Pattern matching and guards for Python functions

## Project description

function-pattern-matching (fpm for short) is a module which introduces Erlang-style multiple clause defined functions and guard sequences to Python.

This module is both Python 2 and 3 compatible.

<nav class="contents" id="table-of-contents" role="doc-toc">

</nav>

## Introduction

Two families of decorators are introduced:

• case: allows multiple function clause definitions and dispatches to correct one. Dispatch happens on the values of call arguments or, more generally, when call arguments’ values match specified guard definitions.

• dispatch: convenience decorator for dispatching on argument types. Equivalent to using case and guard with type checking.

• guard: allows arguments’ values filtering and raises GuardError when argument value does not pass through argument guard.

• rguard: Wrapper for guard which converts first positional decorator argument to relguard. See Relguards.

• raguard: Like rguard, but converts return annotation. See Relguards.

Usage example:

• All Python versions:

import function_pattern_matching as fpm

@fpm.case
def factorial(n=0):
return 1

@fpm.case
@fpm.guard(fpm.is_int & fpm.gt(0))
def factorial(n):
return n * factorial(n - 1)
• Python 3 only:

import function_pattern_matching as fpm

@fpm.case
def factorial(n=0):
return 1

@fpm.case
@fpm.guard
def factorial(n: fpm.is_int & fpm.gt(0)): # Guards specified as annotations
return n * factorial(n - 1)

Of course that’s a poor implementation of factorial, but illustrates the idea in a simple way.

Note: This module does not aim to be used on production scale or in a large sensitive application (but I’d be happy if someone decided to use it in his/her project). I think of it more as a fun project which shows how flexible Python can be (and as a good training for myself).

I’m aware that it’s somewhat against duck typing and EAFP (easier to ask for forgiveness than for permission) philosophy employed by the language, but obviously there are some cases when preliminary checks are useful and make code (and life) much simpler.

## Installation

function-pattern-matching can be installed with pip:

\$ pip install function-pattern-matching

Module will be available as function_pattern_matching. It is recommended to import as fpm.

## Usage

### Guards

With guard decorator it is possible to filter function arguments upon call. When argument value does not pass through specified guard, then GuardError is raised.

When global setting strict_guard_definitions is set True (the default value), then only GuardFunc instances can be used in guard definitions. If it’s set to False, then any callable is allowed, but it is not recommended, as guard behaviour may be unexpected (RuntimeWarning is emitted), e.g. combining regular callables will not work.

GuardFunc objects can be negated with ~ and combined together with &, | and ^ logical operators. Note however, that xor isn’t very useful here.

Note: It is not possible to put guards on varying arguments (*args, **kwargs).

#### List of provided guard functions

Every following function returns/is a callable which takes only one parameter - the call argument that is to be checked.

• _ - Catch-all. Returns True for any input. Actually, this can take any number of arguments.

• eq(val) - checks if input is equal to val

• ne(val) - checks if input is not equal to val

• lt(val) - checks if input is less than val

• le(val) - checks if input is less or equal to val

• gt(val) - checks if input is greater than val

• ge(val) - checks if input is greater or equal to val

• Is(val) - checks if input is val (uses is operator)

• Isnot(val) - checks if input is not val (uses is not operator)

• isoftype(_type) - checks if input is instance of _type (uses isintance function)

• isiterable - checks if input is iterable

• eTrue - checks if input evaluates to True (converts input to bool)

• eFalse - checks if input evaluates to False (converts input to bool)

• In(val) - checks if input is in val (uses in operator)

• notIn(val) - checks if input is not in val (uses not in operator)

#### Custom guards

Although it is not advised (at least for simple checks), you can create your own guards:

• by using makeguard decorator on your test function.

• by writing a function that returns a GuardFunc object initialised with a test function.

Note that a test function must have only one positional argument.

Examples:

# use decorator
@fpm.makeguard
def is_not_zero_nor_None(inp):
return inp != 0 and inp is not None

# return GuardFunc object
def is_not_val_nor_specified_thing(val, thing):
return GuardFunc(lambda inp: inp != val and inp is not thing)

# equivalent to (fpm.ne(0) & fpm.Isnot(None)) | (fpm.ne(1) & fpm.Isnot(some_object))
@fpm.guard(is_not_zero_nor_None | is_not_val_nor_specified_thing(1, some_object))
def guarded(argument):
pass

The above two are very similar, but the second one allows creating function which takes multiple arguments to construct actual guard.

Note: It is not recommended to create your own guard functions. In most cases combinations of the ones shipped with fpm should be all you need.

#### Define guards for function arguments

There are two ways of defining guards:

• As decorator arguments

• positionally: guards order will match decoratee’s (the function that is to be decorated) arguments order.

@fpm.guard(fpm.isoftype(int) & fpm.ge(0), fpm.isiterable)
def func(number, iterable):
pass
• as keyword arguments: e.g. guard under name a will guard decoratee’s argument named a.

@fpm.guard(
number = fpm.isoftype(int) & fpm.ge(0),
iterable = fpm.isiterable
)
def func(number, iterable):
pass
• As annotations (Python 3 only)

@fpm.guard
def func(
number: fpm.isoftype(int) & fpm.ge(0),
iterable: fpm.isiterable
): # this is NOT an emoticon
pass

If you try to declare guards using both methods at once, then annotations get ignored and are left untouched.

