Parsr is a simple parser combinator library in pure python.


Keywords
parser, parser-combinators, parsing, parsing-combinators, python
License
Apache-2.0
Install
pip install parsr==0.4.2

Documentation

Documentation Status Test Status Coverage Status

parsr

parsr is a little library for parsing simple, mostly context free grammars that might require knowledge of indentation or matching tags.

It contains a small set of combinators that perform recursive decent with backtracking. Fancy tricks like rewriting left recursions and optimizations like packrat are not implemented since the goal is a library that's small yet sufficient for parsing non-standard configuration files. It also includes a generic data model that parsers can target to take advantage of an embedded query system.

To see how a handwritten parser might evolve to something like this project, check out the lesson.

parser.query contains the common data model and query system.

Install

  1. Ensure python2.7, python3.6, or python3.7 is installed.
  2. python3.7 -m venv myproject && cd myproject
  3. source bin/activate
  4. pip install parsr

Examples

Primitives

These are the building blocks for matching individual characters, sets of characters, and a few convenient objects like numbers. All matching is case sensitive except for the ignore_case option with Literal.

Char

Match a single character.

a = Char("a")     # parses a single "a"
val = a("a")      # produces an "a" from the data.
val = a("b")      # raises an exception

InSet

Match any single character in a set.

vowel = InSet("aeiou")  # or InSet(set("aeiou"))
val = vowel("a")  # okay
val = vowel("e")  # okay
val = vowel("i")  # okay
val = vowel("o")  # okay
val = vowel("u")  # okay
val = vowel("y")  # raises an exception

String

Match one or more characters in a set. Matching is greedy.

vowels = String("aeiou")
val = vowels("a")            # returns "a"
val = vowels("u")            # returns "u"
val = vowels("aaeiouuoui")   # returns "aaeiouuoui"
val = vowels("uoiea")        # returns "uoiea"
val = vowels("oouieaaea")    # returns "oouieaaea"
val = vowels("ga")           # raises an exception

StringUntil

Matches any number of characters until a predicate is seen. You may set lower and upper bounds. Both are inclusive. The characters that match the predicate are not consumed.

su  = StringUntil(Char("="))  # parses any number of characters until '='
val = su("ab=")               # produces "ab" from the data.
val = su("ab")                # raises an exception

su  = StringUntil(Char("="), lower=2)  # parses at least two characters until '='
val = su("ab=")                        # produces "ab" from the data.
val = su("a=")                         # raises an exception

su  = StringUntil(Char("="), upper=2)  # parses at most two characters until '='
val = su("ab=")                        # produces "ab" from the data.
val = su("a=")                         # produces "a"
val = su("abc=")                       # raises an exception

Regex

Match characters against a regular expression.

identifier = Regex("[a-zA-Z]([a-zA-Z0-9])*")
identifier("abcd1") # returns "abcd1"
identifier("1bcd1") # raises an exception

Literal

Match a literal string. The value keyword lets you return a python value instead of the matched input. The ignore_case keyword makes the match case insensitive.

lit = Literal("true")
val = lit("true")  # returns "true"
val = lit("True")  # raises an exception
val = lit("one")   # raises an exception

lit = Literal("true", ignore_case=True)
val = lit("true")  # returns "true"
val = lit("TRUE")  # returns "TRUE"
val = lit("one")   # raises an exception

t = Literal("true", value=True)
f = Literal("false", value=False)
val = t("true")  # returns the boolean True
val = t("True")  # raises an exception

val = f("false") # returns the boolean False
val = f("False") # raises and exception

t = Literal("true", value=True, ignore_case=True)
f = Literal("false", value=False, ignore_case=True)
val = t("true")  # returns the boolean True
val = t("True")  # returns the boolean True

val = f("false") # returns the boolean False
val = f("False") # returns the boolean False

Number

Match a possibly negative integer or simple floating point number and return the python int or float for it.

val = Number("123")  # returns 123
val = Number("-12")  # returns -12
val = Number("12.4")  # returns 12.4
val = Number("-12.4")  # returns -12.4

parsr also provides SingleQuotedString, DoubleQuotedString, QuotedString, EOL, EOF, WS, AnyChar, and several other primitives. See the bottom of parsr/__init__.py

Combinators

There are several ways of combining primitives and their combinations.

Sequence

Require expressions to be in order.

