The newer version of this question is located here: Cyther: The Cross Platform Cython/Python Compiler (Take 2)

I am currently writing a Python library (soon to be published)that automatically compiles a Cython/Python file to pure -O3 C, without the user having to do anything other than:

C:\Project_dir\> cyther example.pyx

C:\Project_dir\> cyther example.py

from cyther import main

My 'auto-compiler' is named Cyther, and it organizes and polymorphesizes Cython and gcc together in an attempt to make a platform independent, stable, and easy to use compiler.

Cyther was specifically designed to combat the vcvarsall.bat not found error on windows, but still work on every other system with exactly the same performance.

I need to make sure that it is completely cross-platform.

I've tried my best, but alas, I may have slipped up here and there. I am also not very familiar with the oddities of other operating systems. I was thinking about using a Virtual Box to run a few other operating systems to test Cyther on, but I can only get so far by trial and error.

The issue: Cyther makes a few assumptions about your system that I'm not sure how to handle:

  1. Your environment path variable is named 'PATH'
  2. Cython, Python and gcc are all '.exe'
  3. libpythonXY.a exists in your libs directory
  4. mingw32 is up to date with the latest Python release (Windows)
  5. 'cyther' is callable from the command line
  6. Any gcc compiled C program will work on Windows

__author__ = 'Nicholas C. Pandolfi'

license = '''
Copyright (c) 2016 Nicholas C Pandolfi ALL RIGHTS RESERVED (MIT)

Permission is hereby granted, free of charge, to any person
obtaining a copy of this software and associated documentation
files (the "Software"), to deal in the Software without
restriction, including without limitation the rights to use,
copy, modify, merge, publish, distribute, sublicense, and/or
sell copies of the Software, and to permit persons to whom the
Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall
be included in all copies or substantial portions of the Software.


import os
import sys
import subprocess

import argparse
import platform
import errno

import time

#The system constants
ver = str(sys.version_info.major) + str(sys.version_info.minor)
home = sys.exec_prefix
libs = os.path.join(home, 'libs')
include = os.path.join(home, 'include')
a_lib = 'python' + ver
is_windows = platform.platform().split('-')[0] == 'Windows'  # This line and the one below it are delicate
default_output_extension = '.pyd' if is_windows else '.so'


assumptions = """
Assumptions cyther makes about your system:

1) When you put the -z flag to print this very string:
you must also have the filename specified or else it will return an error
2) Your environment path variable is named 'PATH'
3) Cython, Python and gcc are all .exe
4) mingw32 is up to date with the latest python release (windows)
5) 'cyther' is callable from the command line
6) Any gcc compiled c program will work on windows

class CytherError(Exception):
    def __init__(self, message):
        Exception.__init__(self, message)
        self.message = 'CytherError: {}'.format(repr(message))

#Holds each file's attributes
class AttributeHolder:

#My best shot at a raceless make directory function
def raceless_mkdir(directory):
    if not os.path.exists(directory):
        except OSError as error:
            if error.errno != errno.EEXIST:

#gets all files in a specific directory
def get_files(d):
    files = []
    for file in os.listdir(d):
        if os.path.isfile(os.path.join(d, file)):
    return files

#find the program 'program' in the system path
def which(program):
    def is_exe(filename):
        return os.path.isfile(filename) and os.access(filename, os.X_OK)

    filename = os.path.split(program)[0]
    if filename:
        if is_exe(program):
            return program
        for path in os.environ["PATH"].split(os.pathsep):
            path = path.strip('"')
            exe_file = os.path.join(path, program)
            if is_exe(exe_file):
                return exe_file
    return None

#For each file, this constructs all build names and necessary variables to pass along with it to be made
def process_files(args):
    to_process = []
    for filename in args.filenames:
        file = AttributeHolder()
        if os.path.exists(filename) and (filename not in os.listdir(os.getcwd())):
            file.file_path = filename
        elif os.path.exists(os.path.join(os.getcwd(), filename)):
            file.file_path = os.path.join(os.getcwd(), filename)
            raise CytherError("The file '{}' does not exist".format(filename))
        file.file_base_name = os.path.splitext(os.path.basename(file.file_path))[0]
        file.no_extension, file.extension = os.path.splitext(file.file_path)
        if file.extension not in ('.pyx', '.py'):
            raise CytherError("The file '{}' is not a designated cython file".format(file.file_path))
        base_path = os.path.dirname(file.file_path)
        local_build = args.local
        if not local_build:
            cache_name = os.path.join(base_path, '__cythercache__')
            file.c_name = os.path.join(cache_name, file.file_base_name) + '.c'
            file.c_name = file.no_extension + '.c'
        if not args.output_name:
            file.output_name = file.no_extension + default_output_extension
            file.output_name = args.output_name
            if not os.path.exists(os.path.dirname(file.output_name)):
                raise CytherError('The directory specified to write the output file in does not exist')
        file.stamp_if_error = 0
    return to_process

