On Tue, 4 Oct 2005, Leopold Toetsch via RT wrote:
> On Oct 4, 2005, at 19:06, Andrew Dougherty wrote:
> > src/inter_create.c:400: warning: dereferencing type-punned pointer
> > will
> > break strict-aliasing rules
>
> The line reads:
>
> LVALUE_CAST(char *, p) += ALIGNED_CTX_SIZE;
>
> The intent is of course, to bump the context pointer by the needed
> size. The problem is that the needed size does not correlate with the
> size of the struct.
>
> Anyway, does:
>
> p = (struct Parrot_Context *) ( (char *) p + ALIGNED_CTX_SIZE );
>
> help, or better is it "more correct"?
While this does indeed replace the warning by a different warning ("cast
increases required alignment of target type"), it doesn't fix the problem
-- parrot still panics. (And since we're not accessing the memory through
a (char *), I'm not sure it should make any difference. I'm not a
language lawyer, and I haven't read the standard closely.)
Anyway, from what I've been able to gather, gcc's warnings on aliasing are
neither complete nor always 100% on-the-mark. In this case, for example,
I took the apparently offending function, moved it to a different file
(inter_create2.c) and recompiled that function with and without
-fno-strict-aliasing. It made no difference. However, recompiling the
remaining functions in inter_create.c with -fno-strict-aliasing *did* make
the problem go away.
So the compiler's doing some optimization somewhere else in the file that
it's not warning about, and that optimization only happens with
-fstrict-aliasing. I suspect by continuing my divide and conquer strategy
on inter_create.c, one could probably isolate it to a single function, and
then, perhaps, understand whether it's a compiler optimizer error or
whether it's a programming error.
Since this can be reproduced with gcc-3.4 on Intel, I'd appreciate it if
someone with a faster machine and/or a deeper understanding of what the
code is actually trying to do could hunt it down.
> We have a lot of similar code e.g. inside GC, where pointers to PMCs or
> to PObjs are icnremented by the actual size of the object and not by
> the size of some structure.
Yes, I know, but I'm not fluent enough with parrot's internals to follow
it well, so I get lost every time I try to dig in deeply.
--
Andy Dougherty [EMAIL PROTECTED]
