contiki/backyard/tools/chakana/utils.py

737 lines
20 KiB
Python
Raw Normal View History

# Copyright (C) 2003-2007 Swedish Institute of Computer Science.
#
# Please refer to the file named LICENSE in the same directory as this
# file for licensing information.
#
# $Id: utils.py,v 1.1 2009/02/26 13:47:38 fros4943 Exp $
import errno
import math
import operator
import os
import popen2
import re
import string
import shutil
import sys
import tempfile
import threading
import traceback
import types
import chakana.command
import chakana.error
from chakana.debug import *
try:
global False
global True
_ = False
except NameError:
False = 0
True = 1
class DefaultDict(dict):
def __init__(self, defaultCreator, initialiser = (), inject = 0):
dict.__init__(self, initialiser)
if callable(defaultCreator):
self._defaultCreator = defaultCreator
else:
self._defaultCreator = lambda k: defaultCreator
self._inject = inject
def __getitem__(self, key):
try:
return dict.__getitem__(self, key)
except KeyError, err:
value = self._defaultCreator(key)
if self._inject:
self[key] = value
return self[key]
else:
return value
class UnbufferedFile:
def __init__(self, fileObj):
self._fileObj = fileObj
def __getattr__(self, attribute):
return getattr(self._fileObj, attribute)
def write(self, what):
self._fileObj.write(what)
self._fileObj.flush()
def copyFile(src, dest):
debug(Event, "cp " + src + " " + dest)
srcFile = file(src)
destFile = file(dest, "w")
destFile.write(srcFile.read())
srcFile.close()
destFile.close()
def unique(list):
ret = []
inList = {}
for elem in list:
if not elem in inList:
ret.append(elem)
inList[elem] = 1
return ret
def unorderedEqual(listOne, listTwo, predicate = lambda x, y: x == y):
if len(listOne) != len(listTwo):
return 0
list1 = list(listOne)
list2 = list(listTwo)
while list1 != []:
found = 0
for index in range(len(list2)):
if predicate(list1[0], list2[index]):
del list1[0]
del list2[index]
found = 1
break
if not found:
return 0
return 1
def reduceSequences(tupleSeq, oper = operator.add):
"""Return a list, where the item at position i is the reduced value
of all items at position i."""
if len(tupleSeq) == 0:
return []
return [reduce(oper, [s[i] for s in tupleSeq])
for i in range(len(tupleSeq[0]))]
def listRemove(theList, predicate):
"""Remove all items that match predicate in theList."""
if not callable(predicate):
return listRemove(theList, lambda x: x == predicate)
else:
index = 0
while index < len(theList):
if predicate(theList[index]):
del theList[index]
else:
index += 1
return theList
def logn(n, x):
return math.log(x) / math.log(n)
def longRange(* args):
"""As range, but works with long integers."""
if len(args) == 1:
start = 0L
stop = args[0]
step = 1L
elif len(args) == 2:
(start, stop) = args
step = 1L
else:
(start, stop, step) = args
if step == 0:
raise ValueError("longRange() arg 3 must not be zero")
ret = []
while cmp(start, stop) == cmp(0, step):
ret.append(start)
start += step
return ret
integerSuffixes = ("", "K", "M", "G", "T")
def abbreviateInteger(i, factor = 1024, minimise = 0):
suf = 0
while suf < len(integerSuffixes) - 1 and \
((i > 9999) or (minimise and (i != 0) and (i % factor == 0))):
suf += 1
i /= factor
return str(i) + integerSuffixes[suf]
def readSuffixedInt(intString, factor = 1024):
if intString[-1].isalpha():
return int(intString[: -1]) * (
factor ** list(integerSuffixes).index(intString[-1].upper()))
else:
return int(intString)
def normaliseIndex(index, seq):
"""Return a non-negative index relating to a sequence, or negative if
index < - len(seq)."""
if index < 0:
return len(seq) + index
else:
return index
def truncateIndex(index, seq):
"""Return a non-negative index relating to a sequence, less than or equal
to the length of the sequence."""
