Re: Patch: Serialization mods

[Warren Levy <warrenl at cygnus dot com>]

On Thu, 27 Jul 2000, Tom Tromey wrote:

> Warren> FWIW, there's some code in defineclass.cc that verifies the
> Warren> modifier associated with the object; I didn't add any checking
> Warren> for STRICT in there, but that shouldn't hurt anything right
> Warren> now.
> 
> If this change is correct, please make it now while we're aware of
> it.  If we wait then it mutates into an obscure bug...

Yeah, if I knew the patch was right I'd pop it in there.  But given my
other deadlines, I was hoping not to take the time right now to look into
this (it shouldn't be difficult though).  I took the time to merge one of
the Classpath classes (java/io/ObjectStreamClass.java) to avoid
divergence.  The Classpath version still had some native methods in it and
I knew the libgcj version was working so I just substituted the native
methods out.  

java/io/ObjectOutputStream.java also has native methods while libgjc's
doesn't, but I haven't fully verified that the libgcj version is fully
equivalent (and I don't want to introduce bugs if it isn't).  If folks
would rather have me merge this one now at the slight risk of introducing 
a bug into Classpath , I can do that (but I probably should be asking that
of the Classpath list ;-).

> Warren> Libgcj does a much better job now of dealing with
> Warren> serialVersionUID and most classes now match the JDK as they
> Warren> should.
> 
> This is really cool.

Thanks.  It feels really good when long strings of seemingly random
numbers start matching ;-).  Trying to find the problems has been a big
pain.  The Mauve tests still show some problems but things are *much*
better now.
--warrenl