Re: [PATCH 1/3] c++: Track lifetimes in constant evaluation [PR70331, PR96630, PR98675]

[Nathaniel Shead <nathanieloshead@gmail.com>]

On Tue, Mar 28, 2023 at 10:50 PM Jakub Jelinek <jakub@redhat.com> wrote:
>
> On Tue, Mar 28, 2023 at 10:32:28PM +1100, Nathaniel Shead via Gcc-patches wrote:
> >       * tree-core.h (struct tree_decl_common): New flag to check if
> >         value lifetime has expired.
> >       * tree.h (DECL_EXPIRED): Access the new flag.
> >       * print-tree.cc (print_node): Print the new flag.
> > --- a/gcc/tree-core.h
> > +++ b/gcc/tree-core.h
> > @@ -1834,7 +1834,10 @@ struct GTY(()) tree_decl_common {
> >    /* In FIELD_DECL, this is DECL_NOT_FLEXARRAY.  */
> >    unsigned int decl_not_flexarray : 1;
> >
> > -  /* 13 bits unused.  */
> > +  /* In VAR_DECL, PARM_DECL, or RESULT_DECL, this is DECL_EXPIRED.  */
> > +  unsigned int decl_expired_flag : 1;
> > +
> > +  /* 12 bits unused.  */
> >
> >    /* UID for points-to sets, stable over copying from inlining.  */
> >    unsigned int pt_uid;
> > --- a/gcc/tree.h
> > +++ b/gcc/tree.h
> > @@ -2697,6 +2697,12 @@ extern tree vector_element_bits_tree (const_tree);
> >     ??? Need to figure out some way to check this isn't a PARM_DECL.  */
> >  #define DECL_INITIAL(NODE) (DECL_COMMON_CHECK (NODE)->decl_common.initial)
> >
> > +/* Used in a VAR_DECL, PARM_DECL, or RESULT_DECL to indicate whether
> > +   this declaration is currently in lifetime for constant evaluation
> > +   purposes.  */
> > +#define DECL_EXPIRED(NODE) \
> > +  (DECL_COMMON_CHECK(NODE)->decl_common.decl_expired_flag)
> > +
> >  /* Holds the size of the datum, in bits, as a tree expression.
> >     Need not be constant and may be null.  May be less than TYPE_SIZE
> >     for a C++ FIELD_DECL representing a base class subobject with its
>
> While it is true that tree_decl_common has some spare bits, this support
> is for an implementation detail of the C++ FE and as such, I think it is
> highly preferred to use some bit in the lang specific data structures
> rather than wasting bits that we could need later on for things that will
> be needed in the middle-end.
> struct lang_decl_base I think also has 12 spare bits...
>
> Rest I'll defer to C++ maintainers (also whether this is appropriate now
> or should be deferred for GCC 14 stage1).
>
>         Jakub
>

Thanks, I didn't notice that there were lang specific data structures;
I'll take a look at using that instead.

From my point of view I was expecting this would be too late for GCC13
(especially since it should only affect already invalid code, and the
risk of false positives), but I'm happy either way.