docs/sphinxext/apigen.py

"""Attempt to generate templates for module reference with Sphinx

XXX - we exclude extension modules

To include extension modules, first identify them as valid in the
``_uri2path`` method, then handle them in the ``_parse_module`` script.

We get functions and classes by parsing the text of .py files.
Alternatively we could import the modules for discovery, and we'd have
to do that for extension modules.  This would involve changing the
``_parse_module`` method to work via import and introspection, and
might involve changing ``discover_modules`` (which determines which
files are modules, and therefore which module URIs will be passed to
``_parse_module``).

NOTE: this is a modified version of a script originally shipped with the
PyMVPA project, which we've adapted for NIPY use.  PyMVPA is an MIT-licensed
project."""


# Stdlib imports
import ast
import inspect
import os
import re
from importlib import import_module


class Obj(object):
    '''Namespace to hold arbitrary information.'''
    def __init__(self, **kwargs):
        for k, v in kwargs.items():
            setattr(self, k, v)
                                        
class FuncClsScanner(ast.NodeVisitor):
    """Scan a module for top-level functions and classes.
    
    Skips objects with an @undoc decorator, or a name starting with '_'.
    """
    def __init__(self):
        ast.NodeVisitor.__init__(self)
        self.classes = []
        self.classes_seen = set()
        self.functions = []
    
    @staticmethod
    def has_undoc_decorator(node):
        return any(isinstance(d, ast.Name) and d.id == 'undoc' \
                                for d in node.decorator_list)

    def visit_If(self, node):
        if isinstance(node.test, ast.Compare) \
                and isinstance(node.test.left, ast.Name) \
                and node.test.left.id == '__name__':
            return   # Ignore classes defined in "if __name__ == '__main__':"

        self.generic_visit(node)
    
    def visit_FunctionDef(self, node):
        if not (node.name.startswith('_') or self.has_undoc_decorator(node)) \
                and node.name not in self.functions:
            self.functions.append(node.name)
    
    def visit_ClassDef(self, node):
        if not (node.name.startswith('_') or self.has_undoc_decorator(node)) \
                and node.name not in self.classes_seen:
            cls = Obj(name=node.name)
            cls.has_init = any(isinstance(n, ast.FunctionDef) and \
                                n.name=='__init__' for n in node.body)
            self.classes.append(cls)
            self.classes_seen.add(node.name)
    
    def scan(self, mod):
        self.visit(mod)
        return self.functions, self.classes

# Functions and classes
class ApiDocWriter(object):
    ''' Class for automatic detection and parsing of API docs
    to Sphinx-parsable reST format'''

    # only separating first two levels
    rst_section_levels = ['*', '=', '-', '~', '^']

    def __init__(self,
                 package_name,
                 rst_extension='.rst',
                 package_skip_patterns=None,
                 module_skip_patterns=None,
                 names_from__all__=None,
                 ):
        ''' Initialize package for parsing

        Parameters
        ----------
        package_name : string
            Name of the top-level package.  *package_name* must be the
            name of an importable package
        rst_extension : string, optional
            Extension for reST files, default '.rst'
        package_skip_patterns : None or sequence of {strings, regexps}
            Sequence of strings giving URIs of packages to be excluded
            Operates on the package path, starting at (including) the
            first dot in the package path, after *package_name* - so,
            if *package_name* is ``sphinx``, then ``sphinx.util`` will
            result in ``.util`` being passed for earching by these
            regexps.  If is None, gives default. Default is:
            ['\\.tests$']
        module_skip_patterns : None or sequence
            Sequence of strings giving URIs of modules to be excluded
            Operates on the module name including preceding URI path,
            back to the first dot after *package_name*.  For example
            ``sphinx.util.console`` results in the string to search of
            ``.util.console``
            If is None, gives default. Default is:
            ['\\.setup$', '\\._']
        names_from__all__ : set, optional
            Modules listed in here will be scanned by doing ``from mod import *``,
            rather than finding function and class definitions by scanning the
            AST. This is intended for API modules which expose things defined in
            other files. Modules listed here must define ``__all__`` to avoid
            exposing everything they import.
        '''
        if package_skip_patterns is None:
            package_skip_patterns = ['\\.tests$']
        if module_skip_patterns is None:
            module_skip_patterns = ['\\.setup$', '\\._']
        self.package_name = package_name
        self.rst_extension = rst_extension
        self.package_skip_patterns = package_skip_patterns
        self.module_skip_patterns = module_skip_patterns
        self.names_from__all__ = names_from__all__ or set()

    def get_package_name(self):
        return self._package_name

    def set_package_name(self, package_name):
        ''' Set package_name

        >>> docwriter = ApiDocWriter('sphinx')
        >>> import sphinx
        >>> docwriter.root_path == sphinx.__path__[0]
        True
        >>> docwriter.package_name = 'docutils'
        >>> import docutils
        >>> docwriter.root_path == docutils.__path__[0]
        True
        '''
        # It's also possible to imagine caching the module parsing here
        self._package_name = package_name
        self.root_module = import_module(package_name)
        self.root_path = self.root_module.__path__[0]
        self.written_modules = None

    package_name = property(get_package_name, set_package_name, None,
                            'get/set package_name')

    def _uri2path(self, uri):
        ''' Convert uri to absolute filepath

        Parameters
        ----------
        uri : string
            URI of python module to return path for

