python-peps/docutils/transforms/frontmatter.py

381 lines
14 KiB
Python

# Authors: David Goodger, Ueli Schlaepfer
# Contact: goodger@users.sourceforge.net
# Revision: $Revision$
# Date: $Date$
# Copyright: This module has been placed in the public domain.
"""
Transforms related to the front matter of a document (information
found before the main text):
- `DocTitle`: Used to transform a lone top level section's title to
the document title, and promote a remaining lone top-level section's
title to the document subtitle.
- `DocInfo`: Used to transform a bibliographic field list into docinfo
elements.
"""
__docformat__ = 'reStructuredText'
import re
from docutils import nodes, utils
from docutils.transforms import TransformError, Transform
class DocTitle(Transform):
"""
In reStructuredText_, there is no way to specify a document title
and subtitle explicitly. Instead, we can supply the document title
(and possibly the subtitle as well) implicitly, and use this
two-step transform to "raise" or "promote" the title(s) (and their
corresponding section contents) to the document level.
1. If the document contains a single top-level section as its
first non-comment element, the top-level section's title
becomes the document's title, and the top-level section's
contents become the document's immediate contents. The lone
top-level section header must be the first non-comment element
in the document.
For example, take this input text::
=================
Top-Level Title
=================
A paragraph.
Once parsed, it looks like this::
<document>
<section name="top-level title">
<title>
Top-Level Title
<paragraph>
A paragraph.
After running the DocTitle transform, we have::
<document name="top-level title">
<title>
Top-Level Title
<paragraph>
A paragraph.
2. If step 1 successfully determines the document title, we
continue by checking for a subtitle.
If the lone top-level section itself contains a single
second-level section as its first non-comment element, that
section's title is promoted to the document's subtitle, and
that section's contents become the document's immediate
contents. Given this input text::
=================
Top-Level Title
=================
Second-Level Title
~~~~~~~~~~~~~~~~~~
A paragraph.
After parsing and running the Section Promotion transform, the
result is::
<document name="top-level title">
<title>
Top-Level Title
<subtitle name="second-level title">
Second-Level Title
<paragraph>
A paragraph.
(Note that the implicit hyperlink target generated by the
"Second-Level Title" is preserved on the "subtitle" element
itself.)
Any comment elements occurring before the document title or
subtitle are accumulated and inserted as the first body elements
after the title(s).
"""
default_priority = 320
def apply(self):
if self.promote_document_title():
self.promote_document_subtitle()
def promote_document_title(self):
section, index = self.candidate_index()
if index is None:
return None
document = self.document
# Transfer the section's attributes to the document element (at root):
document.attributes.update(section.attributes)
document[:] = (section[:1] # section title
+ document[:index] # everything that was in the
# document before the section
+ section[1:]) # everything that was in the section
return 1
def promote_document_subtitle(self):
subsection, index = self.candidate_index()
if index is None:
return None
subtitle = nodes.subtitle()
# Transfer the subsection's attributes to the new subtitle:
subtitle.attributes.update(subsection.attributes)
# Transfer the contents of the subsection's title to the subtitle:
subtitle[:] = subsection[0][:]
document = self.document
document[:] = (document[:1] # document title
+ [subtitle]
# everything that was before the section:
+ document[1:index]
# everything that was in the subsection:
+ subsection[1:])
return 1
def candidate_index(self):
"""
Find and return the promotion candidate and its index.
Return (None, None) if no valid candidate was found.
"""
document = self.document
index = document.first_child_not_matching_class(
nodes.PreBibliographic)
if index is None or len(document) > (index + 1) or \
not isinstance(document[index], nodes.section):
return None, None
else:
return document[index], index
class DocInfo(Transform):
"""
This transform is specific to the reStructuredText_ markup syntax;
see "Bibliographic Fields" in the `reStructuredText Markup
Specification`_ for a high-level description. This transform
should be run *after* the `DocTitle` transform.
Given a field list as the first non-comment element after the
document title and subtitle (if present), registered bibliographic
field names are transformed to the corresponding DTD elements,
becoming child elements of the "docinfo" element (except for a
dedication and/or an abstract, which become "topic" elements after
"docinfo").
For example, given this document fragment after parsing::
<document>
<title>
Document Title
<field_list>
<field>
<field_name>
Author
<field_body>
<paragraph>
A. Name
<field>
<field_name>
Status
<field_body>
<paragraph>
$RCSfile$
...
After running the bibliographic field list transform, the
resulting document tree would look like this::
<document>
<title>
Document Title
<docinfo>
<author>
A. Name
<status>
frontmatter.py
...
The "Status" field contained an expanded RCS keyword, which is
normally (but optionally) cleaned up by the transform. The sole
contents of the field body must be a paragraph containing an
expanded RCS keyword of the form "$keyword: expansion text $". Any
RCS keyword can be processed in any bibliographic field. The
dollar signs and leading RCS keyword name are removed. Extra
processing is done for the following RCS keywords:
- "RCSfile" expands to the name of the file in the RCS or CVS
repository, which is the name of the source file with a ",v"
suffix appended. The transform will remove the ",v" suffix.
- "Date" expands to the format "YYYY/MM/DD hh:mm:ss" (in the UTC
time zone). The RCS Keywords transform will extract just the
date itself and transform it to an ISO 8601 format date, as in
"2000-12-31".
