488 lines
18 KiB
Python
Executable File
488 lines
18 KiB
Python
Executable File
#
|
|
# Indentation finder, by Philippe Fremy <phil at freehackers dot org>
|
|
# Copyright 2002-2008 Philippe Fremy
|
|
#
|
|
# This program is distributed under the BSD license. You should have received
|
|
# a copy of the file LICENSE.txt along with this software.
|
|
#
|
|
|
|
import sys
|
|
import re
|
|
import argparse
|
|
|
|
help = \
|
|
"""Usage : %s [ --vim-output ] [ --verbose ] file1 file2 ... fileN
|
|
|
|
Display indentation used in the list of files. Possible answers are (with X
|
|
being the number of spaces used for indentation):
|
|
space X
|
|
tab 8
|
|
mixed tab X space Y
|
|
|
|
mixed means that indentation style is tab at the beginning of the line (tab
|
|
being 8 positions) and then spaces to do the indentation, unless you reach 8
|
|
spaces which are replaced by a tab. This is the vim source file indentation
|
|
for example. In my opinion, this is the worst possible style.
|
|
|
|
--vim-output: output suitable to use inside vim:
|
|
set sts=0 | set tabstop=4 | set noexpandtab | set shiftwidth=4
|
|
|
|
"""
|
|
|
|
VERSION='1.4'
|
|
|
|
### Used when indentation is tab, to set tabstop in vim
|
|
DEFAULT_TAB_WIDTH = 8
|
|
|
|
### default values for files where indentation is not meaningful (empty files)
|
|
# possible values:
|
|
# DEFAULT_RESULT = ('space', 4 )
|
|
# DEFAULT_RESULT = ('space', 2 )
|
|
# DEFAULT_RESULT = ('space', 8 )
|
|
# DEFAULT_RESULT = ('tab', DEFAULT_TAB_WIDTH )
|
|
|
|
DEFAULT_RESULT = ('space', 8 )
|
|
|
|
VERBOSE_QUIET = 0
|
|
VERBOSE_INFO = 1
|
|
VERBOSE_DEBUG = 2
|
|
VERBOSE_DEEP_DEBUG = 3
|
|
|
|
DEFAULT_VERBOSITY = VERBOSE_QUIET
|
|
MAX_LINE = 1024
|
|
|
|
###
|
|
class LineType:
|
|
NoIndent = 'NoIndent'
|
|
SpaceOnly = 'SpaceOnly'
|
|
TabOnly = 'TabOnly'
|
|
Mixed = 'Mixed'
|
|
BeginSpace = 'BeginSpace'
|
|
|
|
def info( s ): log( VERBOSE_INFO, s )
|
|
def dbg( s ): log( VERBOSE_DEBUG, s )
|
|
def deepdbg( s ): log( VERBOSE_DEEP_DEBUG, s )
|
|
|
|
def log( level, s ):
|
|
if level <= IndentFinder.VERBOSITY:
|
|
print(s)
|
|
|
|
class IndentFinder:
|
|
"""
|
|
IndentFinder reports the indentation used in a source file. Its approach
|
|
is not tied to any particular language. It was tested successfully with
|
|
python, C, C++ and Java code.
|
|
|
|
How does it work ?
|
|
|
|
It scans each line of the entry file for a space character (white space or
|
|
tab) repeated until a non space character is found. Such a line
|
|
is considered to be a properly indented line of code. Blank lines and
|
|
comments line (starting with # or /* or * ) are ignored. Lines coming
|
|
after a line ending in '\\' have higher chance of being not properly
|
|
indented, and are thus ignored too.
|
|
|
|
Only the increment in indentation are fed in. Dedentation or maintaining
|
|
the same indentation is not taken into account when analysing a file. Increment
|
|
in indentation from zero indentation to some indentation is also ignored because
|
|
it's wrong in many cases (header file with many structures for example, do not always
|
|
obey the indentation of the rest of the code).
|
|
|
|
Each line is analysed as:
|
|
- SpaceOnly: indentation of more than 8 space
|
|
- TabOnly: indentation of tab only
|
|
- Mixed: indentation of tab, then less than 8 spaces
|
|
- BeginSpace: indentation of less than 8 space, that could be either a mixed indentation
|
|
or a pure space indentation.
