Skip to main content

A C99 preprocessor written in pure Python

Project description

A C99 preprocessor written in pure Python

(C) 2018-2019 Niall Douglas and (C) 2007-2019 David Beazley

PyPI: Github: API reference docs:

Travis master branch all tests passing for Python v2 and v3: travis

A pure universal Python C (pre-)preprocessor implementation very useful for pre-preprocessing header only C++ libraries into single file includes and other such build or packaging stage malarky. The implementation can be used as a Python module (see API reference) or as a command line tool pcpp which can stand in for a conventional C preprocessor (i.e. it’ll accept similar arguments).

Your includes can be benchmarked for heft in order to improve your build times! See the --times and --filetimes options.

A very unique facility of this C preprocessor is partial preprocessing so you can programmatically control how much preprocessing is done by pcpp and how much is done by the C or C++ compiler’s preprocessor. The ultimate control is by subclassing the Preprocessor class in Python from which you can do anything you like, however for your convenience the pcpp command line tool comes with the following canned partial preprocessing algorithms:

Pass through but still execute #defines and #undefs if not always removed by preprocessor logic. This ensures that including the output sets exactly the same macros as if you included the original, plus include guards work.
If an #include is not found, pass it through unmodified. This is very useful for passing through includes of system headers.
This is one of the most powerful pass through algorithms. If an expression passed to #if (or its brethern) contains an unknown macro, expand the expression with known macros and pass through unexecuted, and then pass through the remaining block. Each #elif is evaluated in turn and if it does not contain unknown macros, it will be executed immediately. Finally, any #else clause is always passed through unexecuted. Note that include guards normally defeat this algorithm, so those are specially detected and ignored.
A major use case for pcpp is as a preprocessor for the doxygen reference documentation tool whose preprocessor is unable to handle any preprocessing of any complexity. pcpp can partially execute the preprocessing which doxygen is incapable of, thus generating output which produces good results with doxygen. Hence the ability to pass through comments containing doxygen markup is very useful.
Don’t expand __DATE__, __TIME__, __FILE__, __LINE__ nor __COUNTER__.

Standards (non-)compliance

pcpp passes a modified edition of the mcpp unit test suite. Modifications done were to clarify ternary operators with extra brackets, plus those testing the unusual special quirks in expression evaluation (see detailed description below). It also passes the list of “preprocessor torture” expansion fragments in the C11 standard, correctly expanding some very complex recursive macro expansions where expansions cause new macro expansions to be formed. In this, it handily beats the MSVC preprocessor and ought to handle most C99 preprocessor metaprogramming. If you compare its output side-by-side to that of GCC or clang’s preprocessor, results are extremely close indeed with blank line collapsing being the only difference.

The most non-conforming part is #if expression parsing (donations of a proper yacc based parser for executing #if expressions based on are welcome). In practice, in most real world code, you won’t notice the departures and if you do, the application of extra brackets to group subexpressions so Python’s eval() executes right will fix it.

A full, detailed list of known non-conformance with the C99 standard is below. We have been told that pcpp does not pass the Boost.Wave preprocessor test suite, but the chances of that biting most people is low. If it does, pull requests with bug fixes and new unit tests for the fix are welcome.

Note that most of this preprocessor was written originally by David Beazley to show off his excellent Python Lex-Yacc library PLY ( and is hidden in there without being at all obvious given the number of Stackoverflow questions which have asked for a pure Python C preprocessor implementation. This implementation fixes a lot of conformance bugs (the original was never intended to rigidly adhere to the C standard) and adds in a test suite based on the C11 preprocessor torture samples plus the mcpp preprocessor test suite. Still, this project would not be possible without David’s work, so please take off your hat and give a bow towards him.

Command line tool pcpp:

The help from the command line tool pcpp:

usage: pcpp [-h] [-o [path]] [-D macro[=val]] [-U macro] [-N macro] [-I path]
            [--passthru-defines] [--passthru-unfound-includes]
            [--passthru-unknown-exprs] [--passthru-comments]
            [--disable-auto-pragma-once] [--line-directive [form]] [--debug]
            [--time] [--filetimes [path]] [--version]
            [input [input ...]]

