Re: http://gcc.gnu.org/java/done.html

[Andrew Haley <aph at redhat dot com>]

Erik Poupaert writes:
 > 
 > > Erik Poupaert writes:
 > >  > 
 > >  > http://irate.sourceforge.net/index.html
 > >  > 
 > >  > 22 July 2003 - Problems with the server have been sorted. The
 > >  > server has been running as a gcj compiled binary. I've switched to
 > >  > Sun's Java VM and it's running about 100 times faster. I suspect
 > >  > that the problem is garbage collection.
 > > 
 > > I wonder why people always blame the gc?  For what it's worth, I
 > > believe Hans when he says his gc is generally competitive.
 > > 
 > > Anyway, there's not much we can do about reports like this one.  We
 > > need more information.
 
 > Still, the "100 times" is alarming. There is no reason to believe
 > that the reporter, who ran the native and the bytecode versions did
 > not come to this conclusion in good faith.

Oh, of course not.  But with no test case or idea about which gcj
version is being used I can't see what to do.

 > If you or Hans need someone to assist in carrying out the work
 > needed to figure out what's going on, I am available. I am not an
 > expert at all regarding gc, but I also depend on its proper
 > functioning. Chances certainly are that the problem is related to
 > something completely different.

I know we shippped one gcj version that had a horribly broken stack
unwinder.  That could be 100 times slower.

 > That doesn't change anything, however. I feel that early adopters
 > are really important and that I should help with this.

That's great.  A self-contained test case would help.

Andrew.