|
|
- non-significant
|
|
|
|
Then two consecutive significant lines are then considered. The only valid combinations are:
|
|
- (NoIndent, BeginSpace) => space or mixed
|
|
- (NoIndent, Tab) => tab
|
|
- (BeginSpace, BeginSpace) => space or mixed
|
|
- (BeginSpace, SpaceOnly) => space
|
|
- (SpaceOnly, SpaceOnly) => space
|
|
- (TabOnly, TabOnly) => tab
|
|
- (TabOnly, Mixed) => mixed
|
|
- (Mixed, TabOnly) => mixed
|
|
|
|
The increment in number of spaces is then recorded.
|
|
|
|
At the end, the number of lines with space indentation, mixed space and tab indentation
|
|
are compared and a decision is made.
|
|
|
|
If no decision can be made, DEFAULT_RESULT is returned.
|
|
|
|
If IndentFinder ever reports wrong indentation, send me immediately a
|
|
mail, if possible with the offending file.
|
|
"""
|
|
|
|
def __init__(self, default_result=DEFAULT_RESULT):
|
|
self.clear()
|
|
self.default_result = default_result
|
|
|
|
VERBOSITY = DEFAULT_VERBOSITY
|
|
|
|
def parse_file_list( self, file_list ):
|
|
for fname in file_list:
|
|
self.parse_file( fname )
|
|
|
|
def parse_file( self, fname ):
|
|
f = open( fname )
|
|
l = f.readline()
|
|
i = 0
|
|
while( l ):
|
|
self.analyse_line( l )
|
|
l = f.readline()
|
|
if i > MAX_LINE:
|
|
break
|
|
i+=1
|
|
f.close()
|
|
|
|
def clear( self ):
|
|
self.lines = {}
|
|
for i in range(2,9): self.lines['space%d' % i] = 0
|
|
for i in range(2,9): self.lines['mixed%d' % i] = 0
|
|
self.lines['tab'] = 0
|
|
|
|
self.nb_processed_lines = 0
|
|
self.nb_indent_hint = 0
|
|
self.indent_re = re.compile( "^([ \t]+)([^ \t]+)" )
|
|
self.mixed_re = re.compile( "^(\t+)( +)$" )
|
|
self.skip_next_line = False
|
|
self.previous_line_info = None
|
|
|
|
def analyse_line( self, line ):
|
|
if line[-1:] == '\n':
|
|
line = line[:-1]
|
|
deepdbg( 'analyse_line: "%s"' % line.replace(' ', '.' ).replace('\t','\\t') )
|
|
self.nb_processed_lines += 1
|
|
|
|
skip_current_line = self.skip_next_line
|
|
self.skip_next_line = False
|
|
if line[-1:] == '\\':
|
|
deepdbg( 'analyse_line: Ignoring next line!' )
|
|
# skip lines after lines ending in \
|
|
self.skip_next_line = True
|
|
|
|
if skip_current_line:
|
|
deepdbg( 'analyse_line: Ignoring current line!' )
|
|
return
|
|
|
|
ret = self.analyse_line_indentation( line )
|
|
if ret:
|
|
self.nb_indent_hint += 1
|
|
deepdbg( 'analyse_line: Result of line analysis: %s' % str(ret) )
|
|
return ret
|
|
|
|
def analyse_line_type( self, line ):
|
|
'''Analyse the type of line and return (LineType, <indentation part of
|
|
the line>).
|
|
|
|
The function will reject improperly formatted lines (mixture of tab
|
|
and space for example) and comment lines.
