Re: [DotGNU]c compiler: Problems with constants

[Glenn Chambers]

On Tuesday, March 23, 2004, at 08:56 AM, Michal Moskal wrote:

> On Tue, Mar 23, 2004 at 07:38:10PM +1000, Rhys Weatherley wrote:
> > Traditional ANSI C only allows declarations at the head of a block.  
> > C++
> > allows them anywhere in a block.  The latest version of gcc follows 
> > the C++
> > model.  Until recently, such code was considered invalid in C.
>
> I believe this is C99 model, not C++ one.

It's also the C++ model.  Rhys' comment "Until recently" refers to the 
C99
change.  I believe that C++ did it first, and C99 added it later.

(Rhys again)
> > I personally think that it *should* be invalid because it causes 
> > sloppy coders
> > to write C code that won't compile on most of the world's existing C
> > compilers, harming code portability in the process.  This is why I do 
> > most of
> > my development with gcc 2.x, not gcc 3.x - the resulting C code is 
> > more
> > portable.

Since there's no significant C++ element to PNET, this is a reasonable 
position.
(The 3.x C++ compilers have a *vastly* more complete C++ library 
implementation.)

You should probably do a test build every once in a while with a recent 
3.x
just to be sure; the newer compilers have tightened up certain 
restrictions
in the language. (see the flamage on 'strict type aliasing checks' on 
the
GCC lists about a year or so back.)

(Michal this time)

> <flame>Another thing to think about is that C99 is ergh.. 5 years old,
> so maybe there is little point in supporting ancient compilers 
> :-)</flame>

GCC only recently gave up on being buildable with a K&R compiler, and 
those
were 'obsolete' when some of the people on this list were born...

Glenn Chambers
Toledo, OH