abnf 2.5.0

Parsers for ABNF grammars.

ABNF

ABNF is a package that generates parsers from ABNF grammars as described in RFC 5234 and RFC7405. The main purpose of this package is to parse data as specified in RFCs. But it should be able to handle any ABNF grammar.

ABNF was originally written a few years ago for parsing HTTP headers in a web framework. The code herein has been in use in production here and there on the internet since then.

Requirements

ABNF is tested with Python 3.10-14.

Installation

The abnf package is available from PyPI. As of version 2.3.1, abnf uses trusted publishing.

Install it with the Python installer of your choice:

pip install abnf
uv pip install abnf            # uv
uv add abnf                    # uv, inside a project
poetry add abnf                # poetry

For substantially faster parsing, install the optional Rust backend with the rust extra:

pip install 'abnf[rust]'
uv pip install 'abnf[rust]'
uv add 'abnf[rust]'
poetry add 'abnf[rust]'

(Most shells require the quotes around abnf[rust] because [...] is a glob pattern.) See The Rust backend for details and benchmark numbers.

The Rust backend

When the abnf-rust companion is installed (via pip install abnf[rust]), abnf.parser transparently dispatches its combinator primitives to a Rust extension built with PyO3. The public API is unchanged in either case, including every RFC grammar module and every example in this README; set the environment variable ABNF_NO_RUST=1 to force the pure-Python backend at runtime.

Representative benchmarks (median of three runs of pytest tests/benchmarks/, Apple Silicon):

Grammar / input	Pure Python	Rust	Speed-up
RFC 3986 URI   https://user:[e-mail]:8080/a/b/c?q=1&r=2#frag	190 µs	28.4 µs	6.7×
RFC 5322 mailbox   Charles Yeomans <[e-mail]>	422 µs	44.9 µs	9.4×
RFC 7230 request-line   GET /index.html HTTP/1.1\r\n	73 µs	10.4 µs	7.0×
RFC 9051 astring   HelloWorld42	5.9 µs	4.0 µs	1.5×
Fuzz suite (512 cases, inputs up to 138 KB)	9.1 s	1.42 s	6.4×

Where the Rust backend helps most

The Rust backend's biggest advantage is how cheaply it rejects parses that don't match. ABNF parsing is built around alternation and optional groups: on every backtracking step the algorithm tries an alternative, watches it fail, and moves on. The Python combinator's failure path raises a ParseError, propagates it through generator exception machinery, and unwinds Python frames — all comparatively expensive. The Rust equivalent is a single Err(...) return value with no string formatting. Grammars that exercise this path heavily — RFC 5322's deeply-nested FWS and CFWS whitespace handling is the classic example — see the biggest wins.

The advantage compounds with the number of candidate parses the algorithm considers at each step. RFC 3986's URI grammar enumerates dozens of partial-parse candidates (optional [ "?" query ], optional [ "#" fragment ], multiple hier-part alternatives); each candidate that ends up losing costs only a few hundred nanoseconds in Rust.

Where the gap narrows

The Rust backend is less dominant on tight grammars whose work is mostly successful tree-building rather than backtracking. The clearest example among our benchmarks is the RFC 9051 astring row above: parsing "HelloWorld42" against

astring = 1*ASTRING-CHAR / string

is essentially twelve successful ASTRING-CHAR matches plus a single near-instant failure on the string arm. Each successful match builds a small parse-tree node, and CPython optimises that build path extremely well: a small list allocation goes through CPython's list freelist and costs only a handful of nanoseconds. The Rust equivalent allocates an Arc<Vec<NodeKind>> per node and crosses the PyO3 boundary — both fast, but still a few times more expensive than CPython's open-coded path. The Rust backend still wins on astring, but only modestly.

Roughly:

Grammar shape	Typical Rust speed-up
Heavy alternation / optional / deep backtracking	6–10×
Mixed: moderate alternation with tree-building	3–6×
Pure linear success with few alternatives	1.5–2×

In other words: the more an ABNF grammar exercises backtracking — which most real RFC grammars do — the bigger the win. We have not observed a grammar on which the Rust backend is slower than the pure-Python implementation.

When to skip the Rust backend

For most workloads, pip install abnf[rust] is the right choice and is fully transparent after install. The pure-Python implementation remains a complete and supported parser, and is the right choice when:

• The deployment target rejects compiled extensions (e.g. zip-deployed AWS Lambda layers, some constrained container images).
• You want to debug or step through parser internals — the pure-Python code is shorter, generator-based, and trivially traceable.
• ABNF_NO_RUST=1 is convenient for an A/B comparison.

Much of the optimisation work motivated by the Rust port (lazy Rule.lparse, sorted Alternation, Match hash caching, dedup-by-end-position in Repetition) also landed in the pure-Python implementation, so even without the [rust] extra abnf parses meaningfully faster than its historical baseline.

