DHParser 1.9.9

DHParser is a full-stack framework for rapid prototyping and testing of formal grammars (EBNF) and domain specific languages (DSL). See https://dhparser.readthedocs.io/en/latest/ for the full documentation.

Ease of Use

Directly compile existing EBNF-grammars:

DHParser recognizes various dialects of EBNF or PEG-syntax for specifying grammars. For any already given grammar-specification in EBNF or PEG, it is not unlikely that DHParser can generate a parser either right away or with only minor changes or additions.

You can try this by compiling the file XML_W3C_SPEC.ebnf in the examples/XML of the source-tree which contains the official XML-grammar directly extracted from www.w3.org/TR/xml/:

$ dhparser examples/XML/XML_W3C_SPEC.ebnf

This command produces a Python-Script XML_W3C_SPECParser.py in the same directory as the EBNF-file. This file can be run on any XML-file and will yield its concrete syntax tree, e.g.:

$ python examples/XML/XML_W3C_SPECParser.py examples/XML/example.xml

Note, that the concrete syntax tree of an XML file as returned by the generated parser is not the same as the data-tree encoded by that very XML-file. In order to receive the data tree, further transformations are necessary. See examples/XML/XMLParser.py for an example of how this can be done.

Use (small) grammars on the fly in Python code:

Small grammars can also directly be compiled from Python-code. (Here, we use DHParser's preferred syntax which does not require trailing semicolons and uses the tilde ~ as a special sign to denote "insignificant" whitespace.)

key_value_store.py:

#!/usr/bin/env python 
# A mini-DSL for a key value store
from DHParser.dsl import create_parser

# specify the grammar of your DSL in EBNF-notation
grammar = r'''@ drop = whitespace, strings
key_store   = ~ { entry }
entry       = key "="~ value          # ~ means: insignificant whitespace 
key         = /\w+/~                  # Scanner-less parsing: Use regular
value       = /\"[^"\n]*\"/~          # expressions wherever you like'''

# generating a parser is almost as simple as compiling a regular expression
parser = create_parser(grammar)       # parser factory for thread-safety

Now, parse some text and extract the data from the Python shell:

>>> from key_value_store import parser
>>> text = '''
        title    = "Odysee 2001"
        director = "Stanley Kubrick"
    '''
>>> data = parser(text)
>>> for entry in data.select('entry'):
        print(entry['key'], entry['value'])

title "Odysee 2001"
director "Stanley Kubrick"

Or, serialize as XML:

>>> print(data.as_xml())

<key_store>
  <entry>
    <key>title</key>
    <value>"Odysee 2001"</value>
  </entry>
  <entry>
    <key>director</key>
    <value>"Stanley Kubrick"</value>
  </entry>
</key_store>

Add the compiled grammar to your script to save startup-time:

Now, generating a parser with the "create_parser"-function is typically quite slow. If you want to save startup-time, you can simply read out the generated parser's python_src__-field and add the source code to your script.

In order to ensure that generated parser and the grammar stay in sync, you can check if parser needs to be updated by using the grammar_changed()-function as shown in the code-example, below:

#!/usr/bin/env python
# A mini-DSL for a key value store

import re

from DHParser.dsl import create_parser, grammar_changed
from DHParser.parse import *

# specify the grammar of your DSL in EBNF-notation
grammar = r'''@ drop = whitespace, strings
key_store   = ~ { entry }
entry       = key "="~ value          # ~ means: insignificant whitespace 
key         = /[\w]+/~                  # Scanner-less parsing: Use regular
value       = /\"[^"\n]*\"/~          # expressions wherever you like'''

# This class has been generated from the grammar above.
# Do not edit it manually!
class KeyValueGrammar(Grammar):
    r"""Parser for a KeyValue document.

