Tue Jan 22 12:20:42 PST 2008
instruction causes the problem. Fortunately ret <n> is generated
in very few circumstances, e.g. try --- finally() and some kinds
of returns from syncronized blocks. The specific problem goes away
when you recode your source to eliminate the ret <n> instruction.
If this is enough info for someone else, thats good. However, I will
fix it once I find my JVM book that some one has swiped (what timing!).
The good news is after that fix, my app began to run.
The bad news is that once it did hit a reasonable load (50 simultaneous
logins) it crashed miserable, with no useful stack trace and
several "bad file number" exceptions ... the JIT version does not
suffer from that error.
The Solaris THREADSWITCH problem is one I had reported to Tim,
so lets see if the recently posted fix solves it.
> Frank Mehnert
> email: Frank.Mehnert at mchp.siemens.de
More information about the kaffe