Performance work

[madig]

See #733.

[madig]

I was going to rewrite the parse_float_or_int function like so:

def parse_float_or_int(value_string):
    # NOTE: Profile any changes you make to this function, as it is in the hot
    # file loading path. E.g.
    #
    #   In [32]: %timeit parse_float_or_int("2000")
    #       ...: %timeit parse_float_or_int("-2000")
    #       ...: %timeit parse_float_or_int("123.456")
    #       ...: %timeit parse_float_or_int("-123.456")
    #   183 ns ± 1.11 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
    #   186 ns ± 0.614 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
    #   199 ns ± 2.69 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
    #   201 ns ± 1.56 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
    #
    # There's another way of writing it:
    #
    #   In [2]: def parse_float_or_int2(value_string):
    #      ...:     try:
    #      ...:         return int(value_string)
    #      ...:     except ValueError:
    #      ...:         return float(value_string)
    #
    # The usual case of integers parses slighly faster, but as of Python 3.9.7, the
    # exceptional case of floating point coordinates is much slower.
    #
    #     In [36]: %timeit parse_float_or_int2("2000")
    #         ...: %timeit parse_float_or_int2("-2000")
    #         ...: %timeit parse_float_or_int2("123.456")
    #         ...: %timeit parse_float_or_int2("-123.456")
    #       170 ns ± 1.46 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
    #       173 ns ± 2.41 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
    #       1.18 µs ± 10.3 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
    #       1.19 µs ± 7.65 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
    #
    if "." in value_string:
        return float(value_string)
    return int(value_string)

but apparently it is not necessarily called with a string but also floats and ints in various places?????

Another stumbling block is the function behavior, for "100.0" it will return int(100); changing this makes some equivalence tests fail.