Commits on Sep 27, 2010
  1. JRUBY-5011 (jrubyc --java): 'cannot find symbol' if function def has …

    …more than 3 arguments
    enebo committed Sep 27, 2010
  2. JRUBY-4937: Add RubyFile.getUmaskSafe() to cache/sync and avoid troun…

    …cing umask
    Signed-off-by: Hiro Asari <>
    dekellum committed with headius Jul 15, 2010
  3. Bump versions to 1.5.3

    enebo committed Sep 27, 2010
Commits on Sep 25, 2010
Commits on Sep 21, 2010
Commits on Sep 18, 2010
  1. Fix for JRUBY-4944

    yokolet committed Sep 18, 2010
Commits on Sep 17, 2010
  1. Fix JRUBY-4948.

    yokolet committed Sep 17, 2010
  2. Add a fix for JRUBY-5076

    olabini committed with nahi Sep 13, 2010
Commits on Sep 16, 2010
  1. Fix for JRUBY-5093:

    * Re-poll for thread events before locking the ThreadLibrary-related lock. This helps prevent (but does not guarantee) any timeout event from getting "lost" between a poll and the eventual locking.
    * Move RubyThread.interrupt out of the RubyThread.receiveMail synchronized section, since it doesn't need synchronization and it may need to acquire a lock to notify the currentWaitObject. Locking against RubyThread and then currentWaitObject lead to deadlocks in our builds (and probably in real code too).
    headius committed Sep 16, 2010
  2. Hack to work around Scala shipping classes that break Class.getDeclar…

    …edClasses. JRUBY-5092.
    headius committed Sep 16, 2010
Commits on Sep 8, 2010
  1. Selector pool implementation.

    The fixes for JRuby 1.5.2 involved avoiding a per-thread selector in favor of
    always opening and closing the selector around blocking operations. This
    worked for Linux and BSD, where selector lifecycle is fairly clean. But on
    Windows, selector closing involves shutting down a socket, which goes into
    TIME_WAIT state for some amount of time. As a result, heavy IO operations on
    Windows could saturate the max number of socket connections for a given
    process, preventing new selectors from being opened. This fix introduces a
    soft-referenced pool of selector objects. There will only be as many live
    Selector objects as there have been recent concurrent selects, and they will
    age out on a soft-reference lifecycle, which ultimately means unused
    selectors will be dereferenced and eventually finalized at GC boundaries.
    headius committed Sep 3, 2010
  2. Revert "Fix JRUBY-5052: Upgrade to the latest version (1.6.1) of Joda…

    … time"
    This reverts commit 4f2a7e4.
    JodaTime was merged by mistake to jruby-1_5. This is too big a change for a minor release, so we have opted to revert it.
    headius committed Sep 8, 2010
Commits on Sep 2, 2010
Commits on Aug 27, 2010
  1. Not particularly pretty fix for potential NPE if there's an error get…

    …ting selector or setting to non-blocking. This should have thrown at least an IOError before, but now it can throw NPE. Better resource cleanup code here would be really nice.
    headius committed Aug 27, 2010
Commits on Aug 20, 2010
  1. Bump 1.5 branch to upcoming 1.5.2

    enebo committed Aug 20, 2010
Commits on Aug 18, 2010
  1. Fix for JRUBY-5021: imap-over-ssl connections left to GC do not clean…

    … up completely
    headius committed Aug 18, 2010
Commits on Aug 16, 2010
  1. Revert "JRUBY-5007: add 'charset' parameter to mediatype of WEBrick h…

    …ttp server's default error page."
    CRuby's release manager decided to apply the original patch which is
    differ from Apple and RedHat published.
    I'll revert this and apply the following fix by CRuby development team.
    nahi committed Aug 16, 2010
Commits on Aug 12, 2010
  1. JRUBY-5007: add 'charset' parameter to mediatype of WEBrick http serv…

    …er's default error page.
    mediatype should have 'charset' parameter as a conventional manner.
    nahi committed Aug 12, 2010
Commits on Aug 7, 2010
Commits on Aug 3, 2010
  1. Fix for JRUBY-4767: JRuby and open-uri File handle issue

    The fix here is to no longer hold a hard reference to a Selector in each RubyThread, since on some environments (like Tomcat) RubyThread objects can build up a bit. Even a modest number of RubyThread objects will eat 3*N file descriptors for epoll selection, which quickly hits the ulimits on a typical Linux box. By opening and closing them for each select, we pay a small per-select hit, but it fixes the problem.
    headius committed Aug 3, 2010
Commits on Jun 6, 2010
  1. Update maven poms to 1.5.1

    enebo committed Jun 6, 2010
Commits on Jun 4, 2010
  1. Fix for JRUBY-4825: __FILE__ is not expanded when it is used from wit…

    …hin a a compiled ruby script used as the main class of an executable jar
    headius committed Jun 4, 2010
  2. Probable fix for JRUBY-4831: Multiple block args to ActiveRecord asso…

    …ciation proxy are splatted incorrectly
    headius committed Jun 4, 2010
Commits on Jun 3, 2010