ret = normaliseIndex(index, seq)
if ret < 0:
return 0
length = len(seq)
if ret > length:
return length
return ret
class LazySlice:
"""Representation of slice into other sequence."""
def __init__(self, seq, start = None, stop = None, step = None):
self._sequence = seq
if step is None:
self._step = 1
else:
self._step = step
if self._step == 0:
# Provoke ValueError
[][0:1:0]
if self._step > 0:
startDefault = 0
stopDefault = len(seq)
indexShift = 0
else:
startDefault = len(seq) - 1
stopDefault = -1
indexShift = 1
if start is None:
self._start = startDefault
else:
self._start = truncateIndex(start + indexShift, seq) - indexShift
if stop is None:
self._stop = stopDefault
else:
self._stop = truncateIndex(stop + indexShift, seq) - indexShift
if (self._step > 0) != (self._start < self._stop):
self._stop = self._start
def __getitem__(self, key):
# debug(Debug2, "LS getitem " + str(key) + ": " + repr(self))
if type(key) == types.SliceType:
ret = LazySlice(self, key.start, key.stop, key.step)
#debug(Debug2, "LS getitem " + str(key) + " = " +
# repr(ret) + ", LS: " + repr(self))
return ret
index = normaliseIndex(key, self)
if index < 0:
# Provoke IndexError
[][1]
seqIndex = self._start + index * self._step
if self._step > 0:
if seqIndex >= self._stop:
[][1]
elif seqIndex <= self._stop:
[][1]
ret = self._sequence[seqIndex]
#debug(Debug2, "LS getitem " + str(key) + " (" + str(seqIndex) + ") = " +
# str(ret) + ", LS: " + repr(self))
return ret
def __len__(self):
return (abs(self._stop - self._start) + abs(self._step) - 1) / \
abs(self._step)
def __str__(self):
return str(list(self))
def __repr__(self):
return "LazySlice(start = " + str(self._start) + ", stop = " + \
str(self._stop) + ", step = " + str(self._step) + ",\n " + \
repr(self._sequence) + ")"
def getItem(sequence, key):
"""Obtain key in sequence. Key may be an integer or a slice object."""
if type(key) == types.SliceType:
if key.step is None:
return sequence[key.start : key.stop]
return sequence[key.start : key.stop : key.step]
return sequence[key]
def binarySearch(list, value, start = 0, end = None):
"Return the position where value should be inserted in a sorted list."
if end is None:
end = len(list)
if start == end:
return end
middle = start + (end - start) / 2
if list[middle] < value:
return binarySearch(list, value, middle + 1, end)
else:
return binarySearch(list, value, start, middle)
def binaryFind(list, value, *args, ** kwArgs):
"Return position of value in sorted list, or None if not found."
position = binarySearch(list, value, *args, ** kwArgs)
if (position == len(list)) or (list[position] != value):
return None
else:
return position
def compactIntListRepr(input, sort = 0):
if sort:
list = input[:]
list.sort()
else:
list = input
if __debug__:
tmp = list[:]
tmp.sort()
assert(tmp == list)
index = 0
ret = "["
sep = ""
compact = lambda l, i: (i + 1 < len(l)) and (l[i] + 1 == l[i + 1])
while index < len(list):
ret += sep + str(list[index])
if compact(list, index):
ret += "-"
while compact(list, index):
index += 1
ret += str(list[index])
index += 1
sep = ","
return ret + "]"
def predicatedIndex(sequence, predicate):
for i in range(len(sequence)):
if predicate(sequence[i]):
return i
raise ValueError()
class ArgumentBinder:
def __init__(self, function, argument, position = 0):
self.__function = function
self.__argument = argument
self.__position = position
def __call__(self, *args, **keywords):
newArgs = list(args)
newArgs[self.__position : self.__position] = [self.__argument]
return self.__function(*newArgs, **keywords)
class ArgumentTupleBinder:
def __init__(self, function, arguments):
self._function = function
self._arguments = arguments
def __call__(self, * args, ** kwArgs):
return self._function(* (self._arguments + args), ** kwArgs)
class KeywordArgumentBinder:
def __init__(self, function, ** keywords):
self._function = function
self._keywords = keywords
def __call__(self, * args, ** kwArgs):
kw = self._keywords
kw.update(kwArgs)
return self._function(* args, ** kw)
cppKeywords = ["asm", "do", "inline", "short", "typeid", "auto",
"double", "int", "signed", "typename", "bool",
"dynamic_cast", "long", "sizeof", "union", "break",
