panda3d/direct/src/showbase/PythonUtil.py

1106 lines
34 KiB
Python
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

import types
import string
import re
import math
import operator
import inspect
import os
import sys
import random
if __debug__:
import traceback
from direct.directutil import Verify
# NOTE: ifAbsentPut has been replaced with Python's dictionary's builtin setdefault
# before:
# ifAbsentPut(dict, key, defaultValue)
# after:
# dict.setdefault(key, defaultValue)
# Please use setdefault instead -- Joe
def unique(L1, L2):
"""Return a list containing all items in 'L1' that are not in 'L2'"""
L2 = dict([(k,None) for k in L2])
return [item for item in L1 if item not in L2]
def indent(stream, numIndents, str):
"""
Write str to stream with numIndents in front of it
"""
# To match emacs, instead of a tab character we will use 4 spaces
stream.write(' ' * numIndents + str)
def writeFsmTree(instance, indent = 0):
if hasattr(instance, 'parentFSM'):
writeFsmTree(instance.parentFSM, indent-2)
elif hasattr(instance, 'fsm'):
name = ''
if hasattr(instance.fsm, 'state'):
name = instance.fsm.state.name
print "%s: %s"%(instance.fsm.name, name)
if __debug__:
class StackTrace:
def __init__(self, label="", start=0, limit=None):
"""
label is a string (or anything that be be a string)
that is printed as part of the trace back.
This is just to make it easier to tell what the
stack trace is referring to.
start is an integer number of stack frames back
from the most recent. (This is automatically
bumped up by one to skip the __init__ call
to the StackTrace).
limit is an integer number of stack frames
to record (or None for unlimited).
"""
self.label = label
self.trace = traceback.extract_stack(sys._getframe(1+start), limit=10)
def __str__(self):
r = "Debug stack trace of %s (back %s frames):\n"%(
self.label, len(self.trace),)
for i in traceback.format_list(self.trace):
r+=i
return r
def traceFunctionCall(frame):
"""
return a string that shows the call frame with calling arguments.
e.g.
foo(x=234, y=135)
"""
f = frame
co = f.f_code
dict = f.f_locals
n = co.co_argcount
if co.co_flags & 4: n = n+1
if co.co_flags & 8: n = n+1
r=f.f_code.co_name+'('
comma=0 # formatting, whether we should type a comma.
for i in range(n):
name = co.co_varnames[i]
if name=='self':
continue
if comma:
r+=', '
else:
# ok, we skipped the first one, the rest get commas:
comma=1
r+=name
r+='='
if dict.has_key(name):
v=str(dict[name])
if len(v)>200:
r+="<too big for debug>"
else:
r+=str(dict[name])
else: r+="*** undefined ***"
return r+')'
def traceParentCall():
return traceFunctionCall(sys._getframe(2))
def printThisCall():
print traceFunctionCall(sys._getframe(1))
return 1 # to allow "assert printThisCall()"
def tron():
sys.settrace(trace)
def trace(frame, event, arg):
if event == 'line':
pass
elif event == 'call':
print traceFunctionCall(sys._getframe(1))
elif event == 'return':
print "returning"
elif event == 'exception':
print "exception"
return trace
def troff():
sys.settrace(None)
def apropos(obj, *args):
"""
Obsolete, use pdir
"""
print 'Use pdir instead'
def getClassLineage(obj):
"""
print object inheritance list
"""
if type(obj) == types.DictionaryType:
# Just a dictionary, return dictionary
return [obj]
elif type(obj) == types.InstanceType:
# Instance, make a list with the instance and its class interitance
return [obj] + getClassLineage(obj.__class__)
elif type(obj) == types.ClassType:
# Class, see what it derives from
lineage = [obj]
for c in obj.__bases__:
lineage = lineage + getClassLineage(c)
return lineage
else:
# Not what I'm looking for
return []
def pdir(obj, str = None, fOverloaded = 0, width = None,
fTruncate = 1, lineWidth = 75, wantPrivate = 0):
# Remove redundant class entries
uniqueLineage = []
