199 lines
6.2 KiB
Python
Executable File
199 lines
6.2 KiB
Python
Executable File
#! /usr/bin/env python
|
|
|
|
# Fix Python source files to use the new equality test operator, i.e.,
|
|
# if x = y: ...
|
|
# is changed to
|
|
# if x == y: ...
|
|
# The script correctly tokenizes the Python program to reliably
|
|
# distinguish between assignments and equality tests.
|
|
#
|
|
# Command line arguments are files or directories to be processed.
|
|
# Directories are searched recursively for files whose name looks
|
|
# like a python module.
|
|
# Symbolic links are always ignored (except as explicit directory
|
|
# arguments). Of course, the original file is kept as a back-up
|
|
# (with a "~" attached to its name).
|
|
# It complains about binaries (files containing null bytes)
|
|
# and about files that are ostensibly not Python files: if the first
|
|
# line starts with '#!' and does not contain the string 'python'.
|
|
#
|
|
# Changes made are reported to stdout in a diff-like format.
|
|
#
|
|
# Undoubtedly you can do this using find and sed or perl, but this is
|
|
# a nice example of Python code that recurses down a directory tree
|
|
# and uses regular expressions. Also note several subtleties like
|
|
# preserving the file's mode and avoiding to even write a temp file
|
|
# when no changes are needed for a file.
|
|
#
|
|
# NB: by changing only the function fixline() you can turn this
|
|
# into a program for a different change to Python programs...
|
|
|
|
import sys
|
|
import re
|
|
import os
|
|
from stat import *
|
|
import string
|
|
|
|
err = sys.stderr.write
|
|
dbg = err
|
|
rep = sys.stdout.write
|
|
|
|
def main():
|
|
bad = 0
|
|
if not sys.argv[1:]: # No arguments
|
|
err('usage: ' + sys.argv[0] + ' file-or-directory ...\n')
|
|
sys.exit(2)
|
|
for arg in sys.argv[1:]:
|
|
if os.path.isdir(arg):
|
|
if recursedown(arg): bad = 1
|
|
elif os.path.islink(arg):
|
|
err(arg + ': will not process symbolic links\n')
|
|
bad = 1
|
|
else:
|
|
if fix(arg): bad = 1
|
|
sys.exit(bad)
|
|
|
|
ispythonprog = re.compile('^[a-zA-Z0-9_]+\.py$')
|
|
def ispython(name):
|
|
return ispythonprog.match(name) >= 0
|
|
|
|
def recursedown(dirname):
|
|
dbg('recursedown(%r)\n' % (dirname,))
|
|
bad = 0
|
|
try:
|
|
names = os.listdir(dirname)
|
|
except os.error as msg:
|
|
err('%s: cannot list directory: %r\n' % (dirname, msg))
|
|
return 1
|
|
names.sort()
|
|
subdirs = []
|
|
for name in names:
|
|
if name in (os.curdir, os.pardir): continue
|
|
fullname = os.path.join(dirname, name)
|
|
if os.path.islink(fullname): pass
|
|
elif os.path.isdir(fullname):
|
|
subdirs.append(fullname)
|
|
elif ispython(name):
|
|
if fix(fullname): bad = 1
|
|
for fullname in subdirs:
|
|
if recursedown(fullname): bad = 1
|
|
return bad
|
|
|
|
def fix(filename):
|
|
## dbg('fix(%r)\n' % (dirname,))
|
|
try:
|
|
f = open(filename, 'r')
|
|
except IOError as msg:
|
|
err('%s: cannot open: %r\n' % (filename, msg))
|
|
return 1
|
|
head, tail = os.path.split(filename)
|
|
tempname = os.path.join(head, '@' + tail)
|
|
g = None
|
|
# If we find a match, we rewind the file and start over but
|
|
# now copy everything to a temp file.
|
|
lineno = 0
|
|
while 1:
|
|
line = f.readline()
|
|
if not line: break
|
|
lineno = lineno + 1
|
|
if g is None and '\0' in line:
|
|
# Check for binary files
|
|
err(filename + ': contains null bytes; not fixed\n')
|
|
f.close()
|
|
return 1
|
|
if lineno == 1 and g is None and line[:2] == '#!':
|
|
# Check for non-Python scripts
|
|
words = string.split(line[2:])
|
|
if words and re.search('[pP]ython', words[0]) < 0:
|
|
msg = filename + ': ' + words[0]
|
|
msg = msg + ' script; not fixed\n'
|
|
err(msg)
|
|
f.close()
|
|
return 1
|
|
while line[-2:] == '\\\n':
|
|
nextline = f.readline()
|
|
if not nextline: break
|
|
line = line + nextline
|
|
lineno = lineno + 1
|
|
newline = fixline(line)
|
|
if newline != line:
|
|
if g is None:
|
|
try:
|
|
g = open(tempname, 'w')
|
|
except IOError as msg:
|
|
f.close()
|
|
err('%s: cannot create: %r\n' % (tempname, msg))
|
|
return 1
|
|
f.seek(0)
|
|
lineno = 0
|
|
rep(filename + ':\n')
|
|
continue # restart from the beginning
|
|
rep(repr(lineno) + '\n')
|
|
rep('< ' + line)
|
|
rep('> ' + newline)
|
|
if g is not None:
|
|
g.write(newline)
|
|
|
|
# End of file
|
|
f.close()
|
|
if not g: return 0 # No changes
|
|
|
|
# Finishing touch -- move files
|
|
|
|
# First copy the file's mode to the temp file
|
|
try:
|
|
statbuf = os.stat(filename)
|
|
os.chmod(tempname, statbuf[ST_MODE] & 07777)
|
|
except os.error as msg:
|
|
err('%s: warning: chmod failed (%r)\n' % (tempname, msg))
|
|
# Then make a backup of the original file as filename~
|
|
try:
|
|
os.rename(filename, filename + '~')
|
|
except os.error as msg:
|
|
err('%s: warning: backup failed (%r)\n' % (filename, msg))
|
|
# Now move the temp file to the original file
|
|
try:
|
|
os.rename(tempname, filename)
|
|
except os.error as msg:
|
|
err('%s: rename failed (%r)\n' % (filename, msg))
|
|
return 1
|
|
# Return succes
|
|
return 0
|
|
|
|
|
|
from tokenize import tokenprog
|
|
|
|
match = {'if':':', 'elif':':', 'while':':', 'return':'\n', \
|
|
'(':')', '[':']', '{':'}', '`':'`'}
|
|
|
|
def fixline(line):
|
|
# Quick check for easy case
|
|
if '=' not in line: return line
|
|
|
|
i, n = 0, len(line)
|
|
stack = []
|
|
while i < n:
|
|
j = tokenprog.match(line, i)
|
|
if j < 0:
|
|
# A bad token; forget about the rest of this line
|
|
print '(Syntax error:)'
|
|
print line,
|
|
return line
|
|
a, b = tokenprog.regs[3] # Location of the token proper
|
|
token = line[a:b]
|
|
i = i+j
|
|
if stack and token == stack[-1]:
|
|
del stack[-1]
|
|
elif match.has_key(token):
|
|
stack.append(match[token])
|
|
elif token == '=' and stack:
|
|
line = line[:a] + '==' + line[b:]
|
|
i, n = a + len('=='), len(line)
|
|
elif token == '==' and not stack:
|
|
print '(Warning: \'==\' at top level:)'
|
|
print line,
|
|
return line
|
|
|
|
if __name__ == "__main__":
|
|
main()
|