Usage

The main class of abnf is Rule. You should think of a Rule subclass as corresponding to an ABNF grammar. Then instances of that subclass represent the rules of that grammar.

The Rule class is initialized with the core ABNF rules OCTET, BIT, HEXDIG, CTL, HTAB, LWSP, CR, VCHAR, DIGIT, WSP, DQUOTE, LF, SP, CRLF, CHAR, ALPHA, and so are available in any subclass of Rule.

Create a Rule object using the class method create.

rule = Rule.create('double-quoted-string = DQUOTE *(%x20-21 / %x23-7E / %x22.22) DQUOTE')

To later retrieve the object just created:

rule = Rule('double-quoted-string')

Rule objects are cached, so Rule('double-quoted-string') should always return the same object, though you might not want to depend on that.

ABNF includes several grammars. The Rule subclass ABNFGrammarRule implements the rules for ABNF. The package abnf.grammars includes grammars from several RFCs.

from abnf.grammars import rfc7232
src = 'W/"moof"'
node, start = rfc7232.Rule('ETag').parse(src)
print(str(node))

The output is

Node(
    name=ETag, 
    children=
        [
        Node(
            name=entity-tag, 
            children=
                [
                Node(
                    name=weak, 
                    children=
                        [
                        Node(
                            name=literal, 
                            offset=0, 
                            value="W/"
                            )
                        ]
                    ), 
                    Node(
                        name=opaque-tag, 
                        children=
                            [
                            Node(
                                name=DQUOTE, 
                                children=
                                    [
                                    Node(
                                        name=literal, 
                                        offset=2, 
                                        value="""
                                        )
                                    ]
                                ), 
                            Node(
                                name=etagc, 
                                children=
                                    [
                                    Node(
                                        name=literal, 
                                        offset=3, 
                                        value="m"
                                        )
                                    ]
                                ), 
                            Node(
                                name=etagc, 
                                children=
                                    [
                                    Node(
                                        name=literal, 
                                        offset=4, 
                                        value="o"
                                        )
                                    ]
                                ), 
                            Node(
                                name=etagc, 
                                children=
                                    [
                                    Node(
                                        name=literal, 
                                        offset=5, 
                                        value="o"
                                        )
                                    ]
                                ), 
                            Node(
                                name=etagc, 
                                children=
                                [
                                Node(
                                    name=literal, 
                                    offset=6, 
                                    value="f"
                                    )
                                ]
                            ), 
                        Node(
                            name=DQUOTE, 
                            children=
                                [
                                Node(
                                    name=literal, 
                                    offset=7, 
                                    value="""
                                    )
                                ]
                            )
                        ]
                    )
                ]
            )
        ]
    )'

The modules in abnf.grammars may serve as an example for writing other Rule subclasses. In particular, some of the RFC grammars incorporate rules by reference from other RFC. abnf.grammars.rfc7230 shows a way to import rules from another Rule subclass.

You can also load a grammar from a text file using Rule.from_file. This class function accepts either a str or pathlib.Path. The text file must contain an ABNF rulelist.

class FromFileRule(Rule):
    pass
    
FromFileRule.from_file('/path/to/grammar.abnf')

ABNF uses CRLF as a delimiter for rules in a rulelist. Beware that many text editors (e.g. BBEdit) substitute line endings without telling the user.

Errors

abnf implements two exception subclasses, ParseError and GrammarError.

A GrammarError is raised when parsing encounters an undefined rule, or a prose-value in a grammar.

A ParseError is raised when parsing fails for some reason. Error reporting is nothing more than a stack trace, but that usually allows one to get to the source of the problem.

Examples

Validate an email address

The code below validates an arbitrary email address. If src is not syntactically valid, a ParseError is raised.

from abnf.grammars import rfc5322

src = '[e-mail]'
parser = rfc5322.Rule('address')
parser.parse_all(src)

Extract the actual address from an email address

from abnf.grammars import rfc5322

def get_address(node):
    """Do a breadth-first search of the tree for addr-spec node.  If found, 
    return its value."""
    queue = [node]
    while queue:
        n, queue = queue[0], queue[1:]
        if n.name == 'addr-spec':
            return n.value
        else:
            queue.extend(n.children)
    return None

src = 'John Doe <[e-mail]>'
parser = rfc5322.Rule('address')
node = parser.parse_all(src)
address = get_address(node)

    
for x in node_iterator(node):
    if x.name == 'addr-spec':
        print(x.value)
        break

Extract authentication information from an HTTP Authorization header.

