On Fri, Feb 25, 2022 at 11:12:13AM +0100, Richard Biener wrote:
> On Fri, 25 Feb 2022, Jakub Jelinek wrote:
>
> > On Fri, Feb 25, 2022 at 10:31:50AM +0100, Richard Biener wrote:
> > > I think it's used as fallback for UNKNOWN_LOCATION, but if we "invent"
> > > a creative location for the artificial stmt it will of course
> > > affect other stmts/expressions using that location.
> > >
> > > > I think it will work.
> > >
> > > Yes, I think so. OTOH the uninit pass does
> > >
> > > /* Avoid warning if we've already done so or if the warning has been
> > > suppressed. */
> > > if (((warning_suppressed_p (context, OPT_Wuninitialized)
> > > || (gimple_assign_single_p (context)
> > > && get_no_uninit_warning (gimple_assign_rhs1 (context)))))
> > > || (var && get_no_uninit_warning (var))
> > > || (var_name_str
> > > && warning_suppressed_p (var_def_stmt, OPT_Wuninitialized)))
> > > return;
> > >
> > > that's a mightly complicated way to test and I'm not sure we get
> > > to the bit on the stmt reliably. So maybe TREE_NO_WARNING on the
> > > reference (or making sure it has UNKNOWN_LOCATION and using
> > > suppress_warning on it) is a better idea after all...
> >
> > So
> > SET_EXPR_LOCATION (tmp_dst, UNKNOWN_LOCATION);
> > suppress_warning (tmp_dst, OPT_Wuninitialized);
> > with a comment explaing why we do that?
>
> Yes. Depending on the context tmp_dst will already reliably have
> UNKNOWN_LOCATION? If tmp_dst might be a plain decl we have to
> be more careful of course.
tmp_dst is:
tree dst = build2_loc (buf->loc, MEM_REF, atype, buf->base,
build_int_cst (buf->alias_type, off));
...
tmp_dst = unshare_expr (dst);
We could build2 the dst and set its location only after unshare_expr
too, but I guess clearing it with a comment on why we do that is better.
Jakub
