Exception handling, the functional way.
MIT License
A simple, Try
implementation inspired by
scala.util.Try
Wrap functions with arguments:
>>> from tryingsnake import Try, Try_, Success, Failure
>>> from operator import add, truediv
>>> Try(add, 0, 1)
Success(1)
>>> Try(truediv, 1, 0) # doctest:+ELLIPSIS
Failure(ZeroDivisionError(...))
Avoid sentinel values:
>>> def mean_1(xs):
... try:
... return sum(xs) / len(xs)
... except ZeroDivisionError as e:
... return float("inf") # What does it mean?
>>> mean_1([])
inf
vs.
>>> def mean_2(xs):
... return sum(xs) / len(xs)
>>> Try(mean_2, []) # doctest:+ELLIPSIS
Failure(ZeroDivisionError(...))
>>> Try(mean_2, ["foo", "bar"]) # doctest:+ELLIPSIS
Failure(TypeError(...))
Follow the happy path:
>>> def inc(x): return x + 1
>>> def inv(x): return 1. / x
>>> Success(1).map(inc).map(inv)
Success(0.5)
>>> Failure(Exception("e")).map(inc).map(inv) # doctest:+ELLIPSIS
Failure(Exception(...))
>>> Success(-1).map(inc).map(inv) # doctest:+ELLIPSIS
Failure(ZeroDivisionError(...))
Recover:
>>> def get(url):
... if "mirror" in url:
... raise IOError("No address associated with hostname")
... return url
>>> mirrors = ["http://mirror1.example.com", "http://example.com"]
>>> Try(get, mirrors[0]).recover(lambda _: get(mirrors[1]))
Success('http://example.com')
Let them fail:
>>> from operator import getitem
>>> Try(getitem, [], 0) # doctest:+ELLIPSIS
Failure(IndexError(...))
>>> Try_.set_unhandled([IndexError])
>>> Try(getitem, [], 0)
Traceback (most recent call last):
...
IndexError: list index out of range
Make things (relatively) simple:
>>> import math
>>> xs = [1.0, 0.0, "-1", -3, 2, 1 + 2j]
>>> sqrts = [Try(math.sqrt, x) for x in xs]
>>> [x.get() for x in sqrts if x.isSuccess]
[1.0, 0.0, 1.4142135623730951]
>>> def get_etype(e):
... return Try(lambda x: type(x).__name__, e)
>>> [x.recoverWith(get_etype).get() for x in sqrts if x.isFailure]
['TypeError', 'ValueError', 'TypeError']
Inline exception handling:
>>> from tryingsnake.curried import Try
>>> map(Try(str.split), ["foo bar", None]) # doctest:+ELLIPSIS
<map at ...>
Decorate your functions:
>>> from tryingsnake.curried import Try as try_
>>> @try_
... def scale_imag(x):
... return complex(x.real, x.imag * 2)
>>> [scale_imag(x) for x in [1 + 2j, "3", 42 + 0j]]
[Success((1+4j)), Failure(AttributeError("'str' object has no attribute 'real'")), Success((42+0j))]
Wrap generator objects:
>>> def get_nth(xs, i):
... yield xs[i]
>>> xs = [1, 3, 5, 7]
>>> Try(get_nth(xs, 3))
Success(7)
>>> Try(get_nth(xs, 11))
Failure(IndexError('list index out of range'))
>>> def f():
... divisor = 1
... while True:
... divisor_ = yield 1 / divisor
... divisor = divisor_ if divisor_ is not None else 1
>>> g = f()
>>> next(g) # Should be primed
1.0
>>> Try(g, 2)
Success(0.5)
>>> Try(g, 0)
Failure(ZeroDivisionError('division by zero'))
This package is available on PYPI:
pip install tryingsnake
and conda-forge:
conda install -c conda-forge tryingsnake
tryingsnake
supports Python 3.6 or later and
requires no external dependencies.
MIT, See LICENSE
Q: Is this project production-ready?
A: Sure, for some definition of production-ready. It is a toy project. It has decent test coverage, stable API, and in general seems to do what is expected to do. But it is not widely used, and the API design and overall idea are rather unpythonic.
Q: Why to use mixedCase method names instead of lowercase recommended by PEP8?
A: Mostly to make switching between Python and Scala code as painless as possible.
Q: What is the runtime cost?
A: As of 0088286 (releases 0.3 and 0.4 suffered from
severe performance regression caused by using typing.Generic
as a base of
try. See #18 for details)
rough numbers for simple tasks look as follows:
Python 3.7.5 (default, Oct 27 2019, 15:43:29)
Type 'copyright', 'credits' or 'license' for more information
IPython 7.11.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: def identity(x): return x
In [2]: from tryingsnake import Try
In [3]: %timeit for i in range(1_000_000): identity(i)
59.8 ms ± 683 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
In [4]: %timeit for i in range(1_000_000): Try(identity, i)
408 ms ± 4.14 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
and execution time is dominated by the initializer:
In [5]: import cProfile
In [6]: cProfile.run("for i in range(1_000_000): Try(identity, i)")
4000003 function calls in 0.961 seconds
Ordered by: standard name
ncalls tottime percall cumtime percall filename:lineno(function)
1000000 0.078 0.000 0.078 0.000 <ipython-input-1-abafd771428d>:1(identity)
1 0.263 0.263 0.961 0.961 <string>:1(<module>)
1000000 0.094 0.000 0.094 0.000 __init__.py:234(__init__)
1000000 0.480 0.000 0.698 0.000 __init__.py:352(Try)
1000000 0.046 0.000 0.046 0.000 {built-in method builtins.callable}
1 0.000 0.000 0.961 0.961 {built-in method builtins.exec}
1 0.000 0.000 0.000 0.000 {method 'disable' of '_lsprof.Profiler' objects}
This is quite a lot for simple functions so you should probably avoid it in such cases, where raw performance is important. It is still possible to amortize the cost in such cases, for example using composition:
from toolz.functoolz import compose
from tryingsnake import Try
Try(compose(str.split, str.lower, str.strip), " Foo BAR FooBar ")
Memory overhead (as measured by memory-profiler) looks as follows:
Line # Mem usage Increment Line Contents
================================================
6 37.9 MiB 37.9 MiB @profile
7 def f():
8 155.5 MiB 0.8 MiB [Try(identity, i) for i in range(1_000_000)]
compared to:
Line # Mem usage Increment Line Contents
================================================
6 37.9 MiB 37.9 MiB @profile
7 def f():
8 77.4 MiB 1.0 MiB [identity(i) for i in range(1_000_000)]