summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorRobin H. Johnson <robbat2@gentoo.org>2015-08-08 13:49:04 -0700
committerRobin H. Johnson <robbat2@gentoo.org>2015-08-08 17:38:18 -0700
commit56bd759df1d0c750a065b8c845e93d5dfa6b549d (patch)
tree3f91093cdb475e565ae857f1c5a7fd339e2d781e /dev-python/python-levenshtein/files
downloadgentoo-56bd759df1d0c750a065b8c845e93d5dfa6b549d.tar.gz
gentoo-56bd759df1d0c750a065b8c845e93d5dfa6b549d.tar.bz2
gentoo-56bd759df1d0c750a065b8c845e93d5dfa6b549d.zip
proj/gentoo: Initial commit
This commit represents a new era for Gentoo: Storing the gentoo-x86 tree in Git, as converted from CVS. This commit is the start of the NEW history. Any historical data is intended to be grafted onto this point. Creation process: 1. Take final CVS checkout snapshot 2. Remove ALL ChangeLog* files 3. Transform all Manifests to thin 4. Remove empty Manifests 5. Convert all stale $Header$/$Id$ CVS keywords to non-expanded Git $Id$ 5.1. Do not touch files with -kb/-ko keyword flags. Signed-off-by: Robin H. Johnson <robbat2@gentoo.org> X-Thanks: Alec Warner <antarus@gentoo.org> - did the GSoC 2006 migration tests X-Thanks: Robin H. Johnson <robbat2@gentoo.org> - infra guy, herding this project X-Thanks: Nguyen Thai Ngoc Duy <pclouds@gentoo.org> - Former Gentoo developer, wrote Git features for the migration X-Thanks: Brian Harring <ferringb@gentoo.org> - wrote much python to improve cvs2svn X-Thanks: Rich Freeman <rich0@gentoo.org> - validation scripts X-Thanks: Patrick Lauer <patrick@gentoo.org> - Gentoo dev, running new 2014 work in migration X-Thanks: Michał Górny <mgorny@gentoo.org> - scripts, QA, nagging X-Thanks: All of other Gentoo developers - many ideas and lots of paint on the bikeshed
Diffstat (limited to 'dev-python/python-levenshtein/files')
-rw-r--r--dev-python/python-levenshtein/files/genextdoc.py214
1 files changed, 214 insertions, 0 deletions
diff --git a/dev-python/python-levenshtein/files/genextdoc.py b/dev-python/python-levenshtein/files/genextdoc.py
new file mode 100644
index 000000000000..8692297236d8
--- /dev/null
+++ b/dev-python/python-levenshtein/files/genextdoc.py
@@ -0,0 +1,214 @@
+#!/usr/bin/python
+# Simple Python extension module doc generator.
+# @(#) $Id$
+# Written by Yeti <yeti@physics.muni.cz>
+# This program is in the public domain.
+import re, sys, types, inspect
+from cgi import escape as q
+
+args = sys.argv
+args.pop(0)
+if not args: sys.exit(0)
+
+selfcontained = False
+if args[0].startswith('-') and 'selfcontained'.startswith(args[0].strip('-')):
+ selfcontained = True
+ args.pop(0)
+if not args: sys.exit(0)
+modname = args.pop(0)
+plain_docs = args
+
+html_head = """\
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN"
+ "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
+<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en">
+<head><title>%s %s</title></head>
+<body>
+"""
+
+html_foot = """\
+</body>
+</html>
+"""
+
+def split_para(doc):
+ p = []
+ for x in doc.split('\n\n'):
+ x = x.strip()
+ if x.find('\n>>>') > -1:
+ h, t = x.split('\n>>>', 1)
+ p.append(h)
+ p.append('>>>' + t)
+ else:
+ p.append(x)
+ return p
+
+def get_doc(members):
+ try: doc = members['__doc__']
+ except KeyError: pass
+ if doc: return doc
+ try: doc = 'Python module %s' % members['__name__']
+ except KeyError: pass
+ if doc: return doc
+ else: return 'A Python module'
+
+def format_synopsis(synopsis, link=False, classname=None):
+ lst = synopsis.split('\n')
+ for i, s in zip(range(len(lst)), lst):
+ m = re.match(r'(?P<func>\w+)(?P<args>.*)', s)
+ args = re.sub(r'([a-zA-Z]\w+)', r'<var>\1</var>', m.group('args'))
+ func = m.group('func')
+ if link:
+ if classname:
+ func = '<a href="#%s-%s">%s</a>' % (classname, func, func)
+ else:
+ func = '<a href="#%s">%s</a>' % (func, func)
+ lst[i] = func + args
+ return '<br/>\n'.join(lst)
+
+def format_para(p):
+ if not p: return ''
+ doc = ''
+ if p.startswith('>>>'): doc += '<pre>\n%s\n</pre>\n' % q(p)
+ else:
+ if not re.search('^- ', p, re.M): doc += '<p>%s</p>\n' % q(p)
+ else:
+ p = re.split('(?m)^- ', p)
+ if p[0]: doc += '<p>%s</p>\n' % q(p[0].strip())
