From: Brent Roman <brent@...>
Date: 2009-03-25T11:15:13+09:00
Subject: [ruby-core:23017] MBARI patches for 1.8 HEAD


Hi,

I recently pushed the  MBARI patches to my repo on github.
They are finer grained than the original MBARI patch sequence.
Patches that add features are separate from those that fix bugs and/or
improve performance.
All API changes are conditionally compiled on MBARI_API, which can be set
via
the --enable-mbari-api configure parameter.  Or, you could choose not to
apply
those particular patches.

Anyone interested can fetch these from here:

git://github.com/brentr/matzruby.git

Each patch is also available a web interface for reveiw, here:

http://github.com/brentr/matzruby/commits/ruby_1_8-mbari/

Questions/Comments:

I did notice that the new thread anchor code is slowing down my, rather
extreme, threading test cases.
I think I know why and will work on improving this.
Performance on these tests is still significantly better than unpatched.

The patched version, rebased to the ruby_1_8 I pulled this evening, runs the
test suite without error.
Run time when stack clearing is disabled is slightly faster.  With stack
clearing enabled, its slightly
slower, but uses less memory.

Note that there is one test that does fail in all MBARI patched versions. 
The file:
test/ruby/suicide.rb does this:

 at_exit{Process.kill(:INT, $$)}

This signal is not trapped because it is the very last thing eval'd.  The
process terminates before
the trap is executed.  I believe this is inherently a race condition.  One
could CHECKINTS
more often, but that slows things down for no good reason.  So, I changed
the test to:

 at_exit{Process.kill(:INT, $$); sleep 0}  #to keep the interpreter alive
long enough to get the signal.

- brent


-- 
View this message in context: http://www.nabble.com/MBARI-patches-for-1.8-HEAD-tp22694009p22694009.html
Sent from the ruby-core mailing list archive at Nabble.com.