Bug #9737: Non-ASCII characters in the path to ruby executable break require paths

• Status: Closed
• Assignee: windows
• Target version: -
• ruby -v: ruby 2.2.0dev (2014-04-12 trunk 45576) [i386-mswin32_100]
• Backport: 2.0.0: UNKNOWN, 2.1: UNKNOWN
• Tags: wip-patch, encoding

[spatulasnout (B Kelly)]

Hi,

On Windows, if non-ASCII characters exist the path on which the ruby
interpreter is invoked, the character encoding is not handled properly
when the require paths in $: are created, making ruby unable to require
libraries in its standard lib paths.

In the first two examples, the working directory of the shell is not
a factor; what matters is the path by which the ruby interpreter is
invoked.

Here we invoke ruby using a path containing non-ASCII characters:

$ M:\dev\ruby-build\zz-können2\bin\ruby.exe -v --disable-gems -e "p $:.first.encoding, $:.first; require 'uri'"
ruby 2.2.0dev (2014-04-12 trunk 45576) [i386-mswin32_100]
#<Encoding:ASCII-8BIT>
"M:/dev/ruby-build/zz-k\xF6nnen2/lib/ruby/site_ruby/2.2.0"
-e:1:in `require': cannot load such file -- uri (LoadError)
        from -e:1:in `<main>'

Here we invoke ruby in the same location, but instead using the
NTFS "short" name on the path so that it is ASCII only:

$ M:\dev\ruby-build\ZZ-KNN~2\bin\ruby.exe -v --disable-gems -e "p $:.first.encoding, $:.first; require 'uri'"
ruby 2.2.0dev (2014-04-12 trunk 45576) [i386-mswin32_100]
#<Encoding:IBM437>
"M:/dev/ruby-build/ZZ-KNN~2/lib/ruby/site_ruby/2.2.0"

In the second two examples, we change the working dir instead of
specifying the interpreter path directly. The results are the same:

# working dir: M:\dev\ruby-build\zz-können2\bin
$ .\ruby.exe -v --disable-gems -e "p $:.first.encoding, $:.first; require 'uri'"
ruby 2.2.0dev (2014-04-12 trunk 45576) [i386-mswin32_100]
#<Encoding:ASCII-8BIT>
"M:/dev/ruby-build/zz-k\xF6nnen2/lib/ruby/site_ruby/2.2.0"
-e:1:in `require': cannot load such file -- uri (LoadError)
        from -e:1:in `<main>'

# working dir: M:\dev\ruby-build\ZZ-KNN~2\bin
$ .\ruby.exe -v --disable-gems -e "p $:.first.encoding, $:.first; require 'uri'"
ruby 2.2.0dev (2014-04-12 trunk 45576) [i386-mswin32_100]
#<Encoding:IBM437>
"M:/dev/ruby-build/ZZ-KNN~2/lib/ruby/site_ruby/2.2.0"

I'm guessing this will be related to the GetModuleFileName() call in:

win32/stub.c:    lenexe = (size_t)GetModuleFileName(NULL, exename, sizeof exename);

which would presumably need to become the Unicode GetModuleFileNameW()
version instead?

Regards,

Bill

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)

WIP, no tests yet: https://github.com/nobu/ruby/compare/win32-utf8-loadpath?expand=1

[jeremyevans0 (Jeremy Evans)]

• Status changed from Open to Closed

This issue appears to have been fixed, probably in Ruby 3.0. Using the latest RubyInstaller-head:

D:\zz-können2>D:\zz-können2\Ruby31-x64\bin\ruby.exe -v --disable-gems -e "p $:.first.encoding, $:.first; require 'uri'"
ruby 3.1.0dev (2021-06-27 master e724857f42) [x64-mingw-ucrt]
#<Encoding:IBM437>
"D:/zz-k\x94nnen2/Ruby31-x64/lib/ruby/site_ruby/3.1.0"