    Instantiate this class and then call the instance with the
    source code as argument in order to use the parser, e.g.:
        parser = KeyValue()
        syntax_tree = parser(source_code)
    """
    source_hash__ = "d6d82738894cbdaa5b14ba8f4254e666"
    disposable__ = re.compile('$.')
    static_analysis_pending__ = []  # type: List[bool]
    parser_initialization__ = ["upon instantiation"]
    COMMENT__ = r''
    comment_rx__ = RX_NEVER_MATCH
    WHITESPACE__ = r'[ \t]*(?:\n[ \t]*(?![ \t]*\n))?'
    WSP_RE__ = mixin_comment(whitespace=WHITESPACE__, comment=COMMENT__)
    wsp__ = Whitespace(WSP_RE__)
    dwsp__ = Drop(Whitespace(WSP_RE__))
    value = Series(RegExp('\\"[^"\\n]*\\"'), dwsp__)
    key = Series(RegExp('[\\w]+'), dwsp__)
    entry = Series(key, Drop(Text("=")), dwsp__, value)
    key_store = Series(dwsp__, ZeroOrMore(entry))
    root__ = key_store

# Check if the grammar has changed. If so, recompile the gramar
# and raise an Exception displaying the updated parser-class.
if grammar_changed(KeyValueGrammar, grammar):
    parser = create_parser(grammar, branding="KeyValue")
    raise AssertionError(
        "Grammar changed! Please, update your source code with:\n\n" \
        + parser.python_src__)

if __name__ == "__main__":
    parser = KeyValueGrammar()
    example = '''
        title    = "Odysee 2001"
        director = "Stanley Kubrick"'''
    print(parser(example).as_xml())

Set up DSL-projects with unit-tests:

For larger DSL projects that require testing and incremental grammar development, use:

$ dhparser NEW_PROJECT_NAME 

to set up a project-directory with all the scaffolding that a serious DSL-project needs, including DHParser's unit-testing-framework for EBNF-grammars.

Features

• Memoizing packrat-parser based on Parsing Expression Grammars. This means:

  • Linear parsing time guaranteed, not matter how complex the grammar.

  • Any EBNF-grammar supported, including left-recursive grammars (supported via "seed and grow"-algorithm - experimental support for Luke Hutchinson's squirrel parser LR-algorithm.)

  • Unlimited look ahead and look behind.

• Macros to avoid code-repetition within grammars

• Declarative tree-transformations for post-processing syntax-trees

• Unit testing framework and post-mortem-debuger for test-driven grammar development and rapid-prototyping of grammars

• Customizable error reporting, recovery after syntax errors and support for fail-tolerant parsers

• Support for Language-servers

• Workflow-support and data-processing-pipelines

• XML-support like mapping flat-text to the DOM-tree ("node-tree" in DHParser's terminology) and adding markup in arbitrary places, even if this requires splitting tags.

• Full Unicode support

• No dependencies except the Python Standard Library

• Extensive documentation and examples

Application Domains

1. Developing parsers for domain specific languages, either existing languages like, LaTeX, or newly created DSLs, like the Medieval-Latin-Dictionary-DSL.

   Typically, these languages are strict formal languages with a context free grammar. (In cases where this does not hold like TeX, it is often still possible to describe a reasonably large subset of the formal language with a context free grammar.)

2. Developing parsers for semi-structured or informally structured text-data.

   This kind of data is typically what you get when retro-digitizing textual data like printed bibliographies, or reference works or dictionaries. Often such works can be captured with a formal grammar, but these grammars require a lot of iterations and tests to develop and usually become much more ramified than the grammars of well-designed formal languages. Thus, DHParser's elaborated testing and debugging-framework for grammars.

   (See Florian Zacherl's Dissertation on the retro-digitalization of dictionary data for an interesting case study. I am confident that the development of a suitable formal grammar is much easier with an elaborated framework like DHParser than with the PHP-parsing-expression-grammar-kit that Florian Zacherl has used.)

3. Developing processing pipelines for tree-structured data.

   In many digital humanities applications one wants to produce different forms of output (say, printed, online-human-readable, online-machine-readable) from one and the same source of data. Therefore, the parsing stage (if the data source is structured text-data) will be followed by more or less intricate bifurcated processing pipelines.

