Mercurial > hg
view hgext/highlight/highlight.py @ 23610:9266d1dd6a6e
namespaces: generate template keyword when registering a namespace
For any namespace, we generate a template keyword. For example, given a
namespace 'babar', we automatically have the ability to use it in a template:
hg log -r . -T '{babars % "King: {babar}\n"}'
Furthermore, we only generate this keyword for a namespace if one doesn't
already exist. This is necessary for 'branches' and 'bookmarks' since both of
those have concepts of 'current' (something outside the namespace api) and also
allows extensions to override default behavior if desired.
author | Sean Farley <sean.michael.farley@gmail.com> |
---|---|
date | Thu, 16 Oct 2014 23:19:09 -0700 |
parents | bcdfb6078b9f |
children | 7b8ff3fd11d3 |
line wrap: on
line source
# highlight.py - highlight extension implementation file # # Copyright 2007-2009 Adam Hupp <adam@hupp.org> and others # # This software may be used and distributed according to the terms of the # GNU General Public License version 2 or any later version. # # The original module was split in an interface and an implementation # file to defer pygments loading and speedup extension setup. from mercurial import demandimport demandimport.ignore.extend(['pkgutil', 'pkg_resources', '__main__']) from mercurial import util, encoding from pygments import highlight from pygments.util import ClassNotFound from pygments.lexers import guess_lexer, guess_lexer_for_filename, TextLexer from pygments.formatters import HtmlFormatter SYNTAX_CSS = ('\n<link rel="stylesheet" href="{url}highlightcss" ' 'type="text/css" />') def pygmentize(field, fctx, style, tmpl): # append a <link ...> to the syntax highlighting css old_header = tmpl.load('header') if SYNTAX_CSS not in old_header: new_header = old_header + SYNTAX_CSS tmpl.cache['header'] = new_header text = fctx.data() if util.binary(text): return # Pygments is best used with Unicode strings: # <http://pygments.org/docs/unicode/> text = text.decode(encoding.encoding, 'replace') # To get multi-line strings right, we can't format line-by-line try: lexer = guess_lexer_for_filename(fctx.path(), text[:1024], stripnl=False) except (ClassNotFound, ValueError): try: lexer = guess_lexer(text[:1024], stripnl=False) except (ClassNotFound, ValueError): lexer = TextLexer(stripnl=False) formatter = HtmlFormatter(style=style) colorized = highlight(text, lexer, formatter) # strip wrapping div colorized = colorized[:colorized.find('\n</pre>')] colorized = colorized[colorized.find('<pre>') + 5:] coloriter = (s.encode(encoding.encoding, 'replace') for s in colorized.splitlines()) tmpl.filters['colorize'] = lambda x: coloriter.next() oldl = tmpl.cache[field] newl = oldl.replace('line|escape', 'line|colorize') tmpl.cache[field] = newl