        Returns
        -------
        path : None or string
            Returns None if there is no valid path for this URI
            Otherwise returns absolute file system path for URI

        Examples
        --------
        >>> docwriter = ApiDocWriter('sphinx')
        >>> import sphinx
        >>> modpath = sphinx.__path__[0]
        >>> res = docwriter._uri2path('sphinx.builder')
        >>> res == os.path.join(modpath, 'builder.py')
        True
        >>> res = docwriter._uri2path('sphinx')
        >>> res == os.path.join(modpath, '__init__.py')
        True
        >>> docwriter._uri2path('sphinx.does_not_exist')

        '''
        if uri == self.package_name:
            return os.path.join(self.root_path, '__init__.py')
        path = uri.replace('.', os.path.sep)
        path = path.replace(self.package_name + os.path.sep, '')
        path = os.path.join(self.root_path, path)
        # XXX maybe check for extensions as well?
        if os.path.exists(path + '.py'): # file
            path += '.py'
        elif os.path.exists(os.path.join(path, '__init__.py')):
            path = os.path.join(path, '__init__.py')
        else:
            return None
        return path

    def _path2uri(self, dirpath):
        ''' Convert directory path to uri '''
        relpath = dirpath.replace(self.root_path, self.package_name)
        if relpath.startswith(os.path.sep):
            relpath = relpath[1:]
        return relpath.replace(os.path.sep, '.')

    def _parse_module(self, uri):
        ''' Parse module defined in *uri* '''
        filename = self._uri2path(uri)
        if filename is None:
            # nothing that we could handle here.
            return ([],[])
        with open(filename, 'rb') as f:
            mod = ast.parse(f.read())
        return FuncClsScanner().scan(mod)

    def _import_funcs_classes(self, uri):
        """Import * from uri, and separate out functions and classes."""
        ns = {}
        exec('from %s import *' % uri, ns)
        funcs, classes = [], []
        for name, obj in ns.items():
            if inspect.isclass(obj):
                cls = Obj(name=name, has_init='__init__' in obj.__dict__)
                classes.append(cls)
            elif inspect.isfunction(obj):
                funcs.append(name)

        return sorted(funcs), sorted(classes, key=lambda x: x.name)

    def find_funcs_classes(self, uri):
        """Find the functions and classes defined in the module ``uri``"""
        if uri in self.names_from__all__:
            # For API modules which expose things defined elsewhere, import them
            return self._import_funcs_classes(uri)
        else:
            # For other modules, scan their AST to see what they define
            return self._parse_module(uri)

    def generate_api_doc(self, uri):
        '''Make autodoc documentation template string for a module

        Parameters
        ----------
        uri : string
            python location of module - e.g 'sphinx.builder'

        Returns
        -------
        S : string
            Contents of API doc
        '''
        # get the names of all classes and functions
        functions, classes = self.find_funcs_classes(uri)
        if not len(functions) and not len(classes):
            #print ('WARNING: Empty -', uri)  # dbg
            return ''

        # Make a shorter version of the uri that omits the package name for
        # titles
        uri_short = re.sub(r'^%s\.' % self.package_name,'',uri)

        ad = '.. AUTO-GENERATED FILE -- DO NOT EDIT!\n\n'

        # Set the chapter title to read 'Module:' for all modules except for the
        # main packages
        if '.' in uri:
            chap_title = 'Module: :mod:`' + uri_short + '`'
        else:
            chap_title = ':mod:`' + uri_short + '`'
        ad += chap_title + '\n' + self.rst_section_levels[1] * len(chap_title)

        ad += '\n.. automodule:: ' + uri + '\n'
        ad += '\n.. currentmodule:: ' + uri + '\n'
        
        if classes:
            subhead = str(len(classes)) + (' Classes' if len(classes) > 1 else ' Class')
            ad += '\n'+ subhead + '\n' + \
                  self.rst_section_levels[2] * len(subhead) + '\n'

        for c in classes:
            ad += '\n.. autoclass:: ' + c.name + '\n'
            # must NOT exclude from index to keep cross-refs working
            ad += '  :members:\n' \
                  '  :show-inheritance:\n'
            if c.has_init:
                  ad += '\n  .. automethod:: __init__\n'
        
        if functions:
            subhead = str(len(functions)) + (' Functions' if len(functions) > 1 else ' Function')
            ad += '\n'+ subhead + '\n' + \
                  self.rst_section_levels[2] * len(subhead) + '\n'
        for f in functions:
            # must NOT exclude from index to keep cross-refs working
            ad += '\n.. autofunction:: ' + uri + '.' + f + '\n\n'
        return ad

    def _survives_exclude(self, matchstr, match_type):
        ''' Returns True if *matchstr* does not match patterns

