Formats text following the MediaWiki syntax.
To return HTML from Wiki:
from wikimarkup.parser import Parser parser = Parser() html = parser.parse(text[, show_toc=True])
To return HTML without certain "annoying" (TODO: define annoying) elements, such as headings:
from wikimarkup.parser import parselite parselite(text)
You can add new tags with the registerTagHook method.:
from wikimarkup.parser import Parser import cgi def blockquoteTagHook(parser_env, body, attributes={}): """<quote[ cite="Person"]>A paragraph of text.</quote>""" text = ['<blockquote>'] if 'cite' in attributes: text.append('<cite>%s</cite>' % (cgi.escape(attributes['cite']),)) text.append(parse(body.strip())) text.append('</blockquote>') return u'\n'.join(text) parser = Parser() parser.registerTagHook('quote', blockquoteTagHook)
You can support [[internal links]]
with the registerInternalLinkHook
method. There is no default handling for internal links. If no hook
handles the link, it will appear unchanged in the output. An internal
link may have a namespace: prefix. Hooks are registered per namespace,
with 'None' for unprefixed links:
def internalLinkHook(parser_env, namespace, body): ... return replacement parser.registerInternalLinkHook(None, internalLinkHook) # called for [[link]] parser.registerInternalLinkHook('Wikipedia', hook) # called for [[Wikipedia: Link]] parser.registerInternalLinkHook(':en', hook) # called for [[:en:link] parser.registerInternalLinkHook(':', hook) # called for [[:any:link]] parser.registerInternalLinkHook('*', hook) # called for [[anything]]
Examples:
from wikimarkup.parser import Parser def wikipediaLinkHook(parser_env, namespace, body): # namespace is going to be 'Wikipedia' (article, pipe, text) = body.partition('|') href = article.strip().capitalize().replace(' ', '_') text = (text or article).strip() return '<a href="http://en.wikipedia.org/wiki/%s">%s</a>' % (href, text) parser = Parser() parser.registerInternalLinkHook('Wikipedia', wikipediaLinkHook) print parser.parse("[[Wikipedia:public transport|public transportation]]") print parser.parse("[[Wikipedia: bus]]") import settings from pytils.translit import slugify from blog.models import Post def byteflowLinkHook(parser_env, namespace, body): (article, pipe, text) = body.partition('|') slug = slugify(article.strip()) text = (text or article).strip() try: post = Post.objects.get(slug=slug) href = post.get_absolute_url() except Post.DoesNotExist: href = '#' return '<a href="%s">%s</a>' % (href, text) parser.registerInternalLinkHook(None, byteflowLinkHook) parser.parse("[[Blog post title]]")