safe-cmp 0.1.1

Examples

Sorting a heterogeneous list:

>>> from safe_cmp import safe_sorted, safe_order
>>> items = [1, None, "foo", {}, object]
>>> list(safe_sorted(items)) # Using "safe_sorted"
[None, {}, 1, 'foo', object]
>>> items.sort(key=safe_order) # Using "safe_order" with key=
>>> items
[None, {}, 1, 'foo', object]

Finding the max of a list which includes nulls:

>>> from safe_cmp import safe_max
>>> safe_max([1, None, 3, None, 4])
4

The rare situation where Python 2 style cmp is useful:

>>> from safe_cmp import safe_cmp
>>> safe_cmp(None, 1)
-1
>>> safe_cmp(None, None)
0
>>> safe_cmp(1, None)
1

Note: safe_cmp will produce Python 2 compatible results when called with nan:

>>> from safe_cmp import safe_cmp
>>> nan = float("NaN")
>>> safe_cmp(nan, 1)
-1
>>> safe_cmp(1, nan)
1
>>> safe_cmp(nan, nan)
0

As will safe_sorted:

>>> from safe_cmp import safe_sorted
>>> list(safe_sorted([nan, 2, nan, 1]))
[nan, 2, nan, 1]

Performance

Currently safe_cmp methods are currently implemented in Python (in contrast to their unsafe builtin counterparts, which are implemented in C), so performance will notable worse for large comparisons:

In [1]: %timeit safe_max(range(10000000))
2.8 s ± 42 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [2]: %timeit max(range(10000000))
345 ms ± 6.23 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

For smaller comparisons, though, the difference will be negligible:

In [1]: %timeit safe_max(1, 2)
682 ns ± 7.12 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

In [2]: %timeit max(1, 2)
218 ns ± 6.87 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

If there is interest in performant implementations, however, they will be straight forward to provide.

Additionally, where obvious, performance optimizations have been implemented (for example, caching the result of key= functions).