On Sun, Dec 19, 2004 at 11:31:13AM -0500, Marc Shapiro wrote: > On Sun, 19 Dec 2004 12:32:12 +1100 Sam Watkins <[EMAIL PROTECTED]> wrote: > > > >there are two different data segments for a C or C++ program, > >the stack and the heap. > > > >stack space is (apparently) limited under Linux. > >If you declare an array like this in a function: > > > > char c[10*1024*1024]; > > > >and write to it all, this will cause a segfault on Linux. > > That was my problem. > > >The heap can get much bigger, you can malloc as much as you want on the > >heap > >(up to the limits of VM and process address space) and it won't segfault, > >e.g. > > > > char *c = malloc(100*1024*1024); > > int i; > > for (i=0; i<size; ++i) > > c[i] = 'x'; > > free(c); > > Thanks, allocating space with malloc worked great. I get my character > array with 68000000 bytes (2000 x 34000) with no problems. > > > > >you need to: > > > >#include <malloc.h> > > > >to use malloc. > > malloc.h must have been included by one of my other include files, as I > did not include it directly. > > >C++ allocates data on the heap when you use "new", and the STL containers > >allocate their elements on the heap. For example these are ok: > > > > char *c = new char[100*1024*1024]; > > // ... > > delete[] c; > > > > vector<int> v; > > int i; > > for (i=0; i<10000000; ++i) > > v.push_back(i); > > Vectors are single dimensional, I believe, so they wont work in this > case, but I will remember them for the future >
When one defines an STL vector with, for instance: vector< int > v; inside a routine, it has the appearance of being allocated on the stack. This is true, but not really true. A small control block is on the stack, but the data 'contained' in the vector is actually on the heap. C++ takes care of deallocating it when it runs its destructor on v. So C++ is more economical in use of stack space than C, and still safe against mistakes that can happen when using globally allocated store. -- Paul E Condon [EMAIL PROTECTED] -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
