Andrew Haley wrote:
It gives performance advantage over what, exactly?
Using malloc/realloc/free gives a performance advantage over new/delete since realloc need not move the block if it can resize inplace. For example: double *d=static_cast<double *>(malloc(1000*sizeof(double))); d=static_cast<double *>(realloc(500*sizeof(double))); the block pointed to by d does not need to be moved and the 500 doubles that are freed are available on the heap. But to do the same with new, double *d=new double[1000]; double *tmp=new double[500]; memcpy(tmp,d,500*sizeof(double)); delete [] d; d=tmp; With new you always have to move the block, if you don't explicitly move when shrinking, the bytes you no longer need are not available for the heap to use. In both cases the same goes for extending the length of the array. Apparently when using new/delete the heap can become more fragmented than when using malloc/realloc/free, and lead to user running out of memory sooner. I was exploring the new/delete avenue because it would seem cleaner to me to stick with C++ mechanism in a C++ code, and we'd get the benefit of bad_alloc exceptions. However, given that new is using malloc under the hood(at least by g++) I suppose using malloc/realloc/free is a non issue. Thanks for the feedback. Burlen