Installation

You can install DHParser from the Python package index pypi.org:

python -m pip install --user DHParser

Alternatively, you can clone the latest version from gitlab.lrz.de/badw-it/DHParser

And these are the steps, if you want set up a DHParser-Project with the latest DHParser-development version:

$ git clone --depth=1 --branch development https://gitlab.lrz.de/badw-it/DHParser
$ python3 DHParser/DHParser/scripts/dhparser.py MyNewParserProject
$ ln -s ../DHParser/DHParser MyNewParserProject/DHParser
$ cd MyNewParserProject

Getting Started

See Introduction.md for the motivation and an overview how DHParser works or jump right into the Step by Step Guide to learn how to set up and use DHParser. Or have a look at the comprehensive overview of DHParser's features to see how DHParser supports the construction of domain specific languages.

Documentation

For the full documentation see: dhparser.readthedocs.io

License

DHParser is open source software under the Apache 2.0 License.

Copyright 2016-2022 Eckhart Arnold, Bavarian Academy of Sciences and Humanities

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Third-Party-Software

DHParser includes the following third-party-software:

• squirrel tests: These test under tests/squirrel_test have been copy and pasted from the Python version of Luke Hutchinson's squirrelparser. The squirrel parser test suite contains many edge cases, in particular with respect to left recursive grammars. (DHParser's left-recursion handling algorithm is highly indepted to Luke Hutchinson's work on this topic!) License: MIT License

• typing_extensions: needed for backwards compatibility with older Python versions. Licensed under the Python Software Foundation License

• shadow_cython: a single module from Cython to avoid error-messages resulting from the use of the cython-decorators when cython is not installed on the system using DHParser. License: Apache 2.0 License

The following are only needed for building the documentation:

• Sphinx book theme: Makes the documentation look better than the built-in sphinx-themes. License: BSD 3-clause license

• pydata sphinx theme: Nedded by sphinx_book_theme, also licensed under the BSD 3-clause license.

• beautifulsoup4: Needed by by pydata_sphinx_theme, licensed unter the MIT-License

• soupsive: Needed by Beautifulsoup, licensed unter the MIT-License

Optional Post-Installation

It is recommended that you install the regex-module (https://bitbucket.org/mrabarnett/mrab-regex). If present, DHParser will use regex instead of the built-in re-module for regular expressions. regex is faster and more powerful than re.

In order to speed up DHParser even more, it can be compiled with the Python to C compiler Cython. Cython version 3.0 or higher is required to compile DHParser. Type:

pip install cython

on the command-line to install cython. Once cython has been built and installed, you can run the "dhparser_cythonize"-script from the command line:

dhparser_cythonize

On Linux-systems, in case you want to use clang instead of the gcc-compiler, type:

export CC=/usr/bin/clang; dhparser_cythonize 

Using clang may also help to circumvent C-errors like "incompatible pointer types".

The Cython-compiled version is about 2-3 times faster than the CPython-interpreted version. Compiling can take quite a while. If you are in a hurry, you can just can also just call dhparser_cythonize_stringview which just compiles the stringview-module, which profits the most from being "cythonized". As of October 2025, cython appears to be incompatible with the InterpreterPoolExecutor, newly added to Python 3.14. Therefore, DHParser will fall back to ProcessPoolExecutors if cython has been used to compile DHParser. (This may change in the future.)

Another way to speed up your parser is by adding "@ optimizations = all" at the beginning of your EBNF-grammar-file. DHParser then tries to compile (some) non-recursive parts of your grammar to entirely to regular rexpressions which yields a 10-20% speedup. Beware that this option is still experimental!

Depending on the use case, e.g. when parsing large files, PyPy3 might yield a considerable speed-up. However, in most other cases pypy is noticeably slower than cpython! To circumvent the longer startup times of pypy3 in comparison to CPython, it is recommended to use the xxxServer.py-scripts rather than calling the xxxParser.py-script each time when parsing many documents subsequently.

The same goes for GraalPy, which seems much slower, most of the time. However, I have only done few experiments with both technologies (pypy and graal), so far.

Sources

Find the sources on gitlab.lrz.de/badw-it/DHParser . Get them with:

git clone https://gitlab.lrz.de/badw-it/DHParser

There exists a mirror of this repository on Github: https://github.com/jecki/DHParser Be aware, though, that the github-mirror may occasionally lag behind a few commits.

