Garbled Gem.user_home with jruby --1.9

ryenus: Thank you for following up!

tested with fresh 1.6.7 release, problem fixed. this should have been in 1.6.7 changelog

What about a test, e.g.: https://gist.github.com/1808755

OK, I merged the pull request. I am not familiar with the release procedure for jcodings, so I'll let Tom and Marcin take care of that. Until the fix is released (I suppose it will be jcodings 1.0.7), the only way to get the fix is to build jruby from source after copying it to build_lib.

[04:28:36 roger@R /forks/jcodings (fix-gbk)]
$ cp target/jcodings.jar /sandbox/jruby/build_lib/jcodings.jar

[04:29:08 roger@R /sandbox/jruby (master)]
$ ant clean jar
Buildfile: E:\sandbox\jruby\build.xml
...
BUILD SUCCESSFUL
Total time: 1 minute 58 seconds

[04:31:12 roger@R /sandbox/jruby (master)]
$ java -jar lib/jruby.jar --1.9 -e "puts File.expand_path '~'"
C:/Users/roger

[04:31:42 roger@R /sandbox/jruby (master)]
$ java -Dfile.encoding=gbk -jar lib/jruby.jar --1.9 -e "puts File.expand_path '~'"
C:/Users/roger

Oh, yeah!

well, that one is probably different, I'm happy to take a look, but definitely later

OK, the bug is in jcodings, see https://github.com/jruby/jcodings/pull/2

ryenus,

Sorry, I skipped over some details of your update above. I will look into this a bit further.

Hmm. This sounds very much like JRUBY-6398. For that bug's fix, I decided to default to ASCII-8BIT if JVM doesn't have a known default Charset. Does that work for you? Can you test one from http://ci.jruby.org/snapshots/release/ ?

Thank you.

I think I've located the problem in JRuby (though very likely an external bug).

In org.jruby.util.OSEnvironment#getAsMapOfRubyStrings, the encoding is retrieved like this: Encoding encoding = runtime.getEncodingService().getLocaleEncoding();

that line of the code is buggy, when java.nio.charset.Charset.defaultCharset().toString() == "GBK", the above code returns CP949Encoding, I suppose it should return GBKEncoding

Specifically, the code 5c for '\' in GBK is different for that in CP949, instead in CP949 it's for the 'WON SIGN' (see here: http://www.fileformat.info/info/charset/x-IBM949/decode.htm, seems CP949 is asymmetric).

I'm just curious, since JVM can correctly report the name of the default charset, why don't we use that to find the correct encoding but use jcoding's code to detect it?

I just tested it against master and jruby-1_6, same result:

$ java -Dfile.encoding=cp936 -jar lib/jruby.jar --1.9 -e "puts RUBY_VERSION, File.expand_path('~')" && git br -v

1.9.3

E:/sandbox/jruby/C:₩Users₩roger

* master      0cd7377 Fix JRUBY-5967

If you can reproduce this with some regularity, can you test it against JRuby master (1.6) or JRuby 1.6.6 (snapshots from http://ci.jruby.org/snapshot). There have been many encoding-related fixes recently that aren't in 1.6.5.

Now the reproducing condition is clear, and it's an encoding related issue.

the issue can be reproduced with jruby -J-Dfile.encoding=cp936 --1.9 -e "puts File.expand_path '~'".

and chcp doesn't help.

I tested again on another box with a fresh downloaded jruby 1.6.5 and cannot reproduce the issue.

It turns out to be a bug in Rubygems, see https://github.com/rubygems/rubygems/pull/229