for l in getClassLineage(obj):
if type(l) == types.ClassType:
if l in uniqueLineage:
break
uniqueLineage.append(l)
# Pretty print out directory info
uniqueLineage.reverse()
for obj in uniqueLineage:
_pdir(obj, str, fOverloaded, width, fTruncate, lineWidth, wantPrivate)
print
def _pdir(obj, str = None, fOverloaded = 0, width = None,
fTruncate = 1, lineWidth = 75, wantPrivate = 0):
"""
Print out a formatted list of members and methods of an instance or class
"""
def printHeader(name):
name = ' ' + name + ' '
length = len(name)
if length < 70:
padBefore = int((70 - length)/2.0)
padAfter = max(0,70 - length - padBefore)
header = '*' * padBefore + name + '*' * padAfter
print header
print
def printInstanceHeader(i, printHeader = printHeader):
printHeader(i.__class__.__name__ + ' INSTANCE INFO')
def printClassHeader(c, printHeader = printHeader):
printHeader(c.__name__ + ' CLASS INFO')
def printDictionaryHeader(d, printHeader = printHeader):
printHeader('DICTIONARY INFO')
# Print Header
if type(obj) == types.InstanceType:
printInstanceHeader(obj)
elif type(obj) == types.ClassType:
printClassHeader(obj)
elif type (obj) == types.DictionaryType:
printDictionaryHeader(obj)
# Get dict
if type(obj) == types.DictionaryType:
dict = obj
else:
dict = obj.__dict__
# Adjust width
if width:
maxWidth = width
else:
maxWidth = 10
keyWidth = 0
aproposKeys = []
privateKeys = []
remainingKeys = []
for key in dict.keys():
if not width:
keyWidth = len(key)
if str:
if re.search(str, key, re.I):
aproposKeys.append(key)
if (not width) and (keyWidth > maxWidth):
maxWidth = keyWidth
else:
if key[:1] == '_':
if wantPrivate:
privateKeys.append(key)
if (not width) and (keyWidth > maxWidth):
maxWidth = keyWidth
else:
remainingKeys.append(key)
if (not width) and (keyWidth > maxWidth):
maxWidth = keyWidth
# Sort appropriate keys
if str:
aproposKeys.sort()
else:
privateKeys.sort()
remainingKeys.sort()
# Print out results
if wantPrivate:
keys = aproposKeys + privateKeys + remainingKeys
else:
keys = aproposKeys + remainingKeys
format = '%-' + `maxWidth` + 's'
for key in keys:
value = dict[key]
if callable(value):
strvalue = `Signature(value)`
else:
strvalue = `value`
if fTruncate:
# Cut off line (keeping at least 1 char)
strvalue = strvalue[:max(1,lineWidth - maxWidth)]
print (format % key)[:maxWidth] + '\t' + strvalue
# Magic numbers: These are the bit masks in func_code.co_flags that
# reveal whether or not the function has a *arg or **kw argument.
_POS_LIST = 4
_KEY_DICT = 8
def _is_variadic(function):
return function.func_code.co_flags & _POS_LIST
def _has_keywordargs(function):
return function.func_code.co_flags & _KEY_DICT
def _varnames(function):
return function.func_code.co_varnames
def _getcode(f):
"""
_getcode(f)
This function returns the name and function object of a callable
object.
"""
def method_get(f):
return f.__name__, f.im_func
def function_get(f):
return f.__name__, f
def instance_get(f):
if hasattr(f, '__call__'):
method = f.__call__
if (type(method) == types.MethodType):
func = method.im_func
else:
func = method
return ("%s%s" % (f.__class__.__name__, '__call__'), func)
else:
s = ("Instance %s of class %s does not have a __call__ method" %
(f, f.__class__.__name__))
raise TypeError, s
def class_get(f):
if hasattr(f, '__init__'):
return f.__name__, f.__init__.im_func
else:
return f.__name__, lambda: None
codedict = { types.UnboundMethodType: method_get,
types.MethodType : method_get,
types.FunctionType : function_get,
types.InstanceType : instance_get,
types.ClassType : class_get,
}
try:
return codedict[type(f)](f)
except KeyError:
if callable(f): # eg, built-in functions and methods
# raise ValueError, "type %s not supported yet." % type(f)
return f.__name__, None
else:
