To be pedantic, in-place sorting algorithms (typically?) take O(log n) space, not O(1). They don't copy the data to be sorted, but they do need some temporary space that isn't fixed, but scales (slowly) with the size of the array being sorted. The usual sources of the log-n space growth come from either a stack recursing on things (explicitly or implicitly) and/or the array indices.
This particular implementation (looking at the C version) uses fixed-size 'long ints' for its temporary data storage, which means it only works on arrays up to LONG_MAX elements. If you had larger arrays, your need for temporary data would grow, e.g. you could upgrade all those long ints to long long ints and accommodate arrays up to LLONG_MAX. Of course, logarithmic growth is very slow.
This particular implementation (looking at the C version) uses fixed-size 'long ints' for its temporary data storage, which means it only works on arrays up to LONG_MAX elements. If you had larger arrays, your need for temporary data would grow, e.g. you could upgrade all those long ints to long long ints and accommodate arrays up to LLONG_MAX. Of course, logarithmic growth is very slow.