You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by "Roy T. Fielding" <fi...@kiwi.ICS.UCI.EDU> on 1997/06/16 07:43:15 UTC

Re: general/736: pointers cast as ints of different size - potential memory problems (fwd)

>I'd like to revisit this problem.  When I posted my patch I proposed a new
>type "ptr_int" which is large enough to hold a pointer or an int.  But it
>was vetoed by Roy because it wasn't "clean enough".  I'd like to submit
>that the type "int" which we presently use is not an abstraction at all,
>and is just a convenience and really isn't clean either.

The problem with your patch was that it changed the interface of general
functions just because the *completely internal* data type for storing
the values was changed.  That is the wrong way to go.

The purpose of the functional interface is to save things for later
cleanup.  It should therefore have a separate interface for each type
of thing (modern languages would just use polymorphing, but with C we
just define separate functions).

>The remaining alternative that I can think of is to have a table_do_int and
>a table_do_ptr.  Then we'd need a register_cleanup_int and a
>register_cleanup_ptr.  And more crap like that.

The existing use of the table functions is already broken because features
use it directly instead of through abstract functions.  We already have
note_cleanups_for_file, note_cleanups_for_fd, and note_cleanups_for_socket
as interfaces -- any casting or union of data types should be done inside
those functions.

>See <http://www.arctic.org/~dgaudet/patches/apache-1.2-64bit.patch> for the
>example patch using ptr_int that I posted a while back.  That will give you
>an example of what needs to be changed.

I would still veto that patch -- an fd is never a pointer, so changing the
functional interface for the fd routines is breaking an abstraction for
no good reason.

....Roy

Re: general/736: pointers cast as ints of different size - potential memory problems (fwd)

Posted by Dean Gaudet <dg...@arctic.org>.
I'm not submitting that patch for consideration again, you already
vetoed it.

The functions involved are generally those of the form "here is a
function I want you to call back, and here is a bit of data I want you
to pass to that callback".  In which case calling it generic_data is a
perfect abstraction.

The note_* functions are not callbacks, they reference callbacks.  For
example, using my proposed solution, note_cleanups_for_fd becomes:

void note_cleanups_for_fd (pool *p, ptr_int fd)
{
    generic_data gd;

    GENERIC_INT (gd) = fd;
    register_cleanup (p, gd, fd_cleanup, fd_cleanup);
}

Should I go ahead and generate a patch like this?  I'm not going to do it
if you'll veto it.  I'm not sure I completely understand all your issues
with the problem.

Dean

On Sun, 15 Jun 1997, Roy T. Fielding wrote:

> >I'd like to revisit this problem.  When I posted my patch I proposed a new
> >type "ptr_int" which is large enough to hold a pointer or an int.  But it
> >was vetoed by Roy because it wasn't "clean enough".  I'd like to submit
> >that the type "int" which we presently use is not an abstraction at all,
> >and is just a convenience and really isn't clean either.
> 
> The problem with your patch was that it changed the interface of general
> functions just because the *completely internal* data type for storing
> the values was changed.  That is the wrong way to go.
> 
> The purpose of the functional interface is to save things for later
> cleanup.  It should therefore have a separate interface for each type
> of thing (modern languages would just use polymorphing, but with C we
> just define separate functions).
> 
> >The remaining alternative that I can think of is to have a table_do_int and
> >a table_do_ptr.  Then we'd need a register_cleanup_int and a
> >register_cleanup_ptr.  And more crap like that.
> 
> The existing use of the table functions is already broken because features
> use it directly instead of through abstract functions.  We already have
> note_cleanups_for_file, note_cleanups_for_fd, and note_cleanups_for_socket
> as interfaces -- any casting or union of data types should be done inside
> those functions.
> 
> >See <http://www.arctic.org/~dgaudet/patches/apache-1.2-64bit.patch> for the
> >example patch using ptr_int that I posted a while back.  That will give you
> >an example of what needs to be changed.
> 
> I would still veto that patch -- an fd is never a pointer, so changing the
> functional interface for the fd routines is breaking an abstraction for
> no good reason.
> 
> ....Roy
>