raise TypeError, ("object %s of type %s is not callable." %
(f, type(f)))
class Signature:
def __init__(self, func):
self.type = type(func)
self.name, self.func = _getcode(func)
def ordinary_args(self):
n = self.func.func_code.co_argcount
return _varnames(self.func)[0:n]
def special_args(self):
n = self.func.func_code.co_argcount
x = {}
#
if _is_variadic(self.func):
x['positional'] = _varnames(self.func)[n]
if _has_keywordargs(self.func):
x['keyword'] = _varnames(self.func)[n+1]
elif _has_keywordargs(self.func):
x['keyword'] = _varnames(self.func)[n]
else:
pass
return x
def full_arglist(self):
base = list(self.ordinary_args())
x = self.special_args()
if x.has_key('positional'):
base.append(x['positional'])
if x.has_key('keyword'):
base.append(x['keyword'])
return base
def defaults(self):
defargs = self.func.func_defaults
args = self.ordinary_args()
mapping = {}
if defargs is not None:
for i in range(-1, -(len(defargs)+1), -1):
mapping[args[i]] = defargs[i]
else:
pass
return mapping
def __repr__(self):
if self.func:
defaults = self.defaults()
specials = self.special_args()
l = []
for arg in self.ordinary_args():
if defaults.has_key(arg):
l.append( arg + '=' + str(defaults[arg]) )
else:
l.append( arg )
if specials.has_key('positional'):
l.append( '*' + specials['positional'] )
if specials.has_key('keyword'):
l.append( '**' + specials['keyword'] )
return "%s(%s)" % (self.name, string.join(l, ', '))
else:
return "%s(?)" % self.name
def aproposAll(obj):
"""
Print out a list of all members and methods (including overloaded methods)
of an instance or class
"""
apropos(obj, fOverloaded = 1, fTruncate = 0)
def doc(obj):
if (isinstance(obj, types.MethodType)) or \
(isinstance(obj, types.FunctionType)):
print obj.__doc__
def adjust(command = None, dim = 1, parent = None, **kw):
"""
adjust(command = None, parent = None, **kw)
Popup and entry scale to adjust a parameter
Accepts any Slider keyword argument. Typical arguments include:
command: The one argument command to execute
min: The min value of the slider
max: The max value of the slider
resolution: The resolution of the slider
text: The label on the slider
These values can be accessed and/or changed after the fact
>>> vg = adjust()
>>> vg['min']
0.0
>>> vg['min'] = 10.0
>>> vg['min']
10.0
"""
# Make sure we enable Tk
from direct.tkwidgets import Valuator
# Set command if specified
if command:
kw['command'] = lambda x: apply(command, x)
if parent is None:
kw['title'] = command.__name__
kw['dim'] = dim
# Create toplevel if needed
if not parent:
vg = apply(Valuator.ValuatorGroupPanel, (parent,), kw)
else:
vg = apply(Valuator.ValuatorGroup,(parent,), kw)
vg.pack(expand = 1, fill = 'x')
return vg
def intersection(a, b):
"""
intersection(list, list):
"""
if not a: return []
if not b: return []
d = []
for i in a:
if (i in b) and (i not in d):
d.append(i)
for i in b:
if (i in a) and (i not in d):
d.append(i)
return d
def union(a, b):
"""
union(list, list):
"""
# Copy a
c = a[:]
for i in b:
if (i not in c):
c.append(i)
return c
def sameElements(a, b):
if len(a) != len(b):
return 0
for elem in a:
if elem not in b:
return 0
for elem in b:
if elem not in a:
return 0
return 1
def list2dict(L, value=None):
"""creates dict using elements of list, all assigned to same value"""
return dict([(k,value) for k in L])
def invertDict(D):
"""creates a dictionary by 'inverting' D; keys are placed in the new
dictionary under their corresponding value in the old dictionary.
Data will be lost if D contains any duplicate values.
>>> old = {'key1':1, 'key2':2}
>>> invertDict(old)
{1: 'key1', 2: 'key2'}
"""
n = {}
for key, value in D.items():
n[value] = key
return n
def invertDictLossless(D):
"""similar to invertDict, but values of new dict are lists of keys from
old dict. No information is lost.
