mirror of
https://git.lyx.org/repos/lyx.git
synced 2024-11-22 01:59:02 +00:00
Remove trailing whitespace in lyx2lyx.
This commit is contained in:
parent
39c22913a2
commit
6fe2e45f97
@ -1256,7 +1256,7 @@ def convert_colorbox(document):
|
||||
|
||||
def revert_colorbox(document):
|
||||
" outputs color settings for boxes as TeX code "
|
||||
|
||||
|
||||
i = 0
|
||||
defaultframecolor = "black"
|
||||
defaultbackcolor = "none"
|
||||
|
@ -1,6 +1,6 @@
|
||||
# This file is part of lyx2lyx
|
||||
# -*- coding: utf-8 -*-
|
||||
# Copyright (C) 2002-2011 Dekel Tsur <dekel@lyx.org>,
|
||||
# Copyright (C) 2002-2011 Dekel Tsur <dekel@lyx.org>,
|
||||
# José Matos <jamatos@lyx.org>, Richard Heck <rgheck@comcast.net>
|
||||
#
|
||||
# This program is free software; you can redistribute it and/or
|
||||
@ -18,17 +18,17 @@
|
||||
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
||||
|
||||
|
||||
'''
|
||||
'''
|
||||
This module offers several free functions to help parse lines.
|
||||
More documentaton is below, but here is a quick guide to what
|
||||
More documentaton is below, but here is a quick guide to what
|
||||
they do. Optional arguments are marked by brackets.
|
||||
|
||||
find_token(lines, token, start[, end[, ignorews]]):
|
||||
Returns the first line i, start <= i < end, on which
|
||||
token is found at the beginning. Returns -1 if not
|
||||
found.
|
||||
token is found at the beginning. Returns -1 if not
|
||||
found.
|
||||
If ignorews is (given and) True, then differences
|
||||
in whitespace do not count, except that there must be no
|
||||
in whitespace do not count, except that there must be no
|
||||
extra whitespace following token itself.
|
||||
|
||||
find_token_exact(lines, token, start[, end]):
|
||||
@ -36,15 +36,15 @@ find_token_exact(lines, token, start[, end]):
|
||||
|
||||
find_tokens(lines, tokens, start[, end[, ignorews]]):
|
||||
Returns the first line i, start <= i < end, on which
|
||||
one of the tokens in tokens is found at the beginning.
|
||||
Returns -1 if not found.
|
||||
one of the tokens in tokens is found at the beginning.
|
||||
Returns -1 if not found.
|
||||
If ignorews is (given and) True, then differences
|
||||
in whitespace do not count, except that there must be no
|
||||
in whitespace do not count, except that there must be no
|
||||
extra whitespace following token itself.
|
||||
|
||||
find_tokens_exact(lines, token, start[, end]):
|
||||
As find_tokens, but with ignorews True.
|
||||
|
||||
|
||||
find_token_backwards(lines, token, start):
|
||||
find_tokens_backwards(lines, tokens, start):
|
||||
As before, but look backwards.
|
||||
@ -54,16 +54,16 @@ find_re(lines, rexp, start[, end]):
|
||||
so it has to be passed as e.g.: re.compile(r'...').
|
||||
|
||||
get_value(lines, token, start[, end[, default]):
|
||||
Similar to find_token, but it returns what follows the
|
||||
Similar to find_token, but it returns what follows the
|
||||
token on the found line. Example:
|
||||
get_value(document.header, "\\use_xetex", 0)
|
||||
will find a line like:
|
||||
\\use_xetex true
|
||||
and, in that case, return "true". (Note that whitespace
|
||||
is stripped.) The final argument, default, defaults to "",
|
||||
is stripped.) The final argument, default, defaults to "",
|
||||
and is what is returned if we do not find anything. So you
|
||||
can use that to set a default.
|
||||
|
||||
|
||||
get_quoted_value(lines, token, start[, end[, default]]):
|
||||
Similar to get_value, but it will strip quotes off the
|
||||
value, if they are present. So use this one for cases
|
||||
@ -82,12 +82,12 @@ del_token(lines, token, start[, end]):
|
||||
Returns True if a line got deleted, otherwise False.
|
||||
|
||||
find_beginning_of(lines, i, start_token, end_token):
|
||||
Here, start_token and end_token are meant to be a matching
|
||||
pair, like "\\begin_layout" and "\\end_layout". We look for
|
||||
Here, start_token and end_token are meant to be a matching
|
||||
pair, like "\\begin_layout" and "\\end_layout". We look for
|
||||
the start_token that pairs with the end_token that occurs
|
||||
on or after line i. Returns -1 if not found.
