Re: Sick macros (was: Action, not words (Re: bug (?) in 3.0-pre1))

["Bart Schaefer" <schaefer@candle.brasslantern.com>]

On Jul 2,  8:24am, Zefram wrote:
} Subject: Sick macros (was: Action, not words (Re: bug (?) in 3.0-pre1))
}
} Bart wrote:
} >2.  Introduce macros for heapalloc/permalloc that start with an open
} >    brace, and a macro for lastalloc that ends with a close brace, so
} >    it's syntatically required that every heapalloc/permalloc have a
} >    matching lastalloc.
} 
} Probably a good idea, but as these macros now introduce new syntax
} let's make them *look* like syntax, rather than function calls.

I would normally have done so, but I was trying to make the change as
minimal as possible.

If that's not a concern, I'd probably do something like this:

#define ALLOC_BEGIN	do { int nonlocal_useheap = useheap; 0
#define ALLOC_RESTORE	do { \
			    if (nonlocal_useheap) global_heapalloc(); \
			    else global_permalloc(); \
			} while (0)
#define ALLOC_END	ALLOC_RESTORE; } while (0)

#define HEAPALLOC	ALLOC_BEGIN; global_heapalloc(); do
#define PERMALLOC	ALLOC_BEGIN; global_permalloc(); do
#define LASTALLOC	while (0); ALLOC_END

And then I'd write:

	HEAPALLOC {
	    /* ... stuff ... */
	} LASTALLOC;

That makes the block context obvious.  The macros are upper-case because
I don't like introducing new "keywords".

However, the above implies changing indentation and all sorts of stuff.
I didn't want the patch to come out that large.

So it could be:

#define HEAPALLOC	ALLOC_BEGIN; global_heapalloc()
#define PERMALLOC	ALLOC_BEGIN; global_permalloc()
#define LASTALLOC	ALLOC_END

Then write:

	HEAPALLOC;
	/* ... stuff ... */
	LASTALLOC;

} :+ # define lastalloc_return \
} :+ 			if (nonlocal_useheap) global_heapalloc(); \
} :+ 			else global_permalloc(); \
} :+ 			return

That's not in my original patch; Zoltan must have come up with it ...

} This won't work as the body of an if or while.  The following is always
} safe:
} 
} #define lastalloc_return \
}   if( (nonlocal_useheap ? global_heapalloc() : global_permalloc()) , 0 ) \
}     ; \
}   else \
}     return

Hmm ... I seem to recall that some compilers don't like having void
expressions (e.g. (?:) where both branches are void functions) used
anywhere in a comma-expression.  In particular, I think AIX either
rejects this or compiles it wrong.  However, I could be confusing that
with something else ... I know for a fact that AIX could not handle
an `if (x,y)' construct we used in zmail, forcing us to rewrite it
as `if (x?0:y)' -- which doesn't work in general, it just happened
that for us x was always 0 to begin with.

If I'm confused and the comma-expression above turns out to be OK, I'd
change ALLOC_RESTORE and add LASTALLOC_RETURN:

#define ALLOC_RESTORE \
	    ((nonlocal_useheap ? global_heapalloc() : global_permalloc()), 0)
#define LASTALLOC_RETURN	if (ALLOC_RESTORE); else return

Otherwise, I'd just leave it up to the programmer to write:

	HEAPALLOC;
	/* ... stuff ... */
	if (getmeoutofhere()) {
	    ALLOC_RESTORE;
	    return(-1);
	}
	/* ... more stuff ... */
	LASTALLOC;



-- 
Bart Schaefer                             Brass Lantern Enterprises
http://www.well.com/user/barts            http://www.nbn.com/people/lantern

New male in /home/schaefer:
>N  2 Justin William Schaefer  Sat May 11 03:43  53/4040  "Happy Birthday"