Enough is enough... truce time. Re: kernel blah blah blah

[Paul Derbyshire <pderbysh at usa dot net>]

Ohh....my...GOD.

*referee whistle blowing sound*

HEY. Lets have a truce here... no more mudslinging and general aggravated
shouting. I'd think that's beneath the station of the likes of Linus
Torvalds and the GCC/EGCS development gurus.

In "A Final Solution." I suggested some ways to change the compiler or
Linux or both to meet everyone's needs and leave everyone happy.

As for who's wrong, I don't think anyone is. There has been a
*misunderstanding* about the ambiguous phrase "like a macro" in the
documentation. Some took this fairly literally regarding faithfulness of it
being expanded inline; others took it to mean "like a macro when it is
inlinable". Chalk it up to ambiguous documentation, and let's bury the
hatchet. And to reiterate, there are 3 things that can be done.

1. Make the dox clearer about "extern inline".
2. Linus, if worse comes to worst, you may just have to replace one-
   off inlines with macros, or at least use -Winline and jiggle it
   until it inlines. This strikes me as rather unsatisfactory
   however.
3. EGCS developers, if "extern inline" can't be and wasn't meant to
   be what Linus had in mind after reading the dox for it, I suggest
   adding an __attribute__ ((always)) that causes an error if the
   function is not inlinable or a label address from a label inside
   the function is used outside it and an __attribute__ ((once)) that
   inlines a function where it's used, making an error if it's used
   more than once; this means that the compiler can inline the
   functions with these attributes without any worry that it will
   lead to ambiguous references to addressed labels.

-- 
   .*.  "Clouds are not spheres, mountains are not cones, coastlines are not
-()  <  circles, and bark is not smooth, nor does lightning travel in a
   `*'  straight line."    -------------------------------------------------
        -- B. Mandelbrot  |http://surf.to/pgd.net
_____________________ ____|________     Paul Derbyshire     pderbysh@usa.net
Programmer & Humanist|ICQ: 10423848|