Question

We recently received a report that our application will occasionally fail to run. I tracked down the problem code to this:

struct ARRAY2D
{
   long[] col;
}

int numRows = 800000;
int numCols = 300;
array = (ARRAY2D*) malloc(numRows * numCols * sizeof(long))

This allocation of 800 Mb can fail if the user doesn’t have a large enough free block. What is the best way to change how I allocate the memory?

Keep in mind that I have a large amount of code that accesses this object like this: array[row].col[colNum], so I need something that requires minor or primarily find & replace editing of the array access code.

Answer 1

You can allocate smaller chunks of memory separately, instead of one huge block.

long** array = NULL;  
array = (long**) malloc(numCols * sizeof(long*));  
for (int i = 0; i < numCols; i++)  
   array[i] = (long*)  malloc(numRows * sizeof(long));

Generally, memory allocation may fail, every allocation. However, let’s say statistically, due to memory fragmentation, allocating a single large block of memory has higher chance to fail more often than allocating N number of smaller blocks.
Although, also the solution above may cause problems as it is a bit like a double-bladed sword because it may lead to further memory fragmentation.

In other words, there is no generally perfect answer and solution depends on details of a system and application.

As from the comments it seems C++ library is a possibility, then solution based on std::vector (i.e. generic vector of vectors in C++) or using Boost.MultiArray