"else", "mutable", "static", "unsigned", "case",
"enum", "namespace", "static_cast", "using", "catch",
"explicit", "new", "struct", "virtual", "char",
"extern", "operator", "switch", "void", "class",
"false", "private", "template", "volatile", "const",
"float", "protected", "this", "wchar_t", "const_cast",
"for", "public", "throw", "while", "continue",
"friend", "register", "true", "default", "goto",
"reinterpret_cast", "try", "delete", "if", "return",
"typedef"]
def isCWord(str):
return not re.match("^[a-zA-Z_][a-zA-Z0-9_]*$", str) is None
def isCppIdentifier(str):
return isCWord(str) and (str not in cppKeywords)
def find(seq, predicate = lambda x: x):
for item in seq:
if predicate(item):
return item
return None
def allBest(seq, comparator = cmp):
if len(seq) < 2:
return seq
seq = seq[:]
ret = [seq.pop()]
for item in seq:
c = comparator(item, ret[0])
if c < 0:
ret = [item]
elif c == 0:
ret.append(item)
return ret
def readFile(fileName):
f = open(fileName)
try:
ret = f.read()
return ret
finally:
f.close()
def writeFile(fileName, contents):
f = open(fileName, "w")
try:
f.write(contents)
finally:
f.close()
def writeFileAtomic(fileName, contents, suffix = ".tmp"):
tmpFile = fileName + suffix
writeFile(tmpFile, contents)
try:
os.rename(tmpFile, fileName)
except:
if os.path.exists(fileName) and os.path.exists(tmpFile):
os.remove(tmpFile)
raise
class LineIndexed:
"""Read-only file object indexed by line number. The file object must
support seeking."""
def __init__(self, fileName, offset = 0):
if type(fileName) == type(""):
self._file = file(fileName)
else:
self._file = fileName
self._offset = offset
self._length = None
self._lineOffsets = [0]
def close(self):
self._file.close()
def __del__(self):
self.close()
def _seekForward(self, index):
if self._length == 0:
raise IndexError()
for index in range(len(self._lineOffsets), index + 1):
if self._lineOffsets[-1] == self._length:
raise IndexError()
self._file.seek(self._lineOffsets[-1] + self._offset)
line = self._file.readline()
if line == "":
self._length = self._lineOffsets[-1]
else:
self._lineOffsets.append(self._lineOffsets[-1] + len(line))
def __getitem__(self, key):
if type(key) == types.SliceType:
ret = LazySlice(self, key.start, key.stop, key.step)
return ret
if key < 0:
positiveLength = len(self) + key
if positiveLength < 0:
raise IndexError()
return self[positiveLength]
self._seekForward(key)
offset = self._lineOffsets[key]
if offset == self._length:
raise IndexError()
self._file.seek(offset + self._offset)
return self._file.readline()
def __len__(self):
if self._length == 0:
return 0
while self._lineOffsets[-1] != self._length:
self._seekForward(len(self._lineOffsets))
return len(self._lineOffsets) - 1
def lineOffset(self, index):
self._seekForward(index)
return self._lineOffsets[index]
def __str__(self):
return str(list(self))
def __repr__(self):
return "LineIndexed(offset = " + str(self._offset) + ", length = " + \
str(self._length) + ", lineOffsets: " + \
str(self._lineOffsets) + ")"
def getLine(fileName, count):
fileObj = file(fileName)
if count < 0:
return fileObj.readlines()[count]
for i in range(count + 1):
ret = fileObj.readline()
if ret == "":
raise IndexError()
return ret
def conditionalJoin(list, separator = " ", predicate = lambda e: e != ""):
if len(list) == 0:
return ""
rest = conditionalJoin(list[1:], separator, predicate)
if not predicate(rest):
return list[0]
if not predicate(list[0]):
return rest
else:
return list[0] + separator + rest
def pathComponents(path):
(head, tail) = os.path.split(path)
if tail == "":
if head == "":
return []
if head == "/":
return ["/"]
else:
return pathComponents(head)
else:
return pathComponents(head) + [tail]
def commonFirstElements(listOfLists):
if len(listOfLists) == 0:
return []
if len(listOfLists) == 1:
return listOfLists[0]
if (len(listOfLists[0]) == 0) or (len(listOfLists[1]) == 0) or \
(listOfLists[0][0] != listOfLists[1][0]):
return []
else:
return commonFirstElements([
[listOfLists[0][0]] +
commonFirstElements([listOfLists[0][1:], listOfLists[1][1:]])]
+ listOfLists[2:])
def commonPrefix(pathList):
"""Similar to os.path.commonprefix, but works on path components instead
of individual characters."""