### Relguards

Relguard is a kind of guard that checks relations between arguments (and/or external variables). fpm implements them as functions (wrapped in RelGuard object) whose arguments are a subset of decoratee’s arguments (no arguments is fine too).

#### Define relguard

There are a few ways of defining a relguard.

• Using guard with the first (and only) positional non-keyword argument of type RelGuard:

@fpm.guard(
fpm.relguard(lambda a, c: a == c), # converts lambda to RelGuard object in-place
a = fpm.isoftype(int) & fpm.eTrue,
b = fpm.Isnot(None)
)
def func(a, b, c):
pass
• Using guard with the return annotation holding a RelGuard object (Python 3 only):

@fpm.guard
def func(a, b, c) -> fpm.relguard(lambda a, b, c: a != b and b < c):
pass
• Using rguard with a regular callable as the first (and only) positional non-keyword argument.

@fpm.rguard(
lambda a, c: a == c, # rguard will try converting this to RelGuard object
a = fpm.isoftype(int) & fpm.eTrue,
b = fpm.Isnot(None)
)
def func(a, b, c):
pass
• Using raguard with a regular callable as the return annotation.

@fpm.raguard
def func(a, b, c) -> lambda a, b, c: a != b and b < c: # raguard will try converting lambda to RelGuard object
pass

As you can see, when using guard you have to manually convert functions to RelGuard objects with relguard method. By using rguard or raguard decorators you don’t need to do it by yourself, and you get a bit cleaner definition.

### Multiple function clauses

With case decorator you are able to define multiple clauses of the same function.

When such a function is called with some arguments, then the first matching clause will be executed. Matching clause will be the one that didn’t raise a GuardError when called with given arguments.

Note: using case or dispatch (discussed later) disables default functionality of default argument values. Functions with varying arguments (*args, **kwargs) and keyword-only arguments (py3-only) are not supported.

Example:

@fpm.case
def func(a=0): print("zero!")

@fpm.case
def func(a=1): print("one!")

@fpm.case
@fpm.guard(fpm.gt(9000))
def func(a): print("IT'S OVER 9000!!!")

@fpm.case
def func(a): print("some var:", a) # catch-all clause

>>> func(0)
'zero!'
>>> func(1)
'one!'
>>> func(9000.1)
"IT'S OVER 9000!!!"
>>> func(1337)
'some var: 1337'

If no clause matches, then MatchError is raised. The example shown above has a catch-all clause, so MatchError will never occur.

Different arities (argument count) are allowed and are dispatched separetely.

Example:

@fpm.case
def func(a=1, b=1, c):
return 1

@fpm.case
def func(a, b, c):
return 2

@fpm.case
def func(a=1, b=1, c, d):
return 3

@fpm.case
def func(a, b, c, d):
return 4

>>> func(1, 1, 'any')
1
>>> func(1, 0, 0.5)
2
>>> func(1, 1, '', '')
3
>>> func(1, 0, 0, '')
4

As you can see, clause order matters only for same-arity clauses. 4-arg catch-all does not affect any 3-arg definition.

#### Define multi-claused functions

There are three ways of defining a pattern for a function clause:

• Specify exact values as decorator arguments (positional and/or keyword)

@fpm.case(1, 2, 3)
def func(a, b, c):
pass

@fpm.case(1, fpm._, 0)
def func(a, b, c):
pass

@fpm.case(b=10)
def func(a, b, c):
pass
• Specify exact values as default arguments

@fpm.case
def func(a=0):
pass

@fpm.case
def func(a=10):
pass

@fpm.case
def func(a=fpm._, b=3):
pass
• Specify guards for clause to match

@fpm.case
@fpm.guard(fpm.eq(0) & ~fpm.isoftype(float))
def func(a):
pass

@fpm.case
@fpm.guard(fpm.gt(0))
def func(a):
pass

@fpm.case
@fpm.guard(fpm.Is(None))
def func(a):
pass

#### dispatch decorator

dispatch decorator is similar to case, but it lets you to define argument types to match against. You can specify types either as decorator arguments or default values (or as guards, of course, but it makes using dispatch pointless).

Example:

@fpm.dispatch(int, int)
def func(a, b):
print("integers")

@fpm.dispatch
def func(a=float, b=float):
print("floats")

>>> func(1, 1)
'integers'
>>> func(1.0, 1.0)
'floats'

## Examples (the useful ones)

Still working on this section!

• Ensure that an argument is a list of strings. Prevent feeding string accidentally, which can cause some headache, since both are iterables.

• Option 1: do not allow strings

# thanks to creshal from HN for suggestion

lookup = {
"foo": 1,
"bar": 2,
"baz": 3
}

@fpm.guard
def getSetFromDict(
dict_, # let it throw TypeError if not a dict. Will be more descriptive than a GuardError.
keys: ~fpm.isoftype(str)
):
"Returns a subset of elements of dict_"
ret_set = set()
for key in keys:
try:
except KeyError:
pass
return ret_set

getSetFromDict(lookup, ['foo', 'baz', 'not-in-lookup']) # will return two-element set
getSetFromDict(lookup, 'foo') # raises GuardError, but would return empty set without guard!

## Project details

Uploaded py2 py3