|
||||
So, in the layout case, this would find the \\begin_layout
|
||||
for the layout line i is in.
|
||||
So, in the layout case, this would find the \\begin_layout
|
||||
for the layout line i is in.
|
||||
Example:
|
||||
ec = find_token(document.body, "</cell", i)
|
||||
bc = find_beginning_of(document.body, ec, \
|
||||
@ -95,7 +95,7 @@ find_beginning_of(lines, i, start_token, end_token):
|
||||
Now, assuming no -1s, bc-ec wraps the cell for line i.
|
||||
|
||||
find_end_of(lines, i, start_token, end_token):
|
||||
Like find_beginning_of, but looking for the matching
|
||||
Like find_beginning_of, but looking for the matching
|
||||
end_token. This might look like:
|
||||
bc = find_token_(document.body, "<cell", i)
|
||||
ec = find_end_of(document.body, bc, "<cell", "</cell")
|
||||
@ -115,7 +115,7 @@ find_end_of_sequence(lines, i):
|
||||
|
||||
is_in_inset(lines, i, inset):
|
||||
Checks if line i is in an inset of the given type.
|
||||
If so, returns starting and ending lines. Otherwise,
|
||||
If so, returns starting and ending lines. Otherwise,
|
||||
returns False.
|
||||
Example:
|
||||
is_in_inset(document.body, i, "\\begin_inset Tabular")
|
||||
@ -127,8 +127,8 @@ is_in_inset(lines, i, inset):
|
||||
will do what you expect.
|
||||
|
||||
get_containing_inset(lines, i):
|
||||
Finds out what kind of inset line i is within. Returns a
|
||||
list containing what follows \begin_inset on the line
|
||||
Finds out what kind of inset line i is within. Returns a
|
||||
list containing what follows \begin_inset on the line
|
||||
on which the inset begins, plus the starting and ending line.
|
||||
Returns False on any kind of error or if it isn't in an inset.
|
||||
So get_containing_inset(document.body, i) might return:
|
||||
@ -180,7 +180,7 @@ def find_token(lines, token, start, end = 0, ignorews = False):
|
||||
|
||||
Return the lowest line where token is found, and is the first
|
||||
element, in lines[start, end].
|
||||
|
||||
|
||||
If ignorews is True (default is False), then differences in
|
||||
whitespace are ignored, except that there must be no extra
|
||||
whitespace following token itself.
|
||||
@ -358,7 +358,7 @@ def set_option_value(line, option, value):
|
||||
def del_token(lines, token, start, end = 0):
|
||||
""" del_token(lines, token, start, end) -> int
|
||||
|
||||
Find the first line in lines where token is the first element
|
||||
Find the first line in lines where token is the first element
|
||||
and delete that line. Returns True if we deleted a line, False
|
||||
if we did not."""
|
||||
|
||||
@ -445,8 +445,8 @@ def is_in_inset(lines, i, inset):
|
||||
|
||||
|
||||
def get_containing_inset(lines, i):
|
||||
'''
|
||||
Finds out what kind of inset line i is within. Returns a
|
||||
'''
|
||||
Finds out what kind of inset line i is within. Returns a
|
||||
list containing (i) what follows \begin_inset on the line
|
||||
on which the inset begins, plus the starting and ending line.
|
||||
Returns False on any kind of error or if it isn't in an inset.
|
||||
@ -472,8 +472,8 @@ def get_containing_inset(lines, i):
|
||||
|
||||
|
||||
def get_containing_layout(lines, i):
|
||||
'''
|
||||
Finds out what kind of layout line i is within. Returns a
|
||||
'''
|
||||
Finds out what kind of layout line i is within. Returns a
|
||||
list containing what follows \begin_layout on the line
|
||||
on which the layout begins, plus the starting and ending line
|
||||
and the start of the paragraph (after all params). I.e, returns:
|
||||
@ -553,4 +553,3 @@ def find_end_of_sequence(lines, i):
|
||||
i = i + 1
|
||||
|
||||
return endlay
|
||||
|
||||
|
@ -89,5 +89,5 @@ class TestParserTools(unittest.TestCase):
|
||||
self.assertEqual(find_tokens(lines, tokens, 0, 4), -1)
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
if __name__ == '__main__':
|
||||
unittest.main()
|
||||
|
Loading…
Reference in New Issue
Block a user