from abnf.parser import NodeVisitor
from abnf.grammars import rfc7235

header_value = 'Basic YWxhZGRpbjpvcGVuc2VzYW1l'
parser = rfc7235.Rule('Authorization')
node, offset = parser.parse(header_value, 0)

class AuthVisitor(NodeVisitor):
    def __init__(self):
        super().__init__()
        self.auth_scheme = None
        self.token = None

    def visit_authorization(self, node):
        for child_node in node.children:
            self.visit(child_node)

    def visit_credentials(self, node):
        for child_node in node.children:
            self.visit(child_node)

    def visit_auth_scheme(self, node):
        self.auth_scheme = node.value

    def visit_token68(self, node):
        self.token = node.value

visitor = AuthVisitor()
visitor.visit(node)

The result is that visitor.auth_scheme = 'Basic', and visitor.token = 'YWxhZGRpbjpvcGVuc2VzYW1l'

Implementation

abnf is implemented using parser combinators. There is a class Literal whose instances are initialized with either a string like 'moof', or a tuple like ('a', 'z') representing a range. The result is a parser that can match the initialized value.

ABNF operations -- alternation, concatenation, repeat, etc. are implemented as classes.
For example, Alternation(Literal('foo'), Literal('bar')) returns a parser that implements the ABNF expression

"foo" / "bar"

The parser bootstraps itself: the RFC 5234 core rules and the ABNF meta-grammar are constructed directly from these combinator classes in code, with no parser available yet to read them from text. ABNFGrammarRule holds the resulting meta-grammar — it is the parser used to read every other grammar, and it can parse its own ABNF source as a self-check.

Backends

abnf has two interchangeable parser implementations behind the same public API:

• Pure Python (abnf._parser_python) — the default. Always available, depends only on the standard library, and serves as the executable reference for the combinator semantics.
• Rust (abnf_rust._ext, installed via pip install abnf[rust]) — a PyO3 extension that reimplements the combinator engine and the ABNF meta-grammar in Rust. When importable, abnf.parser rebinds its combinator primitives (Alternation, Concatenation, Repetition, Option, Literal, Prose, Repeat, Match, Node, LiteralNode) to the Rust pyclasses. Rule, NodeVisitor, ParseError, and GrammarError remain Python in either case so that subclassing, the per-class rule registry, and reflective visitor dispatch continue to work unchanged. See The Rust backend above for benchmark numbers and the kinds of grammars each implementation handles best.

The dispatch happens once at import time in abnf.parser, based on whether the companion abnf_rust extension is importable. Set ABNF_NO_RUST=1 in the environment to force the pure-Python backend even when the extension is installed.

Alternation

RFC 5234 does not specify the precise behavior of alternation. The ABNF definition of ABNF appears to assume longest match. But other grammars expect first match alternation (e.g. dhall). So this behavior is configurable. A class attribute Rule.first_match_alternation allows one to choose a behavior for a particular grammar (as represented by a Rule subclass). When first_match_alternation is False, alternation returns the longest match; in the event of a tie, the first match is returned. When first_match_alternation is True, the first match is returned.

Backtracking

ABNF implements backtracking as of version 2.0.0. There were sufficient changes in behavior that this constituted a breaking change, and so the major version has been bumped.

As is well-known, naive implementations of backtracking typically have exponential worst-case behavior. Here I attempt to reduce that through the use of generators and some caching. In particular, Repetition objects cache parse results.

Version 2.0.0 uses a LRU cache, ParseCache. The code comes wihout any max sizes set for caches, which will obviously result in long-term issues.
My hope is to get feedback from parser usage. ParseCache has a class attribute max_cache_size: int | None that if set to a non-negative integer, will limit cache size.

Development, Testing, etc.

To set up a development environment, install in editable mode with the dev extra. Pick whichever installer you prefer:

pip install -e '.[dev]'

or, with uv (which the project's lockfile and CI both use):

uv sync --extra dev

A good starting point is to run pytest and see that all tests pass:

pytest --cov-report term-missing --cov=abnf

The test suite includes fuzz testing with test data generated using abnfgen. Some of the test rules are long and gruesome, so the tests take a bit of time to complete. Skip the fuzz tests with:

pytest --cov-report term-missing --cov=abnf --ignore=tests/fuzz

Code quality

Pre-commit hooks run ruff, pyright, check-manifest, and tox automatically. Install them once:

pre-commit install

To invoke the checks manually:

ruff check src/abnf       # Lint
pyright                   # Type-check
tox                       # Run pytest across python 3.10-3.14

Working with the Rust backend

To build and install the Rust extension against your dev venv:

pip install -e ./packages/abnf-rust

or, with uv:

uv pip install -e ./packages/abnf-rust

This drives maturin (declared as the PEP 517 build backend) to compile and install the extension; subsequent runs rebuild only what changed. To force the pure-Python backend even with abnf-rust installed:

ABNF_NO_RUST=1 pytest

To run the Rust crate's own unit tests:

cargo test --manifest-path packages/abnf-rust/Cargo.toml

Third-Party Packages

abnf-to-regexp

The program abnf-to-regexp converts augmented Backus-Naur form (ABNF) to a regular expression.