Re: [Cp-tools-discuss] gnu.bytecode

[Elliott Hughes]

On Dec 29, 2004, at 22:36, C. Brian Jones wrote:

> On Wed, 2004-12-29 at 22:57, Elliott Hughes wrote:
> > is this the right/a sensible place to talk about gnu.bytecode?
> >
> > i know there's *a* copy of gnu.bytecode in the repository, but i know
> > it's not *the* copy. for some reason, it was easier to find this list
> > than any equivalent kawa one.
> >
> > i've been writing a tool similar to javap/jcf-dump, but which outputs
> > something in a form suitable for re-assembly into a .class file. aside
> > from the lack of support for 1.5 changes, i've found that the
> > LocalVariableTable support doesn't work (try
> > gnu.classpath.tools.JavapMain or gnu.bytecode.dump on a class compiled
> > with "javac -g" to see). there's a really simple fix for JavapMain and
> > my disassembler, but i don't know if it'll break kawa or anything else
> > that uses gnu.bytecode.
> >
> > is this where i'd find people who're interested?
>
> I'm interested in fixing the local variable table problem.  I think it
> used to work but definitely doesn't now.  Per might be interested on 
> the
> kawa list.

there are two problems. one is that nothing writes to the start_pc and 
end_pc fields in Variable, so even if you could get hold of a Variable, 
you wouldn't know what bytecode indexes it covers. the other problem -- 
and the reason i didn't use the word 'scope' in the previous sentence 
-- is that the code that reads the LocalVariableTable keeps creating 
new Scope instances but doesn't link them up.

i think the intent was probably that the start_pc and end_pc fields 
would be removed from Variable, and the Variable's Scope's start and 
end Label instances' position fields would be used for that 
information. but i'm not convinced that works in general. nothing i've 
seen says that the range of bytecode indexes for which a slot contains 
a given variable has to correspond to a scope in the usual sense.

my fix doesn't create any new Scope instances, and does assign 
Variable's start_pc and end_pc fields.

> Anyway there is a partial library for disassembly in the
> newest japitools which I've receive permission to extend and dump into
> cp-tools.  I will have to add additional lower level support to it but
> that's okay.  The facets of gnu.bytecode that allow for changing actual
> bytecode and writing class files back out are useless to me.

it's useful to have that functionality, though. an assembler would want 
that. other language implementations want that.

it would be nice to have one library good enough for all these projects 
to share. (and it would be best if it were part of the JDK, so it was 
available everywhere, but that's something we can do little about.)

at the same time, a more direct mapping to/from the class file format 
would be handy for my purposes. and something that's a complete and 
direct implementation of chapter 4, and which makes reference to it, 
would be nice.

--
http://www.jessies.org/~enh/