On Fri, May 31, 2019 at 9:48 PM Jeff Law <l...@redhat.com> wrote:
>
> On 1/11/19 1:10 PM, Martin Sebor wrote:
> > The attached patch extends the detection of references to static
> > variables in inline functions to the function signatures, including
> > their return type. Since the declaration of a function need not be
> > yet available at the time the static is referenced (e.g., when it's
> > referenced in the functions return type), the patch introduces
> > the concept of "tentative records" of such references and either
> > completes the records when the full declaration of the function,
> > including its definition, is known to be inline, or removes it
> > when it isn't.
> >
> > Martin
> >
> > gcc-88718.diff
> >
> > PR c/88718 - Strange inconsistency between old style and new style
> > definitions of inline functions.
> >
> > gcc/c/ChangeLog:
> >
> > PR c/88718
> > * c-decl.c (reset_inline_statics): New function.
> > (record_inline_static): Optimize.
> > (check_inline_statics): Handle tentative records for inline
> > declarations without definitions.
> > Print static declaration location.
> > (push_file_scope): Clear records of references to statics.
> > (finish_decl): Add tentative records of references to statics.
> > (finish_function): Same.
> > * c-typeck.c (build_external_ref): Handle all references to statics.
> >
> > gcc/testsuite/ChangeLog:
> >
> > PR c/88718
> > * gcc.dg/inline-40.c: New test.
> > * gcc.dg/inline-41.c: New test.
> >
> > Index: gcc/c/c-decl.c
> > ===================================================================
> > --- gcc/c/c-decl.c (revision 267616)
> > +++ gcc/c/c-decl.c (working copy)
> > @@ -826,14 +826,47 @@ c_finish_incomplete_decl (tree decl)
> > }
> > }
> >
> > -/* Record that inline function FUNC contains a reference (location
> > - LOC) to static DECL (file-scope or function-local according to
> > - TYPE). */
> > +/* Free records of references to static variables gathered so far. */
> >
> > +static void
> > +reset_inline_statics (void)
> > +{
> > + if (!c_inline_statics)
> > + return;
> > +
> > + for (c_inline_static *csi = c_inline_statics, *next = csi->next;
> > + csi; csi = next)
> > + ggc_free (csi);
> > +
> > + c_inline_statics = NULL;
> > +}
> > +
> > +/* Record that inline function FUNC either does contain or may contain
> > + a reference (location LOC) to static DECL (file-scope or function-local
> > + according to TYPE). For a null FUNC, a tentative record is created that
> > + reflects a reference in the function signature and that is either
> > updated
> > + or removed when the function declaration is complete. */
> > +
> > void
> > record_inline_static (location_t loc, tree func, tree decl,
> > enum c_inline_static_type type)
> > {
> > + gcc_assert (decl);
> > +
> > + if (c_inline_statics)
> > + {
> > + /* Avoid adding another tentative record for this DECL if one
> > + already exists. */
> > + for (c_inline_static *csi = c_inline_statics; csi; csi = csi->next)
> > + {
> > + if (c_inline_statics->function)
> > + break;
> > +
> > + if (decl == c_inline_statics->static_decl)
> > + return;
> > + }
> > + }
> > +
> I'm going to assume this isn't called enough for the linear search to
> matter from a compile-time standpoint.
Err - but there must be a better way, no? It's trivial to write a testcase
running into quadratic behavior here.
Richard.
>
> OK for the trunk.
>
> Jeff