+ del p[0]
+ doc += ('<ul>%s</ul>\n'
+ % '\n'.join(['<li>%s</li>' % q(p.strip()) for p in p]))
+ return doc
+
+def preprocess_routine(name, doc):
+ parts = split_para(doc)
+ if parts: summary = parts.pop(0)
+ else: summary = 'FIXME'
+ if parts and re.match(r'\w+\(.*\)', parts[0]): synopsis = parts.pop(0)
+ else: synopsis = name + '()'
+ return {'synopsis': synopsis, 'summary': summary, 'details': parts}
+
+def analyse(obj):
+ members = obj.__dict__
+ if inspect.isclass(obj):
+ main_doc = preprocess_routine(obj.__name__, get_doc(members))
+ bases = [x.__name__ for x in obj.__bases__]
+ else:
+ main_doc = split_para(get_doc(members))
+ bases = []
+ routines = {}
+ classes = {}
+ data = {}
+ for name, m in members.items():
+ if name.startswith('__'): continue
+ try:
+ mro = list(inspect.getmro(m))
+ if mro[0] != m: continue
+ except AttributeError: pass
+ if inspect.isroutine(m):
+ try: doc = m.__doc__
+ except KeyError: pass
+ if not doc: doc = 'FIXME'
+ routines[name] = preprocess_routine(name, doc)
+ continue
+ if inspect.isclass(m):
+ classes[name] = analyse(m)
+ continue
+ t = type(m)
+ if t == types.IntType or t == types.StringType:
+ data[name] = repr(m)
+ else:
+ data[name] = m.__doc__
+ return {'name': obj.__name__, 'doc': main_doc, 'routines': routines,
+ 'classes': classes, 'data': data, 'bases': bases}
+
+def format(tree, level, prefix=''):
+ name = tree['name']
+ if prefix: fullname = '%s-%s' % (prefix, name)
+ else: fullname = name
+ ##### Main doc
+ doc = []
+ if level > 1:
+ doc = ['<h%d id="%s">' % (level, fullname)]
+ try: doc.append(format_synopsis(tree['doc']['synopsis']))
+ except TypeError:
+ doc.append(name)
+ doc.append('</h%d>\n' % level)
+ if tree.has_key('bases'):
+ doc.append('<p>Bases: %s.</p>\n' % ', '.join(tree['bases']))
+ try: lst = [tree['doc']['summary']] + tree['doc']['details']
+ except TypeError: lst = tree['doc']
+ for p in lst: doc.append(format_para(p))
+ ##### Table of contents
+ routines = tree['routines'].keys()
+ classes = tree['classes'].keys()
+ data = tree['data'].keys()
+ if routines:
+ routines.sort()
+ if level == 1: doc.append('<p><b>Functions:</b></p>\n')
+ else: doc.append('<p><b>Methods:</b></p>\n')
+ doc.append('<ul class="ltoc">\n')
+ for r in routines:
+ synopsis = tree['routines'][r]['synopsis']
+ doc.append('<li>%s</li>\n' % format_synopsis(synopsis, True,
+ fullname))
+ doc.append('</ul>\n')
+ if classes:
+ classes.sort()
+ doc.append('<p><b>Classes:</b></p>\n')
+ doc.append('<ul class="ltoc">\n')
+ for r in classes:
+ synopsis = tree['classes'][r]['doc']['synopsis']
+ doc.append('<li>%s</li>\n' % format_synopsis(synopsis, True,
+ fullname))
+ doc.append('</ul>\n')
+ if data:
+ data.sort()
+ doc.append('<p><b>Data:</b></p>\n')
+ doc.append('<ul class="ltoc">\n')
+ for r in data:
+ doc.append('<li>%s = %s</li>\n' % (r, q(tree['data'][r])))
+ doc.append('</ul>\n')
+ ##### Functions
+ if routines:
+ if level == 1: doc.append('<hr/>\n')
+ doc.append('<dl>\n')
+ for r in routines:
+ doc.append('<dt id="%s-%s">' % (fullname, r))
+ rt = tree['routines'][r]
+ doc.append('%s</dt>\n<dd>' % format_synopsis(rt['synopsis']))
+ for p in [rt['summary']] + rt['details']:
+ doc.append(format_para(p))
+ doc.append('</dd>\n')
+ doc.append('</dl>\n')
+ ##### Classes
+ if classes:
+ for r in classes:
+ doc.append('<hr/>\n')
+ doc.append(format(tree['classes'][r], level+1, fullname))
+ return ''.join(doc)
+
+exec 'import %s as __test__' % modname
+doctree = analyse(__test__)
+document = format(doctree, 1)
+print modname + '.html'
+fh = file(modname + '.html', 'w')
+if selfcontained: fh.write(html_head % (modname, 'module API'))
+fh.write(document)
+if selfcontained: fh.write(html_foot)
+fh.close()
+for f in plain_docs:
+ try: fh = file(f, 'r')
+ except: continue
+ document = fh.read()
+ fh.close()
+ print f + '.xhtml'
+ fh = file(f + '.xhtml', 'w')
+ if selfcontained: fh.write(html_head % (modname, f))
+ fh.write('<h1>%s %s</h1>\n\n' % (modname, f))
+ fh.write('<pre class="main">\n')
+ fh.write(document)
+ fh.write('</pre>\n')
+ if selfcontained: fh.write(html_foot)
+ fh.close()