Current File : //opt/imunify360/venv/lib64/python3.11/site-packages/docutils/transforms/peps.py
# $Id: peps.py 9037 2022-03-05 23:31:10Z milde $
# Author: David Goodger <goodger@python.org>
# Copyright: This module has been placed in the public domain.

"""
Transforms for PEP processing.

- `Headers`: Used to transform a PEP's initial RFC-2822 header.  It remains a
  field list, but some entries get processed.
- `Contents`: Auto-inserts a table of contents.
- `PEPZero`: Special processing for PEP 0.
"""

__docformat__ = 'reStructuredText'

import os
import re
import time
from docutils import nodes, utils, languages
from docutils import DataError
from docutils.transforms import Transform
from docutils.transforms import parts, references, misc


class Headers(Transform):

    """
    Process fields in a PEP's initial RFC-2822 header.
    """

    default_priority = 360

    pep_url = 'pep-%04d'
    pep_cvs_url = ('http://hg.python.org'
                   '/peps/file/default/pep-%04d.txt')
    rcs_keyword_substitutions = (
          (re.compile(r'\$' r'RCSfile: (.+),v \$$', re.IGNORECASE), r'\1'),
          (re.compile(r'\$[a-zA-Z]+: (.+) \$$'), r'\1'),)

    def apply(self):
        if not len(self.document):
            # @@@ replace these DataErrors with proper system messages
            raise DataError('Document tree is empty.')
        header = self.document[0]
        if (not isinstance(header, nodes.field_list)
            or 'rfc2822' not in header['classes']):
            raise DataError('Document does not begin with an RFC-2822 '
                            'header; it is not a PEP.')
        pep = None
        for field in header:
            if field[0].astext().lower() == 'pep':  # should be the first field
                value = field[1].astext()
                try:
                    pep = int(value)
                    cvs_url = self.pep_cvs_url % pep
                except ValueError:
                    pep = value
                    cvs_url = None
                    msg = self.document.reporter.warning(
                        '"PEP" header must contain an integer; "%s" is an '
                        'invalid value.' % pep, base_node=field)
                    msgid = self.document.set_id(msg)
                    prb = nodes.problematic(value, value or '(none)',
                                            refid=msgid)
                    prbid = self.document.set_id(prb)
                    msg.add_backref(prbid)
                    if len(field[1]):
                        field[1][0][:] = [prb]
                    else:
                        field[1] += nodes.paragraph('', '', prb)
                break
        if pep is None:
            raise DataError('Document does not contain an RFC-2822 "PEP" '
                            'header.')
        if pep == 0:
            # Special processing for PEP 0.
            pending = nodes.pending(PEPZero)
            self.document.insert(1, pending)
            self.document.note_pending(pending)
        if len(header) < 2 or header[1][0].astext().lower() != 'title':
            raise DataError('No title!')
        for field in header:
            name = field[0].astext().lower()
            body = field[1]
            if len(body) > 1:
                raise DataError('PEP header field body contains multiple '
                                'elements:\n%s' % field.pformat(level=1))
            elif len(body) == 1:
                if not isinstance(body[0], nodes.paragraph):
                    raise DataError('PEP header field body may only contain '
                                    'a single paragraph:\n%s'
                                    % field.pformat(level=1))
            elif name == 'last-modified':
                try:
                    date = time.strftime(
                        '%d-%b-%Y',
                        time.localtime(os.stat(self.document['source'])[8]))
                except OSError:
                    date = 'unknown'
                if cvs_url:
                    body += nodes.paragraph(
                        '', '', nodes.reference('', date, refuri=cvs_url))
            else:
                # empty
                continue
            para = body[0]
            if name == 'author':
                for node in para:
                    if isinstance(node, nodes.reference):
                        node.replace_self(mask_email(node))
            elif name == 'discussions-to':
                for node in para:
                    if isinstance(node, nodes.reference):
                        node.replace_self(mask_email(node, pep))
            elif name in ('replaces', 'replaced-by', 'requires'):
                newbody = []
                space = nodes.Text(' ')
                for refpep in re.split(r',?\s+', body.astext()):
                    pepno = int(refpep)
                    newbody.append(nodes.reference(
                        refpep, refpep,
                        refuri=(self.document.settings.pep_base_url
                                + self.pep_url % pepno)))
                    newbody.append(space)
                para[:] = newbody[:-1]  # drop trailing space
            elif name == 'last-modified':
                utils.clean_rcs_keywords(para, self.rcs_keyword_substitutions)
                if cvs_url:
                    date = para.astext()
                    para[:] = [nodes.reference('', date, refuri=cvs_url)]
            elif name == 'content-type':
                pep_type = para.astext()
                uri = self.document.settings.pep_base_url + self.pep_url % 12
                para[:] = [nodes.reference('', pep_type, refuri=uri)]
            elif name == 'version' and len(body):
                utils.clean_rcs_keywords(para, self.rcs_keyword_substitutions)


class Contents(Transform):