|
|
'''
|
|
mixed_mode = False
|
|
tab_part = ''
|
|
space_part = ''
|
|
|
|
if len(line) > 0 and line[0] != ' ' and line[0] != '\t':
|
|
return (LineType.NoIndent, '')
|
|
|
|
mo = self.indent_re.match( line )
|
|
if not mo:
|
|
deepdbg( 'analyse_line_type: line is not indented' )
|
|
return None
|
|
|
|
indent_part = mo.group(1)
|
|
text_part = mo.group(2)
|
|
|
|
deepdbg( 'analyse_line_type: indent_part="%s" text_part="%s"' %
|
|
(indent_part.replace(' ', '.').replace('\t','\\t').replace('\n', '\\n' ),
|
|
text_part ) )
|
|
|
|
if text_part[0] == '*':
|
|
# continuation of a C/C++ comment, unlikely to be indented correctly
|
|
return None
|
|
|
|
if text_part[0:2] == '/*' or text_part[0] == '#':
|
|
# python, C/C++ comment, might not be indented correctly
|
|
return None
|
|
|
|
if '\t' in indent_part and ' ' in indent_part:
|
|
# mixed mode
|
|
mo = self.mixed_re.match( indent_part )
|
|
if not mo:
|
|
# line is not composed of '\t\t\t ', ignore it
|
|
return None
|
|
mixed_mode = True
|
|
tab_part = mo.group(1)
|
|
space_part = mo.group(2)
|
|
|
|
if mixed_mode:
|
|
if len(space_part) >= 8:
|
|
# this is not mixed mode, this is garbage !
|
|
return None
|
|
return (LineType.Mixed, tab_part, space_part )
|
|
|
|
if '\t' in indent_part:
|
|
return (LineType.TabOnly, indent_part)
|
|
|
|
if ' ' in indent_part:
|
|
if len(indent_part) < 8:
|
|
# this could be mixed mode too
|
|
return (LineType.BeginSpace, indent_part)
|
|
else:
|
|
# this is really a line indented with spaces
|
|
return (LineType.SpaceOnly, indent_part )
|
|
|
|
assert False, 'We should never get there !'
|
|
|
|
def analyse_line_indentation( self, line ):
|
|
previous_line_info = self.previous_line_info
|
|
current_line_info = self.analyse_line_type( line )
|
|
self.previous_line_info = current_line_info
|
|
|
|
if current_line_info == None or previous_line_info == None:
|
|
deepdbg('analyse_line_indentation: Not enough line info to analyse line: %s, %s' % (str(previous_line_info), str(current_line_info)))
|
|
return
|
|
|
|
t = (previous_line_info[0], current_line_info[0])
|
|
deepdbg( 'analyse_line_indentation: Indent analysis: %s %s' % t )
|
|
if (t == (LineType.TabOnly, LineType.TabOnly)
|
|
or t == (LineType.NoIndent, LineType.TabOnly) ):
|
|
if len(current_line_info[1]) - len(previous_line_info[1]) == 1 :
|
|
self.lines['tab'] += 1
|
|
return 'tab'
|
|
|
|
elif (t == (LineType.SpaceOnly, LineType.SpaceOnly)
|
|
or t == (LineType.BeginSpace, LineType.SpaceOnly)
|
|
or t == (LineType.NoIndent, LineType.SpaceOnly) ):
|
|
nb_space = len(current_line_info[1]) - len(previous_line_info[1])
|
|
if 1 < nb_space <= 8:
|
|
key = 'space%d' % nb_space
|
|
self.lines[key] += 1
|
|
return key
|
|
|
|
elif (t == (LineType.BeginSpace, LineType.BeginSpace)
|
|
or t == (LineType.NoIndent, LineType.BeginSpace) ):
|
|
nb_space = len(current_line_info[1]) - len(previous_line_info[1])
|
|
if 1 < nb_space <= 8:
|
|
key1 = 'space%d' % nb_space
|
|
key2 = 'mixed%d' % nb_space
|
|
self.lines[ key1 ] += 1
|
|
self.lines[ key2 ] += 1
|
|
return key1
|
|
|
|
elif t == (LineType.BeginSpace, LineType.TabOnly):
|
|
# we assume that mixed indentation used 8 characters tabs
|
|
if len(current_line_info[1]) == 1:
