Re: Is Linux Unix?

[Roel Schroeven <rschroev_nospam_ml@fastmail.fm>]

Ryo Furue wrote:
> Roel Schroeven <rschroev_nospam_ml@fastmail.fm> wrote in message news:<2oUpO-4mC-5@gated-at.bofh.it>...
>
> > Jeronimo Pellegrini wrote:
>
> [...]
>
> > > Actually, the first version works -- and both C and C++ (tested here with
> > > gcc -- not sure it it became a standard or not)) will dynamically allocate
> > > memory for you. Try this:
> > >
> > >        int size1, size2, i, j;
> > >        cin >> size1;
> > >        cin >> size2;
> > >        double vec[size1][size2];
> > >
> > > And enter something like "1000" and "1000" when asked for the array dimensions.
> > > It compiles, runs, and does not segfault! :-)
> >
> > Your program segfaults on cygwin (using gcc 3.3.1) when using 1000 and 
> > 1000. On my sarge installation (gcc 3.3.4) it seems to work when using 
> > 1000 and 1000, but segfaults when using 1000 and 10000. Either you were 
> > lucky, or it's a feature that's introduced in more recent versions of gcc.
>
>
> There is a posibility that you run out of stackspace.  I guess these
> dynamically sized arrays are allocated on the stack.  You might want to
> try setting the stacksize to "unlimited".

Sounds plausible. double vec[size1][size2] with size1=1000 and 
size2=10000 allocates 80 MB and with size2=1000 still allocates 8 MB, 
which I imagine can be quite hard on the stack.

> I've found that gcc has implemented this at least for seven years or so.
> If I remember correctly I found a news article in the google groups dated
> in 1996 which mentions this feature of gcc.
>
> Also, I found that the C99 standard incorporated this feature:
>
>    http://home.tiscalinet.ch/t_wolf/tw/c/c9x_changes.html

Cool, I didn't know that.

> It's interesting that some of the new features are (or seem to me to be)
> mainly for numerical computations, such as the "restrict" keyword and
> the builtin complex numbers.

Possibly, but isn't this feature's usefullness in numerical computations 
quite limited if you're restricted to the size of the stack?

> I suspect they are influenced by Fortran.
> And it's amusing to see that C99 has *at last* caught up with Fortran 77
> (partly with Fortran 90/95) in terms of dynamically sized multidimensional
> arrays, complex arithmetics, and "restrict"ed pointers, although C99 lags
> far behind Fortran 90/95 in terms of intrinsic array operations. ;-)

Yeah well, it's not a coincidence that C is often called 'portable 
assembler'.

--
"Codito ergo sum"
Roel Schroeven