on RUBY_PLATFORM, which is insufficient for JRuby

• Status: Closed
• Assignee: -
• Target version: 2.6

[headius (Charles Nutter)]

JRuby shares stdlib with MRI, and as a result we've had to patch a number of things. We'd like to get some of these changes back into MRI's copy.

This issue refers to the following check in resolv.rb:

if /mswin|mingw|bccwin/ =~ RUBY_PLATFORM
  require 'win32/resolv'
  DefaultFileName = Win32::Resolv.get_hosts_path
else
  DefaultFileName = '/etc/hosts'
end

Because RUBY_PLATFORM on JRuby is always 'java', this check will use the incorrect "hosts" path on Windows. This was reported as an issue in https://github.com/jruby/jruby/issues/580.

We had patched 1.8 stdlib but not 1.9 stdlib to use the following line for checking platform:

if /mswin|mingw|bccwin/ =~ RUBY_PLATFORM || ::RbConfig::CONFIG['host_os'] =~ /mswin/

I would like to commit this change back to MRI for 1.9.3, 2.0, and 2.1. It does not produce a behavioral change on MRI, since there's no supported platforms that will match /mswin/ for host_os, but it means we won't have to maintain a diff.

[nobu (Nobuyoshi Nakada)]

Why not try to require 'win32/resolv' and rescue?

[headius (Charles Nutter)]

nobu (Nobuyoshi Nakada) wrote:

Why not try to require 'win32/resolv' and rescue?

Yep, that works too. The original logic does not do that, so I just expanded on the original condition. I don't have a strong preference for one way versus the other.

[nobu (Nobuyoshi Nakada)]

=begin
Because we haven't considered the case (({RUBY_PLATFORM})) means somethings different from CPU and OS.

If you check by (({RbConfig})), (({RUBY_PLATFORM})) is useless, so it should be:

require 'rbconfig' # loaded by rubygems in default
if /mswin|mingw|bccwin/ =~ ::RbConfig::CONFIG['target_os']

You should use '((%target_os%))' to see the running platform.
=end

[headius (Charles Nutter)]

I thought host_os is the OS that we're actually running on, and target_os was the OS name Ruby was built for. Was I mistaken?

I like checking RbConfig::CONFIG alone, if we can decide which key to use.

[nobu (Nobuyoshi Nakada)]

"host" has meanings when you build a cross-compiler. You build a
compiler on "build" platform, which runs on "host" platform and
generates code for "target" platform. So we don't need "host" almost,
and often we tweaks "target" in configure.in.

[jonforums (Jon Forums)]

nobu (Nobuyoshi Nakada) wrote:

"host" has meanings when you build a cross-compiler. You build a
compiler on "build" platform, which runs on "host" platform and
generates code for "target" platform. So we don't need "host" almost,
and often we tweaks "target" in configure.in.

I have a different interpretation of the following docs, and also believe host_os, not target_os should be used (at runtime) to determine which OS Ruby is running on.

http://sourceware.org/autobook/autobook/autobook_260.html#SEC260
http://sourceware.org/autobook/autobook/autobook_261.html#SEC261

[akr (Akira Tanaka)]

• Status changed from Open to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r39988.
Charles, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

• lib/resolv.rb: Test Windows platform by detecting LoadError when
  require 'win32/resolv' suggested by Nobuyoshi Nakada [ruby-core:53389].
  [ruby-core:53388] [Feature #8090] Reported by Charles Nutter.