>>> old = {'key1':1, 'key2':2, 'keyA':2}
>>> invertDictLossless(old)
{1: ['key1'], 2: ['key2', 'keyA']}
"""
n = {}
for key, value in D.items():
n.setdefault(value, [])
n[value].append(key)
return n
def uniqueElements(L):
"""are all elements of list unique?"""
return len(L) == len(list2dict(L))
def disjoint(L1, L2):
"""returns non-zero if L1 and L2 have no common elements"""
used = dict([(k,None) for k in L1])
for k in L2:
if k in used:
return 0
return 1
def contains(whole, sub):
"""
Return 1 if whole contains sub, 0 otherwise
"""
if (whole == sub):
return 1
for elem in sub:
# The first item you find not in whole, return 0
if elem not in whole:
return 0
# If you got here, whole must contain sub
return 1
def replace(list, old, new, all=0):
"""
replace 'old' with 'new' in 'list'
if all == 0, replace first occurrence
otherwise replace all occurrences
returns the number of items replaced
"""
if old not in list:
return 0
if not all:
i = list.index(old)
list[i] = new
return 1
else:
numReplaced = 0
for i in xrange(len(list)):
if list[i] == old:
numReplaced += 1
list[i] = new
return numReplaced
def reduceAngle(deg):
"""
Reduces an angle (in degrees) to a value in [-180..180)
"""
return (((deg + 180.) % 360.) - 180.)
def fitSrcAngle2Dest(src, dest):
"""
given a src and destination angle, returns an equivalent src angle
that is within [-180..180) of dest
examples:
fitSrcAngle2Dest(30,60) == 30
fitSrcAngle2Dest(60,30) == 60
fitSrcAngle2Dest(0,180) == 0
fitSrcAngle2Dest(-1,180) == 359
fitSrcAngle2Dest(-180,180) == 180
"""
return dest + reduceAngle(src - dest)
def fitDestAngle2Src(src, dest):
"""
given a src and destination angle, returns an equivalent dest angle
that is within [-180..180) of src
examples:
fitDestAngle2Src(30,60) == 60
fitDestAngle2Src(60,30) == 30
fitDestAngle2Src(0,180) == -180
fitDestAngle2Src(1,180) == 180
"""
return src + (reduceAngle(dest - src))
def closestDestAngle2(src, dest):
# The function above didn't seem to do what I wanted. So I hacked
# this one together. I can't really say I understand it. It's more
# from impirical observation... GRW
diff = src - dest
if diff > 180:
# if the difference is greater that 180 it's shorter to go the other way
return dest - 360
elif diff < -180:
# or perhaps the OTHER other way...
return dest + 360
else:
# otherwise just go to the original destination
return dest
def closestDestAngle(src, dest):
# The function above didn't seem to do what I wanted. So I hacked
# this one together. I can't really say I understand it. It's more
# from impirical observation... GRW
diff = src - dest
if diff > 180:
# if the difference is greater that 180 it's shorter to go the other way
return src - (diff - 360)
elif diff < -180:
# or perhaps the OTHER other way...
return src - (360 + diff)
else:
# otherwise just go to the original destination
return dest
def binaryRepr(number, max_length = 32):
# This will only work reliably for relatively small numbers.
# Increase the value of max_length if you think you're going
# to use long integers
assert number < 2L << max_length
shifts = map (operator.rshift, max_length * [number], \
range (max_length - 1, -1, -1))
digits = map (operator.mod, shifts, max_length * [2])
if not digits.count (1): return 0
digits = digits [digits.index (1):]
return string.join (map (repr, digits), '')
# constant profile defaults
PyUtilProfileDefaultFilename = 'profiledata'
PyUtilProfileDefaultLines = 80
PyUtilProfileDefaultSorts = ['cumulative', 'time', 'calls']
# call this from the prompt, and break back out to the prompt
# to stop profiling
#
# OR to do inline profiling, you must make a globally-visible
# function to be profiled, i.e. to profile 'self.load()', do
# something like this:
#
# def func(self=self):
# self.load()
# import __builtin__
# __builtin__.func = func
# PythonUtil.startProfile(cmd='func()', filename='profileData')
# del __builtin__.func
#
def startProfile(filename=PyUtilProfileDefaultFilename,
lines=PyUtilProfileDefaultLines,
sorts=PyUtilProfileDefaultSorts,
silent=0,
callInfo=1,
cmd='run()'):
import profile
profile.run(cmd, filename)
if not silent:
printProfile(filename, lines, sorts, callInfo)
# call this to see the results again
def printProfile(filename=PyUtilProfileDefaultFilename,
lines=PyUtilProfileDefaultLines,
sorts=PyUtilProfileDefaultSorts,
callInfo=1):
import pstats
s = pstats.Stats(filename)
s.strip_dirs()
for sort in sorts:
s.sort_stats(sort)
s.print_stats(lines)
if callInfo:
s.print_callees(lines)
s.print_callers(lines)
class Functor:
def __init__(self, function, *args, **kargs):
assert callable(function), "function should be a callable obj"
self._function = function
self._args = args
self._kargs = kargs
self.__name__ = 'Functor: %s' % self._function.__name__
self.__doc__ = self._function.__doc__
def __call__(self, *args, **kargs):
"""call function"""
_args = list(self._args)
_args.extend(args)
_kargs = self._kargs.copy()
_kargs.update(kargs)
return apply(self._function,_args,_kargs)
def bound(value, bound1, bound2):
"""
returns value if value is between bound1 and bound2
otherwise returns bound that is closer to value
"""
if bound1 > bound2:
return min(max(value, bound2), bound1)
else:
return min(max(value, bound1), bound2)
def lerp(v0, v1, t):
"""
returns a value lerped between v0 and v1, according to t
t == 0 maps to v0, t == 1 maps to v1
"""
return v0 + (t * (v1 - v0))
def boolEqual(a, b):
"""
returns true if a and b are both true or both false.
