cpython/Lib/tabnanny.py

305 lines
10 KiB
Python
Raw Normal View History

#! /usr/bin/env python
"""The Tab Nanny despises ambiguous indentation. She knows no mercy."""
# Released to the public domain, by Tim Peters, 15 April 1998.
# XXX Note: this is now a standard library module.
# XXX The API needs to undergo changes however; the current code is too
# XXX script-like. This will be addressed later.
__version__ = "6"
import os
import sys
import getopt
import tokenize
if not hasattr(tokenize, 'NL'):
raise ValueError("tokenize.NL doesn't exist -- tokenize module too old")
__all__ = ["check", "NannyNag", "process_tokens"]
verbose = 0
filename_only = 0
def errprint(*args):
sep = ""
for arg in args:
sys.stderr.write(sep + str(arg))
sep = " "
sys.stderr.write("\n")
def main():
global verbose, filename_only
try:
opts, args = getopt.getopt(sys.argv[1:], "qv")
except getopt.error, msg:
errprint(msg)
1998-04-06 11:45:26 -03:00
return
for o, a in opts:
if o == '-q':
filename_only = filename_only + 1
if o == '-v':
verbose = verbose + 1
1998-04-06 11:45:26 -03:00
if not args:
errprint("Usage:", sys.argv[0], "[-v] file_or_directory ...")
1998-04-06 11:45:26 -03:00
return
for arg in args:
check(arg)
class NannyNag:
def __init__(self, lineno, msg, line):
self.lineno, self.msg, self.line = lineno, msg, line
def get_lineno(self):
return self.lineno
def get_msg(self):
return self.msg
def get_line(self):
return self.line
def check(file):
if os.path.isdir(file) and not os.path.islink(file):
if verbose:
print "%s: listing directory" % `file`
names = os.listdir(file)
for name in names:
fullname = os.path.join(file, name)
if (os.path.isdir(fullname) and
not os.path.islink(fullname) or
os.path.normcase(name[-3:]) == ".py"):
check(fullname)
return
try:
f = open(file)
except IOError, msg:
errprint("%s: I/O Error: %s" % (`file`, str(msg)))
return
if verbose > 1:
print "checking", `file`, "..."
try:
process_tokens(tokenize.generate_tokens(f.readline))
except tokenize.TokenError, msg:
errprint("%s: Token Error: %s" % (`file`, str(msg)))
return
except NannyNag, nag:
badline = nag.get_lineno()
line = nag.get_line()
if verbose:
print "%s: *** Line %d: trouble in tab city! ***" % (
`file`, badline)
print "offending line:", `line`
print nag.get_msg()
else:
if ' ' in file: file = '"' + file + '"'
if filename_only: print file
else: print file, badline, `line`
return
if verbose:
print "%s: Clean bill of health." % `file`
class Whitespace:
# the characters used for space and tab
S, T = ' \t'
# members:
# raw
# the original string
# n
# the number of leading whitespace characters in raw
# nt
# the number of tabs in raw[:n]
# norm
# the normal form as a pair (count, trailing), where:
# count
# a tuple such that raw[:n] contains count[i]
# instances of S * i + T
# trailing
# the number of trailing spaces in raw[:n]
# It's A Theorem that m.indent_level(t) ==
# n.indent_level(t) for all t >= 1 iff m.norm == n.norm.
# is_simple
# true iff raw[:n] is of the form (T*)(S*)
def __init__(self, ws):
self.raw = ws
S, T = Whitespace.S, Whitespace.T
count = []
b = n = nt = 0
for ch in self.raw:
if ch == S:
n = n + 1
b = b + 1
elif ch == T:
n = n + 1
nt = nt + 1
if b >= len(count):
count = count + [0] * (b - len(count) + 1)
count[b] = count[b] + 1
b = 0
else:
break
self.n = n
self.nt = nt
self.norm = tuple(count), b
self.is_simple = len(count) <= 1
# return length of longest contiguous run of spaces (whether or not
# preceding a tab)
def longest_run_of_spaces(self):
count, trailing = self.norm
return max(len(count)-1, trailing)
def indent_level(self, tabsize):
# count, il = self.norm
# for i in range(len(count)):
# if count[i]:
# il = il + (i/tabsize + 1)*tabsize * count[i]
# return il
# quicker:
# il = trailing + sum (i/ts + 1)*ts*count[i] =
# trailing + ts * sum (i/ts + 1)*count[i] =
# trailing + ts * sum i/ts*count[i] + count[i] =
# trailing + ts * [(sum i/ts*count[i]) + (sum count[i])] =
# trailing + ts * [(sum i/ts*count[i]) + num_tabs]
# and note that i/ts*count[i] is 0 when i < ts
count, trailing = self.norm
il = 0
for i in range(tabsize, len(count)):
il = il + i/tabsize * count[i]
return trailing + tabsize * (il + self.nt)
# return true iff self.indent_level(t) == other.indent_level(t)
# for all t >= 1
def equal(self, other):
