Re: [Main] Copy user definitions.

[Robert Anisko]

> >> Hi,
> >>
> >> I send you the patch, which copies the user definitions %{...%} in
> >> the bison.simple. Robert tell me what do you think about ?

I like it, there's a nice memory leak!
When do you free definitions?

> >> Index: src/output.c
> >> ===================================================================
> >> RCS file: /cvsroot/bison/bison/src/output.c,v
> >> retrieving revision 1.44
> >> diff -u -r1.44 output.c
> >> --- src/output.c 2001/08/27 14:55:39 1.44
> >> +++ src/output.c 2001/08/28 10:02:48
> >> @@ -1078,6 +1078,15 @@
> >>    MACRO_INSERT_INT ("verbose", 0);
> >>
> >>    MACRO_INSERT_STRING ("filename", infile);
> >> +  {
> >> +    size_t size = obstack_object_size (&attrs_obstack);
> >> +    char *definitions;
> >> +
> >> +    definitions = XMALLOC (char, size + 1);
> >> +    bcopy (obstack_finish (&attrs_obstack), definitions, size);
> >> +
> >> +    MACRO_INSERT_STRING ("definitions", definitions);
> >> +  }
> >>
> >>    MACRO_INSERT_INT ("nnts", nvars);
> >>    MACRO_INSERT_INT ("nrules", nrules);
> >
> >I do not agree with doing it in prepare ().
>
> This is easy, tell why you are disagree.

I disagree with that too. You do that the quick and
dirty way. I think we should treat definitions
as we treat tables: tables are grown in the output_obstack,
finished as soon as possible, and immediately inserted in the
macro table. I mean, we should not change prepare but
the piece of code that grows attrs_obstack (probably somewhere
in reader).

And I don't understand why do you need to copy the
definitions in a malloc'ed place: you malloc definitions,
you finish attrs_obstack, copy its content to definitions,
and put it back to another obstack. Hey, this is hairy.

You should use the real insertion function, macro_insert,
rather than MACRO_INSERT_STRING, which is a more
specific wrapper.

Why don't you directly grow the definitions in the output_obstack,
and then write: macro_insert ("definitions", obstack_finish
(&output_obstack))?

I think this would more efficient, and more consistent
with the current code.