functional-parsing-library 0.0.16

A type-safe applicative parsing library

functional_parsing_library

A small production non-ready Python library implementing basic applicative parsers. Roughly speaking, these are functions with signature str -> T | CouldNotParse transforming strings into structured data. For example, you might have a function integer which will transform "1" and "-1919" to the integers 1 and -1919, and the string "boink" to CouldNotParse().

What makes these functions useful is that they can be combined with so-called parser combinators. This way, complicated parsers can be gradually built up from smaller, simpler parsers. For example, if we already have parsers nonnegative_integer and negative_integer, the integer parser from earlier could be written as integer = nonnegative_integer | negative_integer, where | should be read as "or". This library implements various such combinators, such as many, some, ignore_left, many_till, and so on.

Another piece of structure that makes these functions useful is that they're functorial: If I have a parser p of type Parser[T] (that is, a function which parses strings to objects of type T), and a function f: T -> S, then f * p will be a parser for objects of type S. For example, take len * many(word('borf')), and try to parse "borfborfborf". Here word('borf') will parse "borf" to the string "borf" (and any other string to CouldNotParse), so the parser many(word('borf') will try and match as many "borf"s as possible and parse our string to the list ['borf', 'borf', 'borf']. The length of this list is 3, so len * many(word('borf')) parses our string to the integer 3.

This works with multi-argument functions as well. If f is a function of type [T, S] -> U, and we have parsers p and q for objects of type T and S, then f * p & q will first try to match p, and if this succeeds it will try and match q, and finally it will apply f.

Another feature of this library is its type safety. Running mypy on

from functional_parsing_library.strings import word


def add_strings(one: str, two: str, three: str) -> int:
    return len(one + two + three)


reveal_type(add_strings * word('hi'))
reveal_type(add_strings * word('hi') & word('hi'))
reveal_type(add_strings * word('hi') & word('hi') & word('di'))

will show that the first parser has type MappedParser[int, str, str], the second MappedParser[int, str], and the third Parser[int]. Expressions like

add_strings * word('hi') & word('hi') & integer

or

add_strings * word('hi') & word('hi') & word('hi') & word('hi')

will raise a TypeError, and mypy will catch this.

Documentation

To see some documentation, clone this repo, run

make serve-documentation

and in your browser you can peruse this library's docstrings at port 8000.

Some operator overloading weirdness

This library overloads the operators *, <, >, and so on to implement parser combinators. Usually this results in more readable parsers, but there are some quirks in order of evaluation whiich results in unexpected behavior. For example, let's take three parsers, a, b, and c, which parse the strings "a", "b", and "c", respectively. Then one would expect a > b < c to parse "abc" to the string "b". This is because > parses the left parser and discards the result, and proceeds to parse the right parser. Similarly for <. However, a > b < c produces a failure on "abc" with the error message String "abc" does not start with "b". The parser succeeds on "bca" with result "b" and remainder "a".

TODO list

• Hide ParseResults and CouldNotParse by supporting some function parse of type Callable[[Parser[T], str], T] which raises on CouldNotParse or a non-empty remainder. Decouples ParseResults from casual clients.
• Backport to earlier Python versions, say 3.9 and up.
• Ambiguity in parsing, for example char('a') | word('ab') should parse "ab" as both "a" with remainder "b" and as "ab" with no remainder.
• Endow the parsers with a monadic structure. Probably context managers can be used to craft some makeshift Haskell-like do notation.