Sequences are optimized so only the first object maintains a list of itself and following objects. Be aware that using a sequence in other sequences will cause it to accumulate the elements of the new sequence onto it, which could affect it if it's used in multiple definitions. To ensure a sequence isn't "sticky" after its definition, wrap it in a Wrapper object.

a = Char("a")     # parses a single "a"
b = Char("b")     # parses a single "b"
c = Char("c")     # parses a single "c"

ab = a + b        # parses a single "a" followed by a single "b"
                  # (a + b) creates a "Sequence" object. Using `ab` as an
                  # element in a later sequence would modify its original
                  # definition.

abc = a + b + c   # parses "abc"
                  # (a + b) creates a "Sequence" object to which c is appended

val = ab("ab")    # produces a list ["a", "b"]
val = ab("a")     # raises an exception
val = ab("b")     # raises an exception
val = ab("ac")    # raises an exception
val = ab("cb")    # raises an exception

val = abc("abc")  # produces ["a", "b", "c"]

Choice

Accept one of several alternatives. Alternatives are checked from left to right, and checking stops with the first one to succeed.

Choices are optimized so only the first object maintains a list of alternatives. Be aware that using a choice object as an element in other choices will cause it to accumulate the elemtents of the new choice onto it, which could affect it if it's used in multiple definitions. To ensure a Choice isn't "sticky" after its definition, wrap it in a Wrapper object.

abc = a | b | c   # alternation or choice.
val = abc("a")    # parses a single "a"
val = abc("b")    # parses a single "b"
val = abc("c")    # parses a single "c"
val = abc("d")    # raises an exception

Many

Match zero or more occurences of an expression. Matching is greedy.

Since Many can match zero occurences, it always succeeds. Keep this in mind when using it in a list of alternatives or with FollowedBy or NotFollowedBy.

x = Char("x")
xs = Many(x)      # parses many (or no) x's in a row
val = xs("")      # returns []
val = xs("a")     # returns []
val = xs("x")     # returns ["x"]
val = xs("xxxxx") # returns ["x", "x", "x", "x", "x"]
val = xs("xxxxb") # returns ["x", "x", "x", "x"]

ab = Many(a + b)  # parses "abab..."
val = ab("")      # produces []
val = ab("ab")    # produces [["a", b"]]
val = ab("ba")    # produces []
val = ab("ababab")# produces [["a", b"], ["a", "b"], ["a", "b"]]

ab = Many(a | b)  # parses any combination of "a" and "b" like "aababbaba..."
val = ab("aababb")# produces ["a", "a", "b", "a", "b", "b"]

xs = Many(x, lower=1)     # parses many (or no) x's in a row
val = xs("")      # raises an exception
val = xs("a")     # raises an exception
val = xs("x")     # returns ["x"]
val = xs("xxxxx") # returns ["x", "x", "x", "x", "x"]
val = xs("xxxxb") # returns ["x", "x", "x", "x"]

ab = Many(a + b, lower=1) # parses "abab..."
val = ab("")      # raises an exception
val = ab("ab")    # produces [["a", "b"]]
val = ab("ba")    # raises an exception
val = ab("ababab")# produces [["a", "b"], ["a", "b"], ["a", "b"]]

ab = Many(a | b, lower=1) # parses any combination of "a" and "b" like "aababbaba..."
val = ab("aababb")# produces ["a", "a", "b", "a", "b", "b"]

ab = Many(a | b, upper=2) # parses any combination of "a" and "b" like "aababbaba..."
val = ab("ab")    # produces ["a", "b"]
val = ab("aab")   # raises an exception

Until

Match zero or more occurences of an expression until a predicate matches. Matching is greedy.

Since Until can match zero occurences, it always succeeds. Keep this in mind when using it in a list of alternatives or with FollowedBy or NotFollowedBy.

cs = AnyChar.until(Char("y")) # parses many (or no) characters until a "y" is
                              # encountered.

val = cs("")                  # returns []
val = cs("a")                 # returns ["a"]
val = cs("x")                 # returns ["x"]
val = cs("ccccc")             # returns ["c", "c", "c", "c", "c"]
val = cs("abcdycc")           # returns ["a", "b", "c", "d"]

Followed by

Require an expression to be followed by another, but don't consume the input that matches the latter expression.

ab = Char("a") & Char("b") # matches an "a" followed by a "b", but the "b"
                           # isn't consumed from the input.
val = ab("ab")             # returns "a" and leaves "b" to be consumed.
val = ab("ac")             # raises an exception and doesn't consume "a".

