C constant expressions proposals

["Joseph S. Myers" <jsm at polyomino dot org dot uk>]

I have now prepared detailed semi-formal models of the interpretation of 
the C constant expression rules, and their extension to GNU C, along with 
ordinary English text descriptions of those extensions to constant 
expressions that are de facto present in GNU C (as part of the present 
"whatever the compiler of the day happens to fold to a constant is 
constant") and which seem likely to have been widely used or the 
preservation of which has been specifically requested in past discussions 
of constant expressions.  Also included are some notes on the benefits 
naturally associated with implementation of well-defined constant 
expression rules, apart from better conformance.

http://www.srcf.ucam.org/~jsm28/gcc/#const

Feedback welcome on these models; the (newly revised and expanded) 
discussion of problems with constant expression rules in C99; the 
particular extensions that should be accepted in GNU C; and anything else 
about these proposals.

These proposed models would render invalid the much-discussed 
gcc.c-torture/compile/20010327-1.c when pointers and longs are different 
widths.  I don't have a plausible description in language standard terms 
of what cases should be acceptable otherwise.  Note that very similar to 
the case

static unsigned long x = (unsigned long) &_text - 0x10000000L - 1;

of that testcase, you could force either sign or zero extension of the 
pointer plus an arbitary offset, e.g.

static unsigned long x = (unsigned long) ((uintptr_t)&_text + 123) + 456;
static long y = (long) ((intptr_t)((uintptr_t)&_text + 123)) + 456;

all of which look like extending an address constant to a wider type, then 
adding a constant, but I don't know if all these cases will work when one 
does.

Relevant past discussions to this testcase and constant expressions 
include the following.  (Representative messages from each thread only 
included; broken mailers may mean not everything is linked in to the 
thread; and in some cases threads may cross month boundaries.)

http://gcc.gnu.org/ml/gcc-patches/2000-11/msg00678.html
http://gcc.gnu.org/ml/gcc-patches/2001-11/msg01191.html
http://gcc.gnu.org/ml/gcc-patches/2003-02/msg01153.html
http://gcc.gnu.org/ml/gcc-patches/2003-03/msg00023.html

-- 
Joseph S. Myers               http://www.srcf.ucam.org/~jsm28/gcc/
    jsm@polyomino.org.uk (personal mail)
    jsm28@gcc.gnu.org (Bugzilla assignments and CCs)