returns false otherwise
(a.k.a. xnor -- eXclusive Not OR).
"""
return (a and b) or not (a or b)
def lineupPos(i, num, spacing):
"""
use to line up a series of 'num' objects, in one dimension,
centered around zero
'i' is the index of the object in the lineup
'spacing' is the amount of space between objects in the lineup
"""
assert num >= 1
assert i >= 0 and i < num
pos = float(i) * spacing
return pos - ((float(spacing) * (num-1))/2.)
def formatElapsedSeconds(seconds):
"""
Returns a string of the form "mm:ss" or "hh:mm:ss" or "n days",
representing the indicated elapsed time in seconds.
"""
sign = ''
if seconds < 0:
seconds = -seconds
sign = '-'
# We use math.floor() instead of casting to an int, so we avoid
# problems with numbers that are too large to represent as
# type int.
seconds = math.floor(seconds)
hours = math.floor(seconds / (60 * 60))
if hours > 36:
days = math.floor((hours + 12) / 24)
return "%s%d days" % (sign, days)
seconds -= hours * (60 * 60)
minutes = (int)(seconds / 60)
seconds -= minutes * 60
if hours != 0:
return "%s%d:%02d:%02d" % (sign, hours, minutes, seconds)
else:
return "%s%d:%02d" % (sign, minutes, seconds)
def solveQuadratic(a, b, c):
# quadratic equation: ax^2 + bx + c = 0
# quadratic formula: x = [-b +/- sqrt(b^2 - 4ac)] / 2a
# returns None, root, or [root1, root2]
# a cannot be zero.
if a == 0.:
return None
# calculate the determinant (b^2 - 4ac)
D = (b * b) - (4. * a * c)
if D < 0:
# there are no solutions (sqrt(negative number) is undefined)
return None
elif D == 0:
# only one root
return (-b) / (2. * a)
else:
# OK, there are two roots
sqrtD = math.sqrt(D)
twoA = 2. * a
root1 = ((-b) - sqrtD) / twoA
root2 = ((-b) + sqrtD) / twoA
return [root1, root2]
def stackEntryInfo(depth=0, baseFileName=1):
"""
returns the sourcefilename, line number, and function name of
an entry in the stack.
'depth' is how far back to go in the stack; 0 is the caller of this
function, 1 is the function that called the caller of this function, etc.
by default, strips off the path of the filename; override with baseFileName
returns (fileName, lineNum, funcName) --> (string, int, string)
returns (None, None, None) on error
"""
try:
stack = None
frame = None
try:
stack = inspect.stack()
# add one to skip the frame associated with this function
frame = stack[depth+1]
filename = frame[1]
if baseFileName:
filename = os.path.basename(filename)
lineNum = frame[2]
funcName = frame[3]
result = (filename, lineNum, funcName)
finally:
del stack
del frame
except:
result = (None, None, None)
return result
def lineInfo(baseFileName=1):
"""
returns the sourcefilename, line number, and function name of the
code that called this function
(answers the question: 'hey lineInfo, where am I in the codebase?')
