public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug c++/46043] New: attribute to mark virtual methods that can't be further overriden so they can be devirtualized
@ 2010-10-16 12:04 zsojka at seznam dot cz
  2010-10-16 14:01 ` [Bug c++/46043] " redi at gcc dot gnu.org
                   ` (7 more replies)
  0 siblings, 8 replies; 9+ messages in thread
From: zsojka at seznam dot cz @ 2010-10-16 12:04 UTC (permalink / raw)
  To: gcc-bugs

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46043

           Summary: attribute to mark virtual methods that can't be
                    further overriden so they can be devirtualized
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: zsojka@seznam.cz


In some scenarios, there is one abstract class declaring virtual methods with
several derived classes, all of them overriding those methods. (at least in
OpenTTD that happens often) The important point is that there are no further
derived classes from those. For example:

struct A {
  virtual void f1() = 0;
  virtual void f2() = 0;
  virtual void f3() { ... this->f1(); this->f2(); ... }
};

struct B : A {
  virtual void f1();
  virtual void f2() { ... this->f1(); ... }
};

struct C : A {
  virtual void f1();
  virtual void f2();
  virtual void f3() { ... this->f1(); this->f2(); ... }
};

struct D : A {
  virtual void f1();
  virtual void f2();
};

If there was an information that there are no classes derived from B,C,D, it
would allow many devirtualizations.
One idea I could come with would be omitting the "virtual" keyword causing the
method to lose its "virtual" status. But that would break a lot of existing
code.
Second idea is using an attribute for that. While searching for this issue I
found java uses "final" keyword for this purpose (or at least it seems to me
so), so I will use attribute "final" in the following example:

struct A {
    virtual void f1();
    virtual void f2();
    virtual void f3();
};

/* f1() can't be further overriden */
struct B : A {
    virtual void f1() __attribute__((final));
};

/* no virtual method can further be overriden */
struct C : A {
    virtual void f3();
} __attribute__((final));

struct D : C {
    virtual void f4();
};

struct E : D {
} __attribute__((final));

void foo(struct C *c, struct D *d, struct E *e)
{
    c->f1(); /* c->B::f1() */
    c->f2(); /* c->A::f2() */
    c->f3(); /* c->C::f3() */
    d->f1(); /* d->B::f1() */
    d->f4(); /* not known at compile-time */
    e->f1(); /* e->B::f1() */
    e->f4(); /* e->D::f4() */
}

/* The following results in compile errors */
struct F : D {
    virtual void f1(); /* error */
    virtual void f2(); /* error */
    virtual void f3(); /* error */
    virtual void f4();
};

I searched the bugzilla for this issue and found none open. Given this idea is
quite simple and the gain can be high, there might be something I overlooked.


^ permalink raw reply	[flat|nested] 9+ messages in thread
end of thread, other threads:[~2011-07-17 13:12 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2010-10-16 12:04 [Bug c++/46043] New: attribute to mark virtual methods that can't be further overriden so they can be devirtualized zsojka at seznam dot cz
2010-10-16 14:01 ` [Bug c++/46043] " redi at gcc dot gnu.org
2010-10-16 15:46 ` redi at gcc dot gnu.org
2010-10-16 15:50 ` zsojka at seznam dot cz
2010-10-27 22:12 ` pinskia at gcc dot gnu.org
2010-10-28  8:50 ` redi at gcc dot gnu.org
2011-07-07 22:02 ` jason at gcc dot gnu.org
2011-07-17 11:46 ` paolo.carlini at oracle dot com
2011-07-17 13:12 ` jason at gcc dot gnu.org

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).