return self.norm == other.norm
# return a list of tuples (ts, i1, i2) such that
# i1 == self.indent_level(ts) != other.indent_level(ts) == i2.
# Intended to be used after not self.equal(other) is known, in which
# case it will return at least one witnessing tab size.
def not_equal_witness(self, other):
n = max(self.longest_run_of_spaces(),
other.longest_run_of_spaces()) + 1
a = []
for ts in range(1, n+1):
if self.indent_level(ts) != other.indent_level(ts):
a.append( (ts,
self.indent_level(ts),
other.indent_level(ts)) )
return a
# Return true iff self.indent_level(t) < other.indent_level(t)
# for all t >= 1.
# The algorithm is due to Vincent Broman.
# Easy to prove it's correct.
# XXXpost that.
# Trivial to prove n is sharp (consider T vs ST).
# Unknown whether there's a faster general way. I suspected so at
# first, but no longer.
# For the special (but common!) case where M and N are both of the
# form (T*)(S*), M.less(N) iff M.len() < N.len() and
# M.num_tabs() <= N.num_tabs(). Proof is easy but kinda long-winded.
# XXXwrite that up.
# Note that M is of the form (T*)(S*) iff len(M.norm[0]) <= 1.
def less(self, other):
if self.n >= other.n:
return 0
if self.is_simple and other.is_simple:
return self.nt <= other.nt
n = max(self.longest_run_of_spaces(),
other.longest_run_of_spaces()) + 1
# the self.n >= other.n test already did it for ts=1
for ts in range(2, n+1):
if self.indent_level(ts) >= other.indent_level(ts):
return 0
return 1
# return a list of tuples (ts, i1, i2) such that
# i1 == self.indent_level(ts) >= other.indent_level(ts) == i2.
# Intended to be used after not self.less(other) is known, in which
# case it will return at least one witnessing tab size.
def not_less_witness(self, other):
n = max(self.longest_run_of_spaces(),
other.longest_run_of_spaces()) + 1
a = []
for ts in range(1, n+1):
if self.indent_level(ts) >= other.indent_level(ts):
a.append( (ts,
self.indent_level(ts),
other.indent_level(ts)) )
return a
def format_witnesses(w):
import string
firsts = map(lambda tup: str(tup[0]), w)
prefix = "at tab size"
if len(w) > 1:
prefix = prefix + "s"
return prefix + " " + string.join(firsts, ', ')
def process_tokens(tokens):
INDENT = tokenize.INDENT
DEDENT = tokenize.DEDENT
NEWLINE = tokenize.NEWLINE
JUNK = tokenize.COMMENT, tokenize.NL
indents = [Whitespace("")]
check_equal = 0
for (type, token, start, end, line) in tokens:
2001-01-14 23:26:36 -04:00
if type == NEWLINE:
# a program statement, or ENDMARKER, will eventually follow,
# after some (possibly empty) run of tokens of the form
# (NL | COMMENT)* (INDENT | DEDENT+)?
# If an INDENT appears, setting check_equal is wrong, and will
# be undone when we see the INDENT.
check_equal = 1
elif type == INDENT:
check_equal = 0
thisguy = Whitespace(token)
if not indents[-1].less(thisguy):
witness = indents[-1].not_less_witness(thisguy)
msg = "indent not greater e.g. " + format_witnesses(witness)
raise NannyNag(start[0], msg, line)
indents.append(thisguy)
elif type == DEDENT:
# there's nothing we need to check here! what's important is
# that when the run of DEDENTs ends, the indentation of the
# program statement (or ENDMARKER) that triggered the run is
# equal to what's left at the top of the indents stack
# Ouch! This assert triggers if the last line of the source
# is indented *and* lacks a newline -- then DEDENTs pop out
# of thin air.
# assert check_equal # else no earlier NEWLINE, or an earlier INDENT
check_equal = 1
del indents[-1]
elif check_equal and type not in JUNK:
# this is the first "real token" following a NEWLINE, so it
# must be the first token of the next program statement, or an
# ENDMARKER; the "line" argument exposes the leading whitespace
# for this statement; in the case of ENDMARKER, line is an empty
# string, so will properly match the empty string with which the
# "indents" stack was seeded
check_equal = 0
thisguy = Whitespace(line)
if not indents[-1].equal(thisguy):
witness = indents[-1].not_equal_witness(thisguy)
msg = "indent not equal e.g. " + format_witnesses(witness)
raise NannyNag(start[0], msg, line)
if __name__ == '__main__':
main()