#Figures out whether the source code has a corresponding compile
def should_compile(file):
    if os.path.exists(file.output_name):
        source_time = os.path.getmtime(file.file_path)
        output_time = os.path.getmtime(file.output_name)
        if source_time > output_time:
            return True
        return True
    return False

def core(args, file):
    #Recieves the pass off args
    cython_args = args.cython_args if args.cython_args else []
    cython_pass_off = []

    gcc_args = args.gcc_args if args.gcc_args else []
    gcc_pass_off = []

    #This command and the rest contruct the arguments to use for compiling given a preset
    preset = args.preset
    if not preset:
        preset = 'ninja'#Default

    if preset == 'ninja':
        cython_command = ['cython', '-a', '-p', '-o', file.c_name, file.file_path]
        gcc_command = ['gcc', '-shared', '-w', '-O3', '-I', include, '-L', libs, '-o', file.output_name, file.c_name,
                       '-l', a_lib]
    elif preset == 'beast':
        cython_command = ['cython', '-a', '-l', '-p', '-o', file.c_name, file.file_path]
        gcc_command = ['gcc', '-shared', '-Wall', '-O3', '-I', include, '-L', libs, '-o', file.output_name,
                       file.c_name, '-l', a_lib]
    elif preset == 'minimal':
        cython_command = ['cython', '-o', file.c_name, file.file_path]
        gcc_command = ['gcc', '-shared', '-I', include, '-L', libs, '-o', file.output_name, file.c_name, '-l', a_lib]
        raise CytherError("The format '{}' is not supported".format(preset))

    #Process cython and gcc pass off args
    for offset, process in enumerate((cython_args, gcc_args)):
        for arg in process:
            if arg[0] == '_':
                if arg[1] == '_':
                    arg = '--' + arg[2:]
                    arg = '-' + arg[1:]
            if offset:

    #This filters the args to give to cython
    if cython_pass_off:
        for item in cython_pass_off:
            if item[0] == '-':
                if item not in cython_command:

    #This filters the args to give to gcc
    if gcc_pass_off:
        for item in gcc_pass_off:
            if item[0] == '-':
                if item not in gcc_command:

    if not args.skip:
        #Figure out if the selected software is installed
        python_found = which('python.exe')
        cython_found = which('cython.exe')
        gcc_found = which('gcc.exe')

        #Raise errors if software is not installed
        if not python_found:
            raise CytherError("Python is not able to be called, please add it to the system's path")
        if not cython_found:
                import cython
                raise CytherError("Cython exists and is able to be imported by Python, " + \
                                  "however it is not in the system path. Please add it.")
            except ImportError:
                raise CytherError("Cython is unable to be imported, and is probably not installed")
        if not gcc_found:
            raise CytherError("gcc is not able to be called, please add it to the system's path")

    print(' '.join(cython_command).strip())
    print(' '.join(gcc_command).strip())

    cython_error = subprocess.call(cython_command)
    string = str(cython_error)
    if cython_error:
        if args.watch:
            return -42
            raise CytherError(string)

    gcc_error = subprocess.call(gcc_command)
    string = str(cython_error)
    if gcc_error:
        if args.watch:
            return -42
            raise CytherError(string)

    return 1

def main(args):
    numfiles = len(args.filenames)
    interval = INTERVAL / numfiles

    if args.assumptions:
        return 0

    if type(args) == str:
        args = parser.parse_args(args.split(' '))
    elif type(args) == argparse.Namespace:
        pass#dont do anything
        raise CytherError("Args must be a instance of str or argparse.Namespace, not '{}'".format(str(type(args))))

    files = process_files(args)

    if not args.timestamp and args.watch:
        args.timestamp = True

    while True:
        for file in files:
            if args.timestamp:
                if should_compile(file) and os.path.getmtime(file.file_path) > file.stamp_if_error:
                    if args.watch:
                        if len(args.filenames) > 1:
                            print("Compiling the file '{}'".format(file.file_path))
                            print('Compiling the file')
                        pass#dont print anything, its a single file
                    ret = core(args, file)
                    if ret == -42:
                        file.stamp_if_error = time.time()
                    if args.watch:
                    if not args.watch:
                        if len(args.filenames) > 1:
                            print("Skipping the file '{}'".format(file.file_path))
                            print('Skipping compilation')
                        pass#dont print anything, its a single file

