On Wed, Aug 07, 2024 at 11:03:18AM +0200, Richard Biener wrote:
> > Richi mentioned on IRC that the non-cleaned up names might make things
> > harder to feed stuff back to the GIMPLE FE, but if so, I think it should be
> > the dumping for GIMPLE FE purposes that cleans those up (but at that point
> > it should also verify if some such cleaned up names don't collide with
> > others and somehow deal with those).
>
> My plan was to accept an additional character in identifiers as
> extension with -fgimple and use that. I think we already accept dollars
> but dots are of course problematic and cannot be used. Replacing
> unwanted chars with $ and pre-existing $ with $$ might work. There's
> not many (ASCII) characters available, @ might be.
Some kind of mangling... ;)
> > --- gcc/gimplify.cc.jj 2024-08-05 13:04:53.903116091 +0200
> > +++ gcc/gimplify.cc 2024-08-06 15:27:40.404865291 +0200
> > @@ -5599,6 +5599,18 @@ gimplify_init_constructor (tree *expr_p,
> >
> > DECL_INITIAL (object) = ctor;
> > TREE_STATIC (object) = 1;
> > + if (DECL_NAME (object) && DECL_NAMELESS (object))
> > + {
> > + const char *name = get_name (object);
> > + char *sname = ASTRDUP (name);
> > + clean_symbol_name (sname);
> > + /* If there are any undesirable characters in DECL_NAMELESS
> > + name, just fall back to C.nnn name, we must ensure e.g.
> > + SRA created names with DECL_UIDs don't make it into
> > + assembly. */
> > + if (strcmp (name, sname))
> > + DECL_NAME (object) = NULL_TREE;
> > + }
>
> Did you actually run into such a case?
No, but I haven't collected statistics about that (so didn't add
some fprintf there to log if it happened).
> I'd expect
> gimplify_init_constructor only happening on the original GENERIC IL.
In theory one can construct INIT_EXPR and gimplify it in later passes as
well. But it is unlikely, sure.
> In any case how about the simpler
>
> if (!DECL_NAME (object) || DECL_NAMELESS (object))
> DECL_NAME (object) = create_tmp_var_name ("C");
>
> ?
This is what I had in my first version, but I was worried too many things
would be DECL_NAMELESS during gimplification. I think I'll do another
bootstrap/regtest with statistics gathering to see what is worth.
> Otherwise looks OK to me. I guess this should get quite some
> soaking time before backporting (if that was intended).
Definitely. At least two months I'd say.
Jakub
