Simplify the get_value routines a bit.

git-svn-id: svn://svn.lyx.org/lyx/lyx-devel/trunk@36103 a592a061-630c-0410-9148-cb99ea01b6c8
This commit is contained in:
Richard Heck 2010-11-05 14:26:14 +00:00
parent 96106e6dda
commit cd99f162b3

View File

@ -147,7 +147,8 @@ def get_value(lines, token, start, end = 0, default = ""):
Find the next line that looks like:
token followed by other stuff
Returns "followed by other stuff".
Returns "followed by other stuff" with leading and trailing
whitespace removed.
"""
i = find_token_exact(lines, token, start, end)
@ -155,7 +156,7 @@ def get_value(lines, token, start, end = 0, default = ""):
return default
l = lines[i].split(None, 1)
if len(l) > 1:
return l[1]
return l[1].strip()
return default
@ -166,18 +167,12 @@ def get_value_string(lines, token, start, end = 0, trim = False, default = ""):
token is the first element. When trim is used, the first and last character
of the string is trimmed."""
i = find_token_exact(lines, token, start, end)
if i == -1:
return default
if len(lines[i].split()) > 1:
for k in range (0, len(lines[i])):
if lines[i][k] == ' ':
if trim == False:
return lines[i][k+1:len(lines[i])]
else:
return lines[i][k+2:len(lines[i])-1]
else:
return default
val = get_value(lines, token, start, end, "")
if not val:
return default
if trim:
return val[1:-1]
return val
def del_token(lines, token, start, end):