        if not args.watch:

help_filenames = 'The Cython source file'
help_preset = 'The preset options for using cython and gcc (ninja, verbose, beast)'
help_timestamp = 'If this flag is provided, cyther will not compile files that have a modified time before that of your compiled .pyd or .so files'
help_output = 'Change the name of the output file, default is basename plus .pyd'
help_assumptions = 'Print the list of assumptions cyther makes about your system before running'
help_skip = 'Skip the checking procedures that make sure that the right software is installed on your system (saves significant time if everything has already been checked, but not recommended)'
help_local = 'When not flagged, builds in __omicache__, when flagged, it builds locally in the same directory'
help_watch = 'When given, cyther will watch the directory with the \'t\' option implied and compile, when necessary, the files given'
help_cython = "Arguments to pass to Cython (use '_' or '__' instead of '-' or '--'"
help_gcc = "Arguments to pass to gcc (use '_' or '__' instead of '-' or '--'"

#All the command line processing statements
parser = argparse.ArgumentParser(description = 'Auto compile and build .pyx files in place.', usage = 'cyther [options] input_file')
parser.add_argument('filenames', action = 'store', nargs = '+', type = str, help = help_filenames)
parser.add_argument('-p', '--preset', action = 'store', type = str, default = '', dest = 'preset', help = help_preset)
parser.add_argument('-t', '--timestamp', action = 'store_true', default = False, dest = 'timestamp', help = help_timestamp)
parser.add_argument('-o', '--output', action = 'store', dest = 'output_name', type = str, help = help_output)
parser.add_argument('-z', '--assumptions', action = 'store_true', default = False, dest = 'assumptions', help = help_assumptions)
parser.add_argument('-skip', action = 'store_true', default = False, help = help_skip)
parser.add_argument('-l', '--local', action = 'store_true', dest = 'local', default = False, help = help_local)
parser.add_argument('-w', '--watch', action = 'store_true', dest = 'watch', default = False, help = help_watch)
parser.add_argument('-cython', action = 'store', nargs = '+', dest = 'cython_args', help = help_cython)
parser.add_argument('-gcc', action = 'store', nargs = '+', dest = 'gcc_args', help = help_gcc)

command_line_args = parser.parse_args()
#print the assumptions in the help text

if __name__ == '__main__':

Cyther has been published on pypi.

  • 2
    \$\begingroup\$ Please avoid editing your question with updated code. That invalidates the answers. Feel free to link to your new work, or even make a new question if you want a second review, but please avoid adding the new code into the question. Read What to do when I receive answers in the help centre for more information. \$\endgroup\$ – Quill Mar 18 '16 at 1:42
  • \$\begingroup\$ Thats why I put two copies of the code, a old and a new \$\endgroup\$ – Nick Pandolfi Mar 18 '16 at 1:43
  • \$\begingroup\$ Two copies or not, the rule still applies. \$\endgroup\$ – Quill Mar 18 '16 at 1:43
  • \$\begingroup\$ Ok, sorry, I didn't know that. Thanks though! \$\endgroup\$ – Nick Pandolfi Mar 18 '16 at 1:45

I will give a quick overview of some aspects that I found at a first glance. Unfortunately, I could not spare the time to do an in-depth analysis of the cross-plattform capabilites and/or thorough testing of your program. As I find your idea very interesting, I maybe check back later with more feedback.

Edit: Updated requirements for proper import and added some thoughts on the given assumptions.

Magic values

On the first sight, I was able to spot some magic values (like -42), which maybe should be replaced with a CONSTANT with a nice name (GCC_ERROR?).

Module level constants

The values of assumptions and the helper_* are constants in this module. It is common practice in Python to write such module level constants in ALL_UPPERCASE_LETTERS.

Missing docstrings

The next thing that caught my eye was a sincere lack of class/method/function docstrings. As the code (maybe) will be available to the public, assume someone will mess around with it. You may help others a lot if you tell them a little bit about each function and maybe the inputs it expects. The Python bible PEP257 is a good reference on this topic.

In addition some comments in the code may help to split longer functions into logical groups, which greatly simplifies rewiews and reworks of the code.

Redefinition of built-ins

With a little help from Spyder's synthax highlighting, I was able to find two redifintions of Python built-ins: license and file (talking from a Python 2.7 point of view, maybe that has changed). Refering to a SO answer, __license__ would be the way to go on this, whereas an alternative for file is left as an exercise to the dear reader.

Space around keyword arguments

Following the infamous PEP8, there should be no spaces around keyword arguments. As an example

parser = argparse.ArgumentParser(description = '...', usage = '...')

would end up as

parser = argparse.ArgumentParser(description='...', usage='...')