assert(not isinstance(pathList, str))
components = commonFirstElements(map(pathComponents, pathList))
if components == []:
return ""
else:
return os.path.join(* components)
def pathIsBelow(path, directory):
"""Check if path lies below directory."""
return commonPrefix([path, directory]) == directory
def realPath(path):
"""Similar to os.path.realpath, but handles amd gracefully."""
ret = os.path.realpath(path)
for binDir in os.environ["PATH"].split(":") + ["/usr/sbin", "/sbin"]:
amq = os.path.join(binDir, "amq")
if os.path.isfile(amq):
try:
output = chakana.command.output(amq)
except chakana.error.CommandFailed, err:
debug(Debug, str(err))
# Assume amd is not running
return ret
for line in output.splitlines():
amdDir = line.split()[0]
mountDir = line.split()[-1]
if mountDir[0] == "/":
match = re.match('^(' + re.escape(line) + ')(/$)', ret)
if match:
return amdDir + ret[len(line) :]
return ret
return ret
def removeTree(path, ignoreErrors = False):
debug(Debug, "rm -rf " + path)
if os.path.isdir(path):
shutil.rmtree(path, ignore_errors = ignoreErrors)
def makeSymlink(value, dest, force = True, debugLevel = MinorEvent,
dryRun = False):
if os.path.islink(dest):
if os.readlink(dest) == value:
debug(Debug, "Link " + dest + " already points to " + value)
return
elif force:
debug(debugLevel, "Removing " + dest)
if not dryRun:
os.remove(dest)
else:
raise OSError((errno.EEXIST, "Link already exists", dest))
absValue = os.path.join(os.path.dirname(dest), value)
if (not dryRun) and (not force) and (not os.path.isfile(absValue)):
raise OSError((errno.ENOENT, "Link destination does not exist", absDest))
debug(debugLevel, "Linking " + dest + " to " + value)
if not dryRun:
os.symlink(value, dest)
def copyTree(src, dst, symlinks = False, predicate = lambda path: 1,
preserveTimes = False):
"""Similar to shutil.copytree, but allows existing destination and
passes exceptions."""
names = filter(predicate, os.listdir(src))
if not os.path.isdir(dst):
os.mkdir(dst)
for name in names:
srcname = os.path.join(src, name)
dstname = os.path.join(dst, name)
try:
if symlinks and os.path.islink(srcname):
linkto = os.readlink(srcname)
os.symlink(linkto, dstname)
elif os.path.isdir(srcname):
copyTree(srcname, dstname, symlinks, predicate, preserveTimes)
else:
if preserveTimes:
shutil.copy2(srcname, dstname)
else:
shutil.copy(srcname, dstname)
# XXX What about devices, sockets etc.?
except (IOError, os.error), why:
debug(Error, "Can't copy %s to %s: %s" %
(`srcname`, `dstname`, str(why)))
raise
def makeDirsSafe(dir, mode = 0777):
"""Similar to os.makedirs, but does not fail if another process is making
the directory simultaneously."""