A pure universal Python C (pre-)preprocessor implementation very useful for
pre-preprocessing header only C++ libraries into single file includes and
other such build or packaging stage malarky.

positional arguments:
  input                 Files to preprocess

optional arguments:
  -h, --help            show this help message and exit
  -o [path]             Output to a file instead of stdout
  -D macro[=val]        Predefine name as a macro [with value]
  -U macro              Pre-undefine name as a macro
  -N macro              Never define name as a macro, even if defined during
                        the preprocessing.
  -I path               Path to search for unfound #include's
  --passthru-defines    Pass through but still execute #defines and #undefs if
                        not always removed by preprocessor logic
                        Pass through #includes not found without execution
                        Unknown macros in expressions cause preprocessor logic
                        to be passed through instead of executed by treating
                        unknown macros as 0L
  --passthru-comments   Pass through comments unmodified
                        Pass through double underscore magic macros unmodified
                        Disable the heuristics which auto apply #pragma once
                        to #include files wholly wrapped in an obvious include
                        guard macro
  --line-directive [form]
                        Form of line directive to use, defaults to #line,
                        specify nothing to disable output of line directives
  --debug               Generate a pcpp_debug.log file logging execution
  --time                Print the time it took to #include each file
  --filetimes [path]    Write CSV file with time spent inside each included
                        file, inclusive and exclusive
  --compress            Make output as small as possible
  --version             show program's version number and exit

Note that so pcpp can stand in for other preprocessor tooling, it ignores any
arguments it does not understand.

Quick demo of pass through mode

Let us look at an example for pass through mode. Here is the original:

#if !defined(__cpp_constexpr)
#if __cplusplus >= 201402L
#define __cpp_constexpr 201304  // relaxed constexpr
#define __cpp_constexpr 190000
#if __cpp_constexpr >= 201304
#define BOOSTLITE_CONSTEXPR constexpr

pcpp test.h --passthru-defines --passthru-unknown-exprs will output:

#if !defined(__cpp_constexpr)
#if __cplusplus >= 201402
#define __cpp_constexpr 201304
#define __cpp_constexpr 190000
#if __cpp_constexpr >= 201304
#define BOOSTLITE_CONSTEXPR constexpr

This is because __cpp_constexpr was not defined, so because of the --passthru-unknown-exprs flag we pass through everything inside that if block unexecuted i.e. defines and undefs are NOT executed by pcpp. Let’s define __cpp_constexpr:

pcpp test.h --passthru-defines --passthru-unknown-exprs -D __cpp_constexpr

#line 8 "test.h"


So, big difference now. We execute the entire first if block as __cpp_constexpr is now defined, thus leaving whitespace. Let’s try setting __cpp_constexpr a bit higher:

pcpp test.h --passthru-defines --passthru-unknown-exprs -D __cpp_constexpr=201304

#line 8 "test.h"

#define BOOSTLITE_CONSTEXPR constexpr


As you can see, the lines related to the known __cpp_constexpr are executed and removed, passing through any if blocks with unknown macros in the expression.

What if you want a macro to be known but undefined? The -U (to undefine) flag has an obvious meaning in pass through mode in that it makes a macro no longer unknown, but known to be undefined.

pcpp test.h --passthru-defines --passthru-unknown-exprs -U __cpp_constexpr

#if __cplusplus >= 201402
#define __cpp_constexpr 201304
#define __cpp_constexpr 190000



Here __cpp_constexpr is known to be undefined so the first clause executes, but __cplusplus is unknown so that entire block is passed through unexecuted. In the next test comparing __cpp_constexpr to 201304 it is still known to be undefined, and so 0 >= 201304 is the expressions tested which is false, hence the following stanza is removed entirely.

