On 14 Oct 2006, at 12:05, Edwin Steiner wrote:
On Fri, Oct 13, 2006 at 10:28:04PM +0930, Trevor Glen wrote:
Hi All,
I've been trying to get Java 1.5 running on a hppa box running
ubuntu.
So far I haven't had much luck.
Currently, I'm seeing the behaviour as described in
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28652
Has anyone had any luck with a 1.5 vm and/or classpath on a hppa
linux box?
Hi!
Hm, we really should get my paperwork done...
I had a look at this, and at least the reported Exception is a pure
Java
problem. (/me, vm developer, relaxes with a sly grin ;)
I attach three patches which (on top of each other) make JBoss start
without exceptions with CACAO. Unfortunately the Web console applet
still does not work.
Some comments about the patches: I actually think that in some of
these
cases the callers (JBoss and Tomcat code) make invalid assumptions:
ObjectName.toString() is assumed to return the canonical name by
Tomcat, which is not specified according to what I found.
javax.management.modelmbean.ModelMBeanInfoSupport assumes that
all kinds of arrays retrieved from its superclass can be cast to
a specific derived type. I saw that many places in GNU
Classpath use
the "new array - arraycopy" idiom to copy arrays, which
probably is
the right thing, as you end up with the most general array type
that
makes sense. JBoss code in this case assumes that array
_cloning_ is
used in the superclass constructor, so the more derived type of
the
arrays is preserved. Aaaargh! How much I like C! ;)
The third patch fixes some missing initializations, and a cast in
getKeyPropertyList which I'm not sure how to get around.
CAUTION! Take all these patches with a heap of salt. I don't know what
I'm doing in Java. ;)
-Edwin
<DIFF-1>
<DIFF-2>
<DIFF-3>
Patches one and two look fine on inspection; if they make JBoss work,
so much the better :)
I've only tested this stuff with my own test cases rather than an
full application, so your feedback
is much appreciated.
Three should also be fine. The initialisations look okay (we caught
one like that earlier this week), but you
should probably add 'final' to them now they are initialised early,
to increase efficiency. The worrying part would appear to be
the Hashmap as you are now returning a modifiable copy, but, on
rereading the documentation, it seems that this is fine. The main
issue is whether the internal map may be modified, which it can't
with a new copy.
Mark, if everything is okay paperwork-wise, can this be committed for
0.93?
Sorry for the delay in replying. If you find further issues with the
management code, feel free to CC me.
Thanks,
--
Andrew :)