while not os.path.isdir(dir):
try:
os.makedirs(dir, mode)
except OSError, err:
if err.errno != errno.EEXIST:
raise
def typeAssert(variable, type):
assert(isinstance(variable, type))
def intervalsOverlap(range1, range2):
if range1 > range2:
return intervalsOverlap(range2, range1)
if (range1[0] == range1[1]) or (range2[0] == range2[1]):
return 0
return range2[0] < range1[1]
class LexicalDistance:
def __init__(self, caseChange = 1, whiteSpaceChange = 2,
whiteSpaceRemoval = 3, ampersand = 5, other = 10,
whiteSpace = " _-'\""):
self.__caseChange = caseChange
self.__whiteSpaceChange = whiteSpaceChange
self.__whiteSpaceRemoval = whiteSpaceRemoval
self.__ampersand = ampersand
self.__other = other
self.__whiteSpace = whiteSpace
def removeWhiteSpace(self, str1, str2, limit):
if str1[0] in self.__whiteSpace:
if limit > self.__whiteSpaceRemoval:
return self(str1[1:], str2, limit - self.__whiteSpaceRemoval) + \
self.__whiteSpaceRemoval
return limit
def changeWhiteSpace(self, str1, str2, limit):
if (str1[0] in self.__whiteSpace) and (str2[0] in self.__whiteSpace):
if limit > self.__whiteSpaceChange:
return self(str1[1:], str2[1:], limit - self.__whiteSpaceChange) + \
self.__whiteSpaceChange
return limit
def changeCase(self, str1, str2, limit):
if str1[0].upper() == str2[0].upper():
if limit > self.__caseChange:
return self(str1[1:], str2[1:], limit - self.__caseChange) + \
self.__caseChange
return limit
def changeAmpersand(self, str1, str2, limit):
if (str1[0] == "&") and (str2[:3].lower() == "and"):
if limit > self.__ampersand:
return self(str1[1:], str2[3:], limit - self.__ampersand) + \
self.__ampersand
return limit
def removeOther(self, str1, str2, limit):
if limit > self.__other:
return self(str1[1:], str2, limit - self.__other) + self.__other
return limit
def changeOther(self, str1, str2, limit):
if limit > self.__other:
return self(str1[1:], str2[1:], limit - self.__other) + self.__other
return limit
def __call__(self, str1, str2, limit = 50):
ret = limit
if str1 == "":
if str2 == "":
return 0
ret = self.removeWhiteSpace(str2, str1, ret)
ret = self.removeOther(str2, str1, ret)
elif str2 == "":
ret = self(str2, str1, ret)
else:
if str1[0] == str2[0]:
ret = self(str1[1:], str2[1:], ret)
else:
ret = self.removeWhiteSpace(str1, str2, ret)
ret = self.removeWhiteSpace(str2, str1, ret)
ret = self.changeWhiteSpace(str1, str2, ret)
ret = self.changeCase(str1, str2, ret)
ret = self.changeAmpersand(str1, str2, ret)
ret = self.changeAmpersand(str2, str1, ret)
ret = self.changeOther(str1, str2, ret)
ret = self.removeOther(str1, str2, ret)
ret = self.removeOther(str2, str1, ret)
assert(ret <= limit)
return ret
class TryOperation:
def __init__(self, operation, description = None):
self.__operation = operation
if description is None:
self.__description = str(operation)
else:
self.__description = description
def __call__(self, * args, ** kwargs):
try:
self.__operation(* args, ** kwargs)
except KeyboardInterrupt:
raise
except:
import debug as d
d.debug(d.Error, self.__description + " failed\n" +
d.exceptionDump(sys.exc_info()))
def lineDirective():
fileInfo = traceback.extract_stack(None, 2)[0]
return "#line " + repr(fileInfo[1] + 1) + " \"" + fileInfo[0] + "\""
try:
mkstemp = tempfile.mkstemp
except AttributeError:
def mkstemp(suffix = "", prefix = None, dir = None, text = 0):
if prefix is None:
prefix = tempfile.gettempprefix()
fileName = tempfile.mktemp(suffix)
if dir is None:
dir = os.path.dirname(fileName)
newFileName = os.path.join(dir, prefix + os.path.basename(fileName))
if text:
flags = "w"
else:
flags = "wb"
return (file(newFileName, flags), newFileName)