(Since the source file for this text is itself stored under CVS,
we can't show an example of the "Date" RCS keyword because we
can't prevent any RCS keywords used in this explanation from
being expanded. Only the "RCSfile" keyword is stable; its
expansion text changes only if the file name changes.)
"""
default_priority = 340
def apply(self):
document = self.document
index = document.first_child_not_matching_class(
nodes.PreBibliographic)
if index is None:
return
candidate = document[index]
if isinstance(candidate, nodes.field_list):
biblioindex = document.first_child_not_matching_class(
nodes.Titular)
nodelist = self.extract_bibliographic(candidate)
del document[index] # untransformed field list (candidate)
document[biblioindex:biblioindex] = nodelist
return
def extract_bibliographic(self, field_list):
docinfo = nodes.docinfo()
bibliofields = self.language.bibliographic_fields
labels = self.language.labels
topics = {'dedication': None, 'abstract': None}
for field in field_list:
try:
name = field[0][0].astext()
normedname = utils.normalize_name(name)
if not (len(field) == 2 and bibliofields.has_key(normedname)
and self.check_empty_biblio_field(field, name)):
raise TransformError
biblioclass = bibliofields[normedname]
if issubclass(biblioclass, nodes.TextElement):
if not self.check_compound_biblio_field(field, name):
raise TransformError
utils.clean_rcs_keywords(
field[1][0], self.rcs_keyword_substitutions)
docinfo.append(biblioclass('', '', *field[1][0]))
else: # multiple body elements possible
if issubclass(biblioclass, nodes.authors):
self.extract_authors(field, name, docinfo)
elif issubclass(biblioclass, nodes.topic):
if topics[normedname]:
field[-1] += self.document.reporter.warning(
'There can only be one "%s" field.' % name,
base_node=field)
raise TransformError
title = nodes.title(name, labels[normedname])
topics[normedname] = biblioclass(
'', title, CLASS=normedname, *field[1].children)
else:
docinfo.append(biblioclass('', *field[1].children))
except TransformError:
if len(field[-1]) == 1 \
and isinstance(field[-1][0], nodes.paragraph):
utils.clean_rcs_keywords(
field[-1][0], self.rcs_keyword_substitutions)
docinfo.append(field)
continue
nodelist = []
if len(docinfo) != 0:
nodelist.append(docinfo)
for name in ('dedication', 'abstract'):
if topics[name]:
nodelist.append(topics[name])
return nodelist
def check_empty_biblio_field(self, field, name):
if len(field[-1]) < 1:
field[-1] += self.document.reporter.warning(
'Cannot extract empty bibliographic field "%s".' % name,
base_node=field)
return None
return 1
def check_compound_biblio_field(self, field, name):
if len(field[-1]) > 1:
field[-1] += self.document.reporter.warning(
'Cannot extract compound bibliographic field "%s".' % name,
base_node=field)
return None
if not isinstance(field[-1][0], nodes.paragraph):
field[-1] += self.document.reporter.warning(
'Cannot extract bibliographic field "%s" containing '
'anything other than a single paragraph.' % name,
base_node=field)
return None
return 1
rcs_keyword_substitutions = [
(re.compile(r'\$' r'Date: (\d\d\d\d)/(\d\d)/(\d\d) [\d:]+ \$$',
re.IGNORECASE), r'\1-\2-\3'),
(re.compile(r'\$' r'RCSfile: (.+),v \$$', re.IGNORECASE), r'\1'),
(re.compile(r'\$[a-zA-Z]+: (.+) \$$'), r'\1'),]
def extract_authors(self, field, name, docinfo):
try:
if len(field[1]) == 1:
if isinstance(field[1][0], nodes.paragraph):
authors = self.authors_from_one_paragraph(field)
elif isinstance(field[1][0], nodes.bullet_list):
authors = self.authors_from_bullet_list(field)
else:
raise TransformError
else:
authors = self.authors_from_paragraphs(field)
authornodes = [nodes.author('', '', *author)
for author in authors if author]
if len(authornodes) > 1:
docinfo.append(nodes.authors('', *authornodes))
elif len(authornodes) == 1:
docinfo.append(authornodes[0])
else:
raise TransformError
except TransformError:
field[-1] += self.document.reporter.warning(
'Bibliographic field "%s" incompatible with extraction: '
'it must contain either a single paragraph (with authors '
'separated by one of "%s"), multiple paragraphs (one per '
'author), or a bullet list with one paragraph (one author) '
'per item.'
% (name, ''.join(self.language.author_separators)),
base_node=field)
raise
def authors_from_one_paragraph(self, field):
text = field[1][0].astext().strip()
if not text:
raise TransformError
for authorsep in self.language.author_separators:
authornames = text.split(authorsep)
if len(authornames) > 1:
break
authornames = [author.strip() for author in authornames]
authors = [[nodes.Text(author)] for author in authornames if author]
return authors
def authors_from_bullet_list(self, field):
authors = []
for item in field[1][0]:
if len(item) != 1 or not isinstance(item[0], nodes.paragraph):
raise TransformError
authors.append(item[0].children)
if not authors:
raise TransformError
return authors
def authors_from_paragraphs(self, field):
for item in field[1]:
if not isinstance(item, nodes.paragraph):
raise TransformError
authors = [item.children for item in field[1]]
return authors