|
|
# more than one tab on the line --> not mixed mode !
|
|
nb_space = len(current_line_info[1])*8 - len(previous_line_info[1])
|
|
if 1 < nb_space <= 8:
|
|
key = 'mixed%d' % nb_space
|
|
self.lines[ key ] += 1
|
|
return key
|
|
|
|
elif t == (LineType.TabOnly, LineType.Mixed):
|
|
tab_part, space_part = tuple(current_line_info[1:3])
|
|
if len(previous_line_info[1]) == len(tab_part):
|
|
nb_space = len(space_part)
|
|
if 1 < nb_space <= 8:
|
|
key = 'mixed%d' % nb_space
|
|
self.lines[ key ] += 1
|
|
return key
|
|
|
|
elif t == (LineType.Mixed, LineType.TabOnly):
|
|
tab_part, space_part = previous_line_info[1:3]
|
|
if len(tab_part)+1 == len(current_line_info[1]):
|
|
nb_space = 8-len(space_part)
|
|
if 1 < nb_space <= 8:
|
|
key = 'mixed%d' % nb_space
|
|
self.lines[ key ] += 1
|
|
return key
|
|
else:
|
|
pass
|
|
|
|
return None
|
|
|
|
def results( self ):
|
|
dbg( "Nb of scanned lines : %d" % self.nb_processed_lines )
|
|
dbg( "Nb of indent hint : %d" % self.nb_indent_hint )
|
|
dbg( "Collected data:" )
|
|
for key in self.lines:
|
|
if self.lines[key] > 0:
|
|
dbg( '%s: %d' % (key, self.lines[key] ) )
|
|
|
|
max_line_space = max( [ self.lines['space%d'%i] for i in range(2,9) ] )
|
|
max_line_mixed = max( [ self.lines['mixed%d'%i] for i in range(2,9) ] )
|
|
max_line_tab = self.lines['tab']
|
|
|
|
dbg( 'max_line_space: %d' % max_line_space )
|
|
dbg( 'max_line_mixed: %d' % max_line_mixed )
|
|
dbg( 'max_line_tab: %d' % max_line_tab )
