Files
@ fa4e94ad3dd6
Branch filter:
Location: kallithea/scripts/docs-headings.py - annotation
fa4e94ad3dd6
2.6 KiB
text/x-python
templates: use span elements instead of left floating div
<span> have a more natural (inline) flow, i.e. they already have proper spacing
between the elements without additional styling.
On files browser page, since the form element does not flow inline, it needs to
include every element that should be in the first line.
<span> have a more natural (inline) flow, i.e. they already have proper spacing
between the elements without additional styling.
On files browser page, since the form element does not flow inline, it needs to
include every element that should be in the first line.
f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 ed2fb6e84a02 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 f38b50f8a6a6 | #!/usr/bin/env python2
"""
Consistent formatting of rst section titles
"""
import re
import subprocess
spaces = [
(0, 1), # we assume this is a over-and-underlined header
(2, 1),
(1, 1),
(1, 0),
(1, 0),
]
# http://sphinx-doc.org/rest.html :
# for the Python documentation, this convention is used which you may follow:
# # with overline, for parts
# * with overline, for chapters
# =, for sections
# -, for subsections
# ^, for subsubsections
# ", for paragraphs
pystyles = ['#', '*', '=', '-', '^', '"']
# match on a header line underlined with one of the valid characters
headermatch = re.compile(r'''\n*(.+)\n([][!"#$%&'()*+,./:;<=>?@\\^_`{|}~-])\2{2,}\n+''', flags=re.MULTILINE)
def main():
for fn in subprocess.check_output(['hg', 'loc', 'set:**.rst+kallithea/i18n/how_to']).splitlines():
print 'processing %s:' % fn
s = file(fn).read()
# find levels and their styles
lastpos = 0
styles = []
for markup in headermatch.findall(s):
style = markup[1]
if style in styles:
stylepos = styles.index(style)
if stylepos > lastpos + 1:
print 'bad style %r with level %s - was at %s' % (style, stylepos, lastpos)
else:
stylepos = len(styles)
if stylepos > lastpos + 1:
print 'bad new style %r - expected %r' % (style, styles[lastpos + 1])
else:
styles.append(style)
lastpos = stylepos
# remove superfluous spacing (may however be restored by header spacing)
s = re.sub(r'''(\n\n)\n*''', r'\1', s, flags=re.MULTILINE)
if styles:
newstyles = pystyles[pystyles.index(styles[0]):]
def subf(m):
title, style = m.groups()
level = styles.index(style)
before, after = spaces[level]
newstyle = newstyles[level]
return '\n' * (before + 1) + title + '\n' + newstyle * len(title) + '\n' * (after + 1)
s = headermatch.sub(subf, s)
# remove superfluous spacing when headers are adjacent
s = re.sub(r'''(\n.+\n([][!"#$%&'()*+,./:;<=>?@\\^_`{|}~-])\2{2,}\n\n\n)\n*''', r'\1', s, flags=re.MULTILINE)
# fix trailing space and spacing before link sections
s = s.strip() + '\n'
s = re.sub(r'''\n+((?:\.\. _[^\n]*\n)+)$''', r'\n\n\n\1', s)
file(fn, 'w').write(s)
print subprocess.check_output(['hg', 'diff', fn])
print
if __name__ == '__main__':
main()
|