see stackEntryInfo, above, for info on 'baseFileName' and return types
"""
return stackEntryInfo(1)
def callerInfo(baseFileName=1):
"""
returns the sourcefilename, line number, and function name of the
caller of the function that called this function
(answers the question: 'hey callerInfo, who called me?')
see stackEntryInfo, above, for info on 'baseFileName' and return types
"""
return stackEntryInfo(2)
def lineTag(baseFileName=1, verbose=0, separator=':'):
"""
returns a string containing the sourcefilename and line number
of the code that called this function
(equivalent to lineInfo, above, with different return type)
see stackEntryInfo, above, for info on 'baseFileName'
if 'verbose' is false, returns a compact string of the form
'fileName:lineNum:funcName'
if 'verbose' is true, returns a longer string that matches the
format of Python stack trace dumps
returns empty string on error
"""
fileName, lineNum, funcName = callerInfo()
if fileName is None:
return ''
if verbose:
return 'File "%s", line %s, in %s' % (fileName, lineNum, funcName)
else:
return '%s%s%s%s%s' % (fileName, separator, lineNum, separator,
funcName)
def findPythonModule(module):
# Look along the python load path for the indicated filename.
# Returns the located pathname, or None if the filename is not
# found.
filename = module + '.py'
for dir in sys.path:
pathname = os.path.join(dir, filename)
if os.path.exists(pathname):
return pathname
return None
def describeException(backTrace = 4):
# When called in an exception handler, returns a string describing
# the current exception.
def byteOffsetToLineno(code, byte):
# Returns the source line number corresponding to the given byte
# offset into the indicated Python code module.
import array
lnotab = array.array('B', code.co_lnotab)
line = code.co_firstlineno
for i in range(0, len(lnotab),2):
byte -= lnotab[i]
if byte <= 0:
return line
line += lnotab[i+1]
return line
infoArr = sys.exc_info()
exception = infoArr[0]
exceptionName = getattr(exception, '__name__', None)
extraInfo = infoArr[1]
trace = infoArr[2]
stack = []
while trace.tb_next:
# We need to call byteOffsetToLineno to determine the true
# line number at which the exception occurred, even though we
# have both trace.tb_lineno and frame.f_lineno, which return
# the correct line number only in non-optimized mode.
frame = trace.tb_frame
module = frame.f_globals.get('__name__', None)
lineno = byteOffsetToLineno(frame.f_code, frame.f_lasti)
stack.append("%s:%s, " % (module, lineno))
trace = trace.tb_next
frame = trace.tb_frame
module = frame.f_globals.get('__name__', None)
lineno = byteOffsetToLineno(frame.f_code, frame.f_lasti)
stack.append("%s:%s, " % (module, lineno))
description = ""
for i in range(len(stack) - 1, max(len(stack) - backTrace, 0) - 1, -1):
description += stack[i]
description += "%s: %s" % (exceptionName, extraInfo)
return description
def mostDerivedLast(classList):
"""pass in list of classes. sorts list in-place, with derived classes
appearing after their bases"""
def compare(a,b):
if issubclass(a,b):
result=1
elif issubclass(b,a):
result=-1
else:
result=0
#print a,b,result
return result
classList.sort(compare)
def weightedChoice(choiceList, rng=random.random, sum=None):
"""given a list of (weight,item) pairs, chooses an item based on the
weights. rng must return 0..1. if you happen to have the sum of the
weights, pass it in 'sum'."""
# TODO: add support for dicts
if sum is None:
sum = 0.
for weight, item in choiceList:
sum += weight
rand = rng()
accum = rand * sum
for weight, item in choiceList:
accum -= weight
if accum <= 0.:
return item
# rand is ~1., and floating-point error prevented accum from hitting 0.
# Or you passed in a 'sum' that was was too large.
# Return the last item.
return item
def randFloat(a, b=0., rng=random.random):
"""returns a random float in [a,b]
call with single argument to generate random float between arg and zero
"""
return lerp(a,b,rng())
def normalDistrib(a, b, gauss=random.gauss):
"""
NOTE: assumes a < b
Returns random number between a and b, using gaussian distribution, with
mean=avg(a,b), and a standard deviation that fits ~99.7% of the curve
between a and b. Outlying results are clipped to a and b.
