Using jvm111_18n and HotJava1.0_19 on Solaris Intel or SPARC,
JVM goes into infinite loop by operation described below.
LANG or codeset converter choice don't have relationship
Way of reproduction:
1. Bring up HotJava
2. Type URL "http://www.systems.co.jp/" and load it.
Wait loading finished.
The page contains Japanese characters so it may be displayed
as garabage. But it doesn't relate to the reported problem so
3. Modify the URL to "http://www.systems.co.jp/g-mall/fujioka/" and
load it. Wait loading finished
4. Press "previous page" button
-> Page is cleared and displaying previous page will
be started. But it won't be finished.
Monitor shows one or two lines and Duke works forever.
"Previous page" button keeps pressed shape.
If the bug is not reproduced (i.e. previous page is displayed successfully),
repeat to press "next page" button and "previous page" button
alternately for 10 to 20 times.
I'll attach two files:
stack traces of threds gatherd by attaching dbx to JVM in loop.
logged by "-log" option of HotJava.
I wrote "it's reproducible on s297_32/Intel".
However, I discovered it's also reproducible
on s297_G1(JVM and HJ replaced with latest) or on SPARC.
So I updated synopsis, description, public summary, and hardware version.
It seems easy to reproduce it on multiprocessor machine.
"easy" means it doesn't require to repeat "previous page"
and "next page". First "previous page" action reproduces the problem.
The 2nd page has been moved on server. Please use this combination:
To reproduce, it sometimes need to repeat "previous page" and "next page",
For case the server is not accecible, I'll attach "systems.tar.Z".
It contains copy of some of the Web contents.
I'll attach 2 more files "dbxwhere.jitoff" and "hjlog.jitoff"
which are trace data when JIT is off.