Allocating and freeing inside a double loop can never be cheap. Couldn't gmt_graticule_path() receive static x,y arrays, sufficiently large, set the non-used elements to 0 for safety, and return the number of used elements?

Since you don't know the number needed, I would recommend something like this:

1. Pass the same arrays in, no freeing inside loop
2. Pass in a size_t *n_alloc argument, initially 0
3. Inside we check before each alloc if n < *n_alloc, and only alloc if too small.
4. Increase *n_alloc for each alloc (there are 4 sides)
5. Free arrays outside the loop in the end

This means in practice that we do most allocations early on and then we just recycle. Or as you suggest, start with n_alloc = GMT_LEN256 or something and then perhaps very little allocations happen.

If you have the time to do this - which I think is pretty simple, knock yourself out!

These two weeks are quite busy but I'll see.