------------------------------------------------------------------------
http://www-stat.stanford.edu/~naras/jsm/NormalDensity/NormalDensity.html
The 68-95-99.7% Rule
====================
All normal density curves satisfy the following property which is often
referred to as the Empirical Rule:
68% of the observations fall within 1 standard deviation of the mean.
95% of the observations fall within 2 standard deviations of the mean.
99.7% of the observations fall within 3 standard deviations of the mean.
Thus, for a normal distribution, almost all values lie within 3 standard
deviations of the mean.
------------------------------------------------------------------------
In calculating our standard deviation, we divide (b-a) by 6, since the
99.7% figure includes 3 standard deviations _on_either_side_ of the mean.
"""
return max(a, min(b, gauss((a+b)*.5, (b-a)/6.)))
def randUint31(rng=random.random):
"""returns a random integer in [0..2^31).
rng must return float in [0..1]"""
return int(rng() * 0x7FFFFFFF)
def randInt32(rng=random.random):
"""returns a random integer in [-2147483648..2147483647].
rng must return float in [0..1]
"""
i = int(rng() * 0x7FFFFFFF)
if rng() < .5:
i += 0x80000000
return i
class Enum:
"""Pass in list of strings or string of comma-separated strings.
Items are accessible as instance.item, and are assigned unique,
increasing integer values. Pass in integer for 'start' to override
starting value.
Example:
>>> colors = Enum('red, green, blue')
>>> colors.red
0
>>> colors.green
1
>>> colors.blue
2
>>> colors.getString(colors.red)
'red'
"""
if __debug__:
# chars that cannot appear within an item string.
InvalidChars = string.whitespace
def _checkValidIdentifier(item):
invalidChars = string.whitespace+string.punctuation
invalidChars = invalidChars.replace('_','')
invalidFirstChars = invalidChars+string.digits
if item[0] in invalidFirstChars:
raise SyntaxError, ("Enum '%s' contains invalid first char" %
item)
if not disjoint(item, invalidChars):
for char in item:
if char in invalidChars:
raise SyntaxError, (
"Enum\n'%s'\ncontains illegal char '%s'" %
(item, char))
return 1
_checkValidIdentifier = staticmethod(_checkValidIdentifier)
def __init__(self, items, start=0):
if type(items) == types.StringType:
items = items.split(',')
self._stringTable = {}
# make sure we don't overwrite an existing element of the class
assert(self._checkExistingMembers(items))
assert(uniqueElements(items))
i = start
for item in items:
# remove leading/trailing whitespace
item = string.strip(item)
# is there anything left?
if len(item) == 0:
continue
# make sure there are no invalid characters
assert(Enum._checkValidIdentifier(item))
self.__dict__[item] = i
self._stringTable[i] = item
i += 1
def getString(self, value):
return self._stringTable[value]
def __contains__(self, value):
return value in self._stringTable
def __len__(self):
return len(self._stringTable)
if __debug__:
def _checkExistingMembers(self, items):
for item in items:
if hasattr(self, item):
return 0
return 1
############################################################
# class: Singleton
# Purpose: This provides a base metaclass for all classes
# that require one and only one instance.
#
# Example: class mySingleton:
# __metaclass__ = PythonUtil.Singleton
# def __init__(self,...):
# ...
#
# Note: This class is based on Python's New-Style Class
# design. An error will occur if a defined class
# attemps to inherit from a Classic-Style Class only,
# ie: class myClassX:
# def __init__(self, ...):
# ...
#
# class myNewClassX(myClassX):
# __metaclass__ = PythonUtil.Singleton
# def __init__(self, ...):
# myClassX.__init__(self, ...)
# ...
#
# This causes problems because myNewClassX is a
# New-Style class that inherits from only a
# Classic-Style base class. There are two ways
# simple ways to resolve this issue.
#
# First, if possible, make myClassX a
# New-Style class by inheriting from object
# object. IE: class myClassX(object):
#
# If for some reason that is not an option, make
# myNewClassX inherit from object and myClassX.
# IE: class myNewClassX(object, myClassX):
############################################################
class Singleton(type):
def __init__(cls,name,bases,dic):
super(Singleton,cls).__init__(name,bases,dic)
cls.instance=None
def __call__(cls,*args,**kw):
if cls.instance is None:
cls.instance=super(Singleton,cls).__call__(*args,**kw)
return cls.instance
class SingletonError(ValueError):
""" Used to indicate an inappropriate value for a Singleton."""