Not followed by

Require an expression to not be followed by another.

anb = Char("a") / Char("b") # matches an "a" not followed by a "b".
val = anb("ac")             # returns "a" and leaves "c" to be consumed
val = anb("ab")             # raises an exception and doesn't consume "a".

Keep Left / Keep Right

KeepLeft (<<) and KeepRight (>>) match adjacent expressions but ignore one of their results.

a = Char("a")
q = Char('"')

qa = a << q      # like a + q except only the result of a is returned
val = qa('a"')   # returns "a". Keeps the thing on the left of the << 

qa = q >> a      # like q + a except only the result of a is returned
val = qa('"a')   # returns "a". Keeps the thing on the right of the >> 

qa = q >> a << q # like q + a + q except only the result of the a is returned
val = qa('"a"')  # returns "a".

Opt

Opt wraps a parser and returns a default value of None if it fails. That value can be changed with the default keyword. Input is consumed if the wrapped parser succeeds but not otherwise.

a = Char("a")
o = Opt(a)      # matches an "a" if its available. Still succeeds otherwise but
                # doesn't advance the read pointer.
val = o("a")    # returns "a"
val = o("b")    # returns None. Read pointer is not advanced.

o = Opt(a, default="x") # matches an "a" if its available. Returns "x" otherwise.
val = o("a")    # returns "a"
val = o("b")    # returns "x". Read pointer is not advanced.

map

All parsers have a .map function that allows you to pass a function to evaluate the input they've matched.

def to_number(val):
    # val is like [non_zero_digit, [other_digits]]
    first, rest = val
    s = first + "".join(rest)
    return int(s)

m = NonZeroDigit + Many(Digit)  # returns [nzd, [other digits]]
n = m.map(to_number)  # converts the match to an actual integer
val = n("15")  # returns the int 15

Lift

Allows a multiple parameter function to work on parsers.

def comb(a, b, c):
    """ a, b, and c should be strings. Returns their concatenation."""
    return "".join([a, b, c])

# You'd normally invoke comb like comb("x", "y", "z"), but you can "lift" it for
# use with parsers like this:

x = Char("x")
y = Char("y")
z = Char("z")
p = Lift(comb) * x * y * z

# The * operator separates parsers whose results will go into the arguments of
# the lifted function. I've used Char above, but x, y, and z can be arbitrarily
# complex.

val = p("xyz")  # would return "xyz"
val = p("xyx")  # raises an exception. nothing would be consumed

Forward

Forward allows recursive grammars where a nonterminal's definition includes itself directly or indirectly. You initially create a Forward nonterminal with regular assignment.

expr = Forward()

You later give it its real definition with the <= operator.

expr <= (term + Many(LowOps + term)).map(op)

Arithmetic

Here's an arithmetic parser that ties several concepts together. A progression of this parser from a simple imperative style to what you see below is in the repo.

from parsr import EOF, Forward, InSet, LeftParen, Many, Number, RightParen, WS


def op(args):
    ans, rest = args
    for op, arg in rest:
        if op == "+":
            ans += arg
        elif op == "-":
            ans -= arg
        elif op == "*":
            ans *= arg
        elif op == "/":
            ans /= arg
    return ans


# high precedence operations
HighOps = InSet("*/")

# low precedence operations
LowOps = InSet("+-")

# Operator precedence is handled by having different declarations for each
# prededence level. expr handles low level operations, term handles high level
# operations, and factor handles simple numbers or subexpressions between
# parentheses. Since the first element in expr is term and the first element in
# term is factor, factors are evaluated first, then terms, and then exprs.

# We have to declare expr before its definition since it's used recursively
# through the definition of factor.
expr = Forward()

# A factor is a simple number or a subexpression between parentheses.
factor = WS >> (Number | (LeftParen >> expr << RightParen)) << WS

# A term handles strings of multiplication and division. As written, it would
# convert "1 + 2 - 3 + 4" into [1, [['+', 2], ['-', 3], ['+', 4]]]. The first
# element in the outer list is the initial factor. The second element of the
# outer list is another list, which is the result of the Many. The Many's list
# contains several two-element lists generated from each match of
# (HighOps + factor). We pass the entire structure into the op function with
# map.
term = (factor + Many(HighOps + factor)).map(op)

# expr has the same form and behavior as term.
# Notice that we assign to expr with "<=" instead of "=". This is how you assign
# to nonterminals that have been declared previously as Forward.
expr <= (term + Many(LowOps + term)).map(op)

val = expr("2*(3+4)/3+4")  # returns 8.666666666666668