Helping pcpp using source code annotation

You can achieve a great deal using -D (define), -U (undefine) and -N (never define) on the command line, but for more complex preprocessing it gets hard to pass through the correct logic without some source code annotation.

pcpp lets you annotate which part of an if block being passed through due to use of unknown macros to also be executed in addition to the pass through. For this use __PCPP_ALWAYS_FALSE__ or __PCPP_ALWAYS_TRUE__ which tells pcpp to temporarily start executing the passed through preprocessor commands e.g.

#if !defined(__cpp_constexpr)
#if __cplusplus >= 201402L
#define __cpp_constexpr 201304
#elif !__PCPP_ALWAYS_FALSE__     // pcpp please execute this next block
#define __cpp_constexpr 190000
#if __cpp_constexpr >= 201304
#define BOOSTLITE_CONSTEXPR constexpr

Note that __PCPP_ALWAYS_FALSE__ will always be false in any other preprocessor, and it is also false in pcpp. However it causes pcpp to execute the define of __cpp_constexpr to 190000:

pcpp test.h --passthru-defines --passthru-unknown-exprs

#if !defined(__cpp_constexpr)
#if __cplusplus >= 201402
#define __cpp_constexpr 201304
#elif 1
#define __cpp_constexpr 190000


This is one way of marking up #else clauses so they always execute in a normal preprocessor and also pass through with execution with pcpp. You can, of course, also place || __PCPP_ALWAYS_FALSE__ in any #if stanza to cause it to be passed through with execution, but not affect the preprocessing logic otherwise.

What’s implemented by the Preprocessor class:

  • Digraphs and Trigraphs
  • line continuation operator ‘\
  • C99 correct elimination of comments and maintenance of whitespace in output.
  • __DATE__, __TIME__, __FILE__, __LINE__. Note that __STDC__ et al are NOT defined by default, you need to define those manually before starting preprocessing.
  • __COUNTER__, a very common extension
  • Object #define
  • Function #define macro(…)
    • Retokenisation and reexpansion after expansion is C99 compliant.
  • #undef
  • #include “path”, <path> and PATH
  • defined operator
  • C operators:
    • +, -, !, ~
    • *, /, %
    • +, -
    • <<, >>
    • <, <=, >, >=
    • ==, !=
    • &
    • ^
    • |
    • &&
    • ||
    • x ? y : z (partial support, see known bugs)
  • #if, #ifdef, #ifndef, #elif, #else, #endif
  • Stringizing operator #
  • Token pasting operator ##
  • #pragma once, a very common extension

Additionally implemented by pcpp command line tool:

  • #error (default implementation prints to stderr and increments the exit code)
  • #warning (default implementation prints to stderr)

Not implemented yet (donations of code welcome):

  • #pragma anything other than once.
  • _Pragma used to emit preprocessor calculated #pragma.
  • #line num, num “file” and NUMBER FILE.

Known bugs (ordered from worst to least worst):

Expression evaluation is a bit broken

Currently #if expressions are evaluated by converting them into Python expressions and calling eval() on them. This works surprisingly well most of the time, but because Python is not C, corner cases break. These are the known such broken corner cases:

  • Unary operator evaluation will break for evil expressions such as -!+!9 because logical NOT in Python results in a boolean, not an integer, and a unary plus or negative boolean is invalid syntax in Python
  • Similarly expressions which assume that boolean operations output either a zero or a one will fail e.g. (2 || 3) == 0
  • Python has no concept of an unsigned integer and C expressions relying on unsigned integer semantics will fail badly e.g. -1 <= 0U is supposed to be evaluated as false in the C preprocessor, but it will be evaluated as true under this implementation. To be honest if your preprocessor logic is relying on those sorts of behaviours, you should rewrite it.
  • Without a back tracking parser, the C ternary operator is hard to accurately convert into a Python ternary operation, so you need to help it by using one of these two forms:
    • (x) ? y : z (z gets evaluated according to Python not C precedence)
    • (x ? y : z) (preferred, evaluates correctly, we inject brackets around the subexpessions before sending to Python)

A proper lexing parser based on’s yacc module has been started and can be found in the branch. Time to complete it, is the problem, and any pull requests helping with that are welcome.

