String#slice on strings with multibyte chars fails

Could you attach the test case as a file, rather than as inline code?

Also, could you test it on the master, since there have been a lot of fixes involving encoding stuff there?

Thank you.

The described behaviour is still present on master. Moreover, evaluating the code from the file in IRB does not give quite the same results (see comments in code).

Behavior comparison between JRuby master and Ruby 1.9.3:

system ~/projects/jruby $ jruby ~/Downloads/slice_bug.rb 
> 'å'.slice(0,16)
"å"
> 'å'.slice(0,17)
"å"
> 'å'.slice(1,16)
""
> 'å'.slice(1,17)
Java::JavaLang::ArrayIndexOutOfBoundsException
> '1234567890å'.slice(0,17)
"1234567890å"
> '1234567890å'.slice(1,17)
Java::JavaLang::ArrayIndexOutOfBoundsException
> '1234567890å'.slice(9,17)
Java::JavaLang::ArrayIndexOutOfBoundsException

system ~/projects/jruby $ ruby-1.9.3 ~/Downloads/slice_bug.rb 
> 'å'.slice(0,16)
"å"
> 'å'.slice(0,17)
"å"
> 'å'.slice(1,16)
""
> 'å'.slice(1,17)
""
> '1234567890å'.slice(0,17)
"1234567890å"
> '1234567890å'.slice(1,17)
"234567890å"
> '1234567890å'.slice(9,17)
"0å"

I have a fix. String#slice did not, for the path that led to multibyte slicing, have a guard to ensure lengths outside the actual string length were truncated. Adding a simple check fixes the cases provided.

The behavioral difference in IRB is likely due to strings parsed in IRB having different sizes for their backing store. Where the JRuby compiler (used on scripts run directly) will create strings of exactly the length in code, the JRuby interpreter and IRB parser will often have some room around the edges (since they'll be slicing off chunks of a larger line or the entire file in most cases).