|
|
|
|
### Result analysis
|
|
#
|
|
# 1. Space indented file
|
|
# - lines indented with less than 8 space will fill mixed and space array
|
|
# - lines indented with 8 space or more will fill only the space array
|
|
# - almost no lines indented with tab
|
|
#
|
|
# => more lines with space than lines with mixed
|
|
# => more a lot more lines with space than tab
|
|
#
|
|
# 2. Tab indented file
|
|
# - most lines will be tab only
|
|
# - very few lines as mixed
|
|
# - very few lines as space only
|
|
#
|
|
# => a lot more lines with tab than lines with mixed
|
|
# => a lot more lines with tab than lines with space
|
|
#
|
|
# 3. Mixed tab/space indented file
|
|
# - some lines are tab-only (lines with exactly 8 step indentation)
|
|
# - some lines are space only (less than 8 space)
|
|
# - all other lines are mixed
|
|
#
|
|
# If mixed is tab + 2 space indentation:
|
|
# - a lot more lines with mixed than with tab
|
|
# If mixed is tab + 4 space indentation
|
|
# - as many lines with mixed than with tab
|
|
#
|
|
# If no lines exceed 8 space, there will be only lines with space
|
|
# and tab but no lines with mixed. Impossible to detect mixed indentation
|
|
# in this case, the file looks like it's actually indented as space only
|
|
# and will be detected so.
|
|
#
|
|
# => same or more lines with mixed than lines with tab only
|
|
# => same or more lines with mixed than lines with space only
|
|
#
|
|
|
|
|
|
result = None
|
|
|
|
# Detect space indented file
|
|
if max_line_space >= max_line_mixed and max_line_space > max_line_tab:
|
|
nb = 0
|
|
indent_value = None
|
|
for i in range(8,1,-1):
|
|
if self.lines['space%d'%i] > int( nb * 1.1 ) : # give a 10% threshold
|
|
indent_value = i
|
|
nb = self.lines[ 'space%d' % indent_value ]
|
|
|
|
if indent_value == None: # no lines
|
|
result = self.default_result
|
|
else:
|
|
result = ('space', indent_value )
|
|
|
|
# Detect tab files
|
|
elif max_line_tab > max_line_mixed and max_line_tab > max_line_space:
|
|
result = ('tab', DEFAULT_TAB_WIDTH )
|
|
|
|
# Detect mixed files
|
|
elif max_line_mixed >= max_line_tab and max_line_mixed > max_line_space:
|
|
nb = 0
|
|
indent_value = None
|
|
for i in range(8,1,-1):
|
|
if self.lines['mixed%d'%i] > int( nb * 1.1 ) : # give a 10% threshold
|
|
indent_value = i
|
|
nb = self.lines[ 'mixed%d' % indent_value ]
|
|
|
|
if indent_value == None: # no lines
|
|
result = self.default_result
|
|
else:
|
|
result = ('mixed', (8,indent_value) )
|
|
|
|
else:
|
|
# not enough information to make a decision
|
|
result = self.default_result
|
|
|
|
info( "Result: %s" % str( result ) )
|
|
return result
|
|
|
|
def __str__ (self):
|
|
itype, ival = self.results()
|
|
if itype != 'mixed':
|
|
return '%s %d' % (itype, ival)
|
|
else:
|
|
itab, ispace = ival
|
|
return '%s tab %d space %d' % (itype, itab, ispace)
|
|
|
|
|
|
def vim_output( self ):
|
|
result = self.results()
|
|
indent_type, n = result
|
|
if indent_type == "space":
|
|
# spaces:
|
|
# => set sts to the number of spaces
|
|
# => set tabstop to the number of spaces
|
|
# => expand tabs to spaces
|
|
# => set shiftwidth to the number of spaces
|
|
return "set sts=%d | set tabstop=%d | set expandtab | set shiftwidth=%d \" (%s %d)" % (n,n,n,indent_type,n)
|
|
|
|
elif indent_type == "tab":
|
|
# tab:
|
|
# => set sts to 0
|
|
# => set tabstop to preferred value
|
|
# => set expandtab to false
|
|
# => set shiftwidth to tabstop
|
|
return "set sts=0 | set tabstop=%d | set noexpandtab | set shiftwidth=%d \" (%s)" % (DEFAULT_TAB_WIDTH, DEFAULT_TAB_WIDTH, indent_type )
|
|
|
|
if indent_type == 'mixed':
|
|
tab_indent, space_indent = n
|
|
# tab:
|
|
# => set sts to 0
|
|
# => set tabstop to tab_indent
|
|
# => set expandtab to false
|
|
# => set shiftwidth to space_indent
|
|
return "set sts=4 | set tabstop=%d | set noexpandtab | set shiftwidth=%d \" (%s %d)" % (tab_indent, space_indent, indent_type, space_indent )
|
|
|
|
|
|
|
|
def main():
|
|
VIM_OUTPUT = 0
|
|
|
|
parser = argparse.ArgumentParser('File indent finder')
|
|
parser.add_argument('files', metavar='FILE', nargs='+', help='file to scan')
|
|
parser.add_argument('--vim-output', dest='vim', default=None, help='output suitable to use inside vim', action="store_true")
|
|
parser.add_argument('--default', dest='default', default='space', help='default type of indentation for files where indentation is not meaningful: could be space or tab')
|
|
parser.add_argument('--default-size', dest='default_size', default=8, help='default values for files where indentation is not meaningful')
|
|
parser.add_argument('-v', '--version', dest='version', help='print version', action="store_true")
|
|
args = parser.parse_args()
|
|
|
|
if args.version:
|
|
print('IndentFinder v%s' % VERSION)
|
|
return
|
|
if args.vim:
|
|
VIM_OUTPUT = 1
|
|
|
|
DEFAULT_RESULT = (args.default, args.default_size)
|
|
|
|
file_list = args.files
|
|
|
|
fi = IndentFinder()
|
|
|
|
if len(file_list) > 1:
|
|
# multiple files
|
|
for fname in file_list:
|
|
fi.clear()
|
|
fi.parse_file( fname )
|
|
if VIM_OUTPUT:
|
|
print("%s : %s" % (fname, fi.vim_output()))
|
|
else:
|
|
print("%s : %s" % (fname, str(fi)))
|
|
return
|
|
|
|
else:
|
|
# only one file, don't print filename
|
|
fi.parse_file_list( file_list )
|
|
if VIM_OUTPUT:
|
|
sys.stdout.write( fi.vim_output() )
|
|
else:
|
|
print(str(fi))
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|