Packaging

DHParser uses Poetry for packaging and dependency-management. In order to build a package from the sources, type:

poetry build

on the command line. The packages will then appear in the "dist" subdirectory.

Author

Author: Eckhart Arnold, Bavarian Academy of Sciences Email: arnold@badw.de

How to cite

If you use DHParser for Scientific Work, then please cite it as:

DHParser. A Parser-Generator for Digital-Humanities-Applications,
Division for Digital Humanities Research & Development, Bavarian Academy of Science and Technology, Munich Germany 2017-2025, https://gitlab.lrz.de/badw-it/dhparser

References and Acknowledgement

A very incomplete list of literature that has been used to develop DHParser.

Eckhart Arnold: Domänenspezifische Notationen. Eine (noch) unterschätzte Technologie in den Digitalen Geisteswissenschaften, Präsentation auf dem dhmuc-Workshop: Digitale Editionen und Auszeichnungssprachen, München 2016. Short-URL: tiny.badw.de/2JVT

Brian Ford: Parsing Expression Grammars: A Recognition-Based Syntactic Foundation, Cambridge Massachusetts, 2004. Short-URL:t1p.de/jihs

Richard A. Frost, Rahmatullah Hafiz and Paul Callaghan: Parser Combinators for Ambiguous Left-Recursive Grammars, in: P. Hudak and D.S. Warren (Eds.): PADL 2008, LNCS 4902, pp. 167–181, Springer-Verlag Berlin Heidelberg 2008.

Luke A. D. Hutchinson: The Squirrel Parser: A Linear-Time PEG Packrat Parser Capable of Left Recursion and Optimal Error Recovery, ArxiV, arXiv:2601.05012

Elizabeth Scott and Adrian Johnstone, GLL Parsing, in: Electronic Notes in Theoretical Computer Science 253 (2010) 177–189, dotat.at/tmp/gll.pdf

Dominikus Herzberg: Objekt-orientierte Parser-Kombinatoren in Python, Blog-Post, September, 18th 2008 on denkspuren. gedanken, ideen, anregungen und links rund um informatik-themen, short-URL: t1p.de/bm3k

Dominikus Herzberg: Eine einfache Grammatik für LaTeX, Blog-Post, September, 18th 2008 on denkspuren. gedanken, ideen, anregungen und links rund um informatik-themen, short-URL: t1p.de/7jzh

Dominikus Herzberg: Uniform Syntax, Blog-Post, February, 27th 2007 on denkspuren. gedanken, ideen, anregungen und links rund um informatik-themen, short-URL: t1p.de/s0zk

John MacFarlane, David Greenspan, Vicent Marti, Neil Williams, Benjamin Dumke-von der Ehe, Jeff Atwood: CommonMark. A strongly defined, highly compatible specification of Markdown, 2017. commonmark.org

Stefan Müller: DSLs in den digitalen Geisteswissenschaften, Präsentation auf dem dhmuc-Workshop: Digitale Editionen und Auszeichnungssprachen, München 2016. Short-URL: tiny.badw.de/2JVy

Markus Voelter, Sbastian Benz, Christian Dietrich, Birgit Engelmann, Mats Helander, Lennart Kats, Eelco Visser, Guido Wachsmuth: DSL Engineering. Designing, Implementing and Using Domain-Specific Languages, 2013. dslbook.org/

Redziejowski, R.R. (2023). Left Recursion by Recursive Ascent. In: Schlingloff, BH., Vogel, T., Skowron, A. (eds) Concurrency, Specification and Programming. Studies in Computational Intelligence, vol 1091. Springer, Cham. https://doi.org/10.1007/978-3-031-26651-5_2 URL: https://ceur-ws.org/Vol-2951/paper2.pdf

Christopher Seaton: A Programming Language Where the Syntax and Semantics are Mutuable at Runtime, University of Bristol 2007, chrisseaton.com/katahdin/katahdin.pdf

Vegard Øye: General Parser Combinators in Racket, 2012, epsil.github.io/gll/

and many more...