From my experience it is more common to check the type of an object with isintance instead of type. With that in mind type(args) == str becomes isinstance(args, str), for example.


One could move the whole parser initialization into the if __name__ == 'main': block. If you want to import your program, you will have to move command_line_args = parser.parse_args() to that block. Otherwise the parser will be started on import, which will cause an error because the script is called with no arguments.

The assumptions

1) Your environment path variable is named 'PATH'

All Linux and Windows systems I have worked on had an enviroment variable called PATH. It is an educated guess, but from my point of view that assumptions is valid.

2) Cython, Python and gcc are all '.exe'

This may be a bit problematic. Linux systems do not have .exe files usually. I can start the Python interpreter on Linux and Windows by typing python into the console if the executable is in PATH, but that may be caused by my setup. Maybe that can be exploited to check the pressence of Python and co.

3) libpythonXY.a exists in your libs directory

Linux seems to prefer .so as file extension for libpython (see this to the Ubuntu package repository for example)

4) mingw32 is up to date with the latest python release (windows)

I use Python library collections (Python(x,y) and/or Anaconda), of which at least one (to the best of my knowledge) comes bundled with its own version of MinGW. There is at least one more MinGW on my system. Following Murphy's Law, that may (and probably will) cause some hiccup, depending on which MinGW installation comes first in PATH.

5) 'cyther' is callable from the command line

You will have to take care the cyther is executable then. SO and the internet will give you a lot of advice how this can reached.

6) Any gcc compiled c program will work on windows

My experience says: "Not gonna happen!". Sorry. But I wish you the best of luck thay "any" will be "almost any".


This isn't really organized nor does it have any kind of theme - I just started from the top and went to the bottom, commenting as I went. I also agree with everything listed in Alex Vorndran's answer and didn't repeat anything he said (I think). I will say that this seems to be somewhat far off from actually being highly portable.


You could probably make this less fragile if you wrote something like

is_windows = platform.platform().strip().lower().startswith('windows')

Then you're a little less susceptible to format changes IMO. It would probably be safe to lose the .strip() call as well.


Prefer super to Exception.__init__

super(Exception, self).__init__(message)

It is also super generic - I'm not sure that it actually helps except to identify that the error is coming from Cyther, which I think they should have figured out on their own. Either make more specific errors that inherit this, or just use the standard library exceptions.


If you aren't actually going to give this any values or control what attributes it will have, just use a dictionary instead.


I think you've increased the likelihood of race conditions by adding the os.path.exists check. Just try to make the directory, then move on. I also question that your program should be making directories (this always makes me nervous) that might already exist - either figure out something that you know only yours will ever make (prefix it somehow) or error out if they don't have the directory themselves.


Prefer a list/generator comprehension here, or just a generator

def get_files(d):
    return [file_ for file_ in os.listdir(d) 
            if os.path.isfile(os.path.join(d, file_))]


def get_files(d):
    for file_ in os.listdir(d):
        if os.path.isfile(os.path.join(d, file_)):
            yield file_
    raise StopIteration


Why is is_exe a closure? I don't see any benefit there. I also don't like that you return None - raising an exception seems more logical here.


This could use a lot of whitespace to help with readability.

if file.extension not in ('.pyx', '.py') - You might be surprised at how often people use non-standard extensions, and even more surprised at how often they have decent reasons for it. You might want to make this the default, but make it configurable/extensible. It should probably also be a named constant.

You don't need to assign to a local variable local_build - just use args.local

I think this would be much better as a generator - less of a memory footprint, it won't hang forever building up a potentially large list, etc.


There is no indication as to why these files should or should not compile - is it just checking if they're out of date? If so, then the name should reflect that.


Anytime I see something named core (or similar) that isn't just a few lines I cringe - you're doing too much here. Break it up into smaller functions for readability and maintainability.

You also have lots of magic values here - what, if anything, do they signify? This should probably be documented somewhere.

Is there a use-case for people who want to use your program but not gcc? You don't support that atm.


You have some redundancy in your loop.

 if not args.watch:
     if len(args.filenames) > 1:
         print("Skipping the file '{}'".format(file.file_path))
         print('Skipping compilation')
     pass#dont print anything, its a single file

First of all, leave some space before a comment so it can be read. Secondly, just leave this as

 if not args.watch:
     if len(args.filenames) > 1:
         print("Skipping the file '{}'".format(file.file_path))
         print('Skipping compilation')


I really don't like -long_name flags - prefer -l and --long_name for consistency with most people's expectations, or just use long_name as a subparser.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.