We do not pass the Boost.Wave preprocessor test suite
A lot of bugs have been fixed since this was reported, however the chances are that pcpp still doesn’t pass it. A TODO is to port the Wave test suite to Python and find out how bad things are. We suspect that any failures will be in highly estoric use cases i.e. known illegal input. If you only use valid input then we expect you generally won’t have trouble.

Customising your own preprocessor:

See the API reference docs at

You can find an example of overriding the on_*() processing hooks at


v1.21 (30th September 2019):

  • Fix bug where token pasting two numeric tokens did not yield a numeric token. Thanks to Sei-Lisa for reporting this.

  • BREAKING CHANGE: Paths emitted by pcpp into #line directives now are relative to the working directory of the process when Preprocessor is initialised. This includes added search paths - files included from those locations will be emitted with a sequence of ../ to relativise the path emitted. If no path exists between the working directory and the path of the file being emitted, an absolute path is emitted instead.

    If you wish to disable this new behaviour, or use different behaviour, you can customise the new rewrite_paths member variable of Preprocessor.

  • Fix bug where __LINE__ was expanding into the line number of its definition instead of its use. Thanks to Sei-Lisa for reporting this.

  • Add --passthru-magic-macros command line option.

  • BREAKING CHANGE: The PreprocessorHooks and OutputDirective interface has changed. One now must specify the kind of OutputDirective abort one wants, and one can now both ignore AND remove directives. on_directive_handle() and on_directive_unknown() now take an extra parameter precedingtoks, these are the tokens from the # up to the directive.

  • Fix a corner case where FUNC(void)foo() expanded to voidfoo() and not void foo() which is a very common non-conforming extension of the C preprocessor. Thanks to OmegaDoom for reporting this.

  • Add tokens for all the C operators, to help implementation of an expression evaluator.

  • Updated embedded ply to HEAD (2019-04-25)

  • Fix #include not working if no -I parameters were supplied. Thanks to csm10495 for reporting this.

v1.20 (7th January 2019):

  • Now supports character literals in expressions. Thanks to untaugh for the pull request adding this.
  • Stopped the default collapsing of whitespace in output, and made it optional via a new command line option --compress.
  • Fixed extraneous whitespace in --passthru-comments caused by multiline comments. Thanks to p2k for reporting this.
  • Fixed bug where defining a macro via string did not set the source attribute in the token. Thanks to ZedThree for reporting this.
  • Stop triggering an exception when no arguments are supplied to pcpp. Thanks to virtuald for reporting this.
  • Rebase onto PLY latest from Dec 28th 2018 ( Latest PLY no longer works using pypi packaging, David wants people to include the source of PLY directly. pcpp does this via a git submodule, and has setuptools bundle the submodule.
  • Add a formal LICENSE.txt file, as requested by Sei-Lisa.
  • Fix failure to issue #line directive for first include file in a file. Thanks to Sei-Lisa for reporting this.

v1.1 (19th June 2018):

  • Added the --times and --filetimes features.
  • Fix bug where macros containing operator defined were not being expanded properly.
  • Added the ability to accept multiple inputs, they are concatenated into the output.
  • Fix bug where lines beginning with # and no contents caused an internal preprocessor error.
  • Fix bug where the macro expansion par par##ext was expanding into parext parext.

v1.01 (21st Feb 2018):

  • Fix bug where in pass through mode, an #elif in an #if block inside an #if block in ifpassthru was failing to be passed through.
  • Downgraded failure to evaluate an expression to a warning.
  • Fix missing Readme.rst in pypi package.

v1.00 (13th Mar 2017):

First release

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for pcpp, version 1.21
Filename, size File type Python version Upload date Hashes
Filename, size pcpp-1.21-py2.py3-none-any.whl (80.7 kB) File type Wheel Python version py2.py3 Upload date Hashes View
Filename, size pcpp-1.21.tar.gz (84.1 kB) File type Source Python version None Upload date Hashes View

Supported by

Pingdom Pingdom Monitoring Google Google Object Storage and Download Analytics Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN DigiCert DigiCert EV certificate StatusPage StatusPage Status page