    """
    Insert an empty table of contents topic and a transform placeholder into
    the document after the RFC 2822 header.
    """

    default_priority = 380

    def apply(self):
        language = languages.get_language(self.document.settings.language_code,
                                          self.document.reporter)
        name = language.labels['contents']
        title = nodes.title('', name)
        topic = nodes.topic('', title, classes=['contents'])
        name = nodes.fully_normalize_name(name)
        if not self.document.has_name(name):
            topic['names'].append(name)
        self.document.note_implicit_target(topic)
        pending = nodes.pending(parts.Contents)
        topic += pending
        self.document.insert(1, topic)
        self.document.note_pending(pending)


class TargetNotes(Transform):

    """
    Locate the "References" section, insert a placeholder for an external
    target footnote insertion transform at the end, and schedule the
    transform to run immediately.
    """

    default_priority = 520

    def apply(self):
        doc = self.document
        i = len(doc) - 1
        refsect = copyright = None
        while i >= 0 and isinstance(doc[i], nodes.section):
            title_words = doc[i][0].astext().lower().split()
            if 'references' in title_words:
                refsect = doc[i]
                break
            elif 'copyright' in title_words:
                copyright = i
            i -= 1
        if not refsect:
            refsect = nodes.section()
            refsect += nodes.title('', 'References')
            doc.set_id(refsect)
            if copyright:
                # Put the new "References" section before "Copyright":
                doc.insert(copyright, refsect)
            else:
                # Put the new "References" section at end of doc:
                doc.append(refsect)
        pending = nodes.pending(references.TargetNotes)
        refsect.append(pending)
        self.document.note_pending(pending, 0)
        pending = nodes.pending(misc.CallBack,
                                details={'callback': self.cleanup_callback})
        refsect.append(pending)
        self.document.note_pending(pending, 1)

    def cleanup_callback(self, pending):
        """
        Remove an empty "References" section.

        Called after the `references.TargetNotes` transform is complete.
        """
        if len(pending.parent) == 2:    # <title> and <pending>
            pending.parent.parent.remove(pending.parent)


class PEPZero(Transform):

    """
    Special processing for PEP 0.
    """

    default_priority = 760

    def apply(self):
        visitor = PEPZeroSpecial(self.document)
        self.document.walk(visitor)
        self.startnode.parent.remove(self.startnode)


class PEPZeroSpecial(nodes.SparseNodeVisitor):

    """
    Perform the special processing needed by PEP 0:

    - Mask email addresses.

    - Link PEP numbers in the second column of 4-column tables to the PEPs
      themselves.
    """

    pep_url = Headers.pep_url

    def unknown_visit(self, node):
        pass

    def visit_reference(self, node):
        node.replace_self(mask_email(node))

    def visit_field_list(self, node):
        if 'rfc2822' in node['classes']:
            raise nodes.SkipNode

    def visit_tgroup(self, node):
        self.pep_table = node['cols'] == 4
        self.entry = 0

    def visit_colspec(self, node):
        self.entry += 1
        if self.pep_table and self.entry == 2:
            node['classes'].append('num')

    def visit_row(self, node):
        self.entry = 0

    def visit_entry(self, node):
        self.entry += 1
        if self.pep_table and self.entry == 2 and len(node) == 1:
            node['classes'].append('num')
            p = node[0]
            if isinstance(p, nodes.paragraph) and len(p) == 1:
                text = p.astext()
                try:
                    pep = int(text)
                    ref = (self.document.settings.pep_base_url
                           + self.pep_url % pep)
                    p[0] = nodes.reference(text, text, refuri=ref)
                except ValueError:
                    pass


non_masked_addresses = ('peps@python.org',
                        'python-list@python.org',
                        'python-dev@python.org')


def mask_email(ref, pepno=None):
    """
    Mask the email address in `ref` and return a replacement node.

    `ref` is returned unchanged if it contains no email address.

    For email addresses such as "user@host", mask the address as "user at
    host" (text) to thwart simple email address harvesters (except for those
    listed in `non_masked_addresses`).  If a PEP number (`pepno`) is given,
    return a reference including a default email subject.
    """
    if ref.hasattr('refuri') and ref['refuri'].startswith('mailto:'):
        if ref['refuri'][8:] in non_masked_addresses:
            replacement = ref[0]
        else:
            replacement_text = ref.astext().replace('@', '&#32;&#97;t&#32;')
            replacement = nodes.raw('', replacement_text, format='html')
        if pepno is None:
            return replacement
        else:
            ref['refuri'] += '?subject=PEP%%20%s' % pepno
            ref[:] = [replacement]
            return ref
    else:
        return ref