        ``self.package_name`` removed from front of string if present

        Examples
        --------
        >>> dw = ApiDocWriter('sphinx')
        >>> dw._survives_exclude('sphinx.okpkg', 'package')
        True
        >>> dw.package_skip_patterns.append('^\\.badpkg$')
        >>> dw._survives_exclude('sphinx.badpkg', 'package')
        False
        >>> dw._survives_exclude('sphinx.badpkg', 'module')
        True
        >>> dw._survives_exclude('sphinx.badmod', 'module')
        True
        >>> dw.module_skip_patterns.append('^\\.badmod$')
        >>> dw._survives_exclude('sphinx.badmod', 'module')
        False
        '''
        if match_type == 'module':
            patterns = self.module_skip_patterns
        elif match_type == 'package':
            patterns = self.package_skip_patterns
        else:
            raise ValueError('Cannot interpret match type "%s"'
                             % match_type)
        # Match to URI without package name
        L = len(self.package_name)
        if matchstr[:L] == self.package_name:
            matchstr = matchstr[L:]
        for pat in patterns:
            try:
                pat.search
            except AttributeError:
                pat = re.compile(pat)
            if pat.search(matchstr):
                return False
        return True

    def discover_modules(self):
        ''' Return module sequence discovered from ``self.package_name``


        Parameters
        ----------
        None

        Returns
        -------
        mods : sequence
            Sequence of module names within ``self.package_name``

        Examples
        --------
        >>> dw = ApiDocWriter('sphinx')
        >>> mods = dw.discover_modules()
        >>> 'sphinx.util' in mods
        True
        >>> dw.package_skip_patterns.append('\\.util$')
        >>> 'sphinx.util' in dw.discover_modules()
        False
        >>>
        '''
        modules = [self.package_name]
        # raw directory parsing
        for dirpath, dirnames, filenames in os.walk(self.root_path):
            # Check directory names for packages
            root_uri = self._path2uri(os.path.join(self.root_path,
                                                   dirpath))
            for dirname in dirnames[:]: # copy list - we modify inplace
                package_uri = '.'.join((root_uri, dirname))
                if (self._uri2path(package_uri) and
                    self._survives_exclude(package_uri, 'package')):
                    modules.append(package_uri)
                else:
                    dirnames.remove(dirname)
            # Check filenames for modules
            for filename in filenames:
                module_name = filename[:-3]
                module_uri = '.'.join((root_uri, module_name))
                if (self._uri2path(module_uri) and
                    self._survives_exclude(module_uri, 'module')):
                    modules.append(module_uri)
        return sorted(modules)

    def write_modules_api(self, modules,outdir):
        # write the list
        written_modules = []
        for m in modules:
            api_str = self.generate_api_doc(m)
            if not api_str:
                continue
            # write out to file
            outfile = os.path.join(outdir, m + self.rst_extension)
            with open(outfile, "wt", encoding="utf-8") as fileobj:
                fileobj.write(api_str)
            written_modules.append(m)
        self.written_modules = written_modules

    def write_api_docs(self, outdir):
        """Generate API reST files.

        Parameters
        ----------
        outdir : string
            Directory name in which to store files
            We create automatic filenames for each module

        Returns
        -------
        None

        Notes
        -----
        Sets self.written_modules to list of written modules
        """
        if not os.path.exists(outdir):
            os.mkdir(outdir)
        # compose list of modules
        modules = self.discover_modules()
        self.write_modules_api(modules,outdir)

    def write_index(self, outdir, path='gen.rst', relative_to=None):
        """Make a reST API index file from written files

        Parameters
        ----------
        outdir : string
            Directory to which to write generated index file
        path : string
            Filename to write index to
        relative_to : string
            path to which written filenames are relative.  This
            component of the written file path will be removed from
            outdir, in the generated index.  Default is None, meaning,
            leave path as it is.
        """
        if self.written_modules is None:
            raise ValueError('No modules written')
        # Get full filename path
        path = os.path.join(outdir, path)
        # Path written into index is relative to rootpath
        if relative_to is not None:
            relpath = outdir.replace(relative_to + os.path.sep, '')
        else:
            relpath = outdir
        with open(path, "wt", encoding="utf-8") as idx:
            w = idx.write
            w('.. AUTO-GENERATED FILE -- DO NOT EDIT!\n\n')
            w('.. autosummary::\n'
            '   :toctree: %s\n\n' % relpath)
            for mod in self.written_modules:
                w('   %s\n' % mod)