]> git.ipfire.org Git - thirdparty/Python/cpython.git/commitdiff
Companion module for new profile.py
authorGuido van Rossum <guido@python.org>
Thu, 23 Jun 1994 11:42:52 +0000 (11:42 +0000)
committerGuido van Rossum <guido@python.org>
Thu, 23 Jun 1994 11:42:52 +0000 (11:42 +0000)
Lib/pstats.py [new file with mode: 0644]

diff --git a/Lib/pstats.py b/Lib/pstats.py
new file mode 100644 (file)
index 0000000..5a4601a
--- /dev/null
@@ -0,0 +1,535 @@
+#
+# Class for printing reports on profiled python code. rev 1.0  4/1/94
+#
+# Based on prior profile module by Sjoerd Mullender...
+#   which was hacked somewhat by: Guido van Rossum
+#
+# see jprofile.doc and jprofile.py for more info.
+
+# Copyright 1994, by InfoSeek Corporation, all rights reserved.
+# Written by James Roskind
+# 
+# Permission to use, copy, modify, and distribute this Python software
+# and its associated documentation for any purpose (subject to the
+# restriction in the following sentence) without fee is hereby granted,
+# provided that the above copyright notice appears in all copies, and
+# that both that copyright notice and this permission notice appear in
+# supporting documentation, and that the name of InfoSeek not be used in
+# advertising or publicity pertaining to distribution of the software
+# without specific, written prior permission.  This permission is
+# explicitly restricted to the copying and modification of the software
+# to remain in Python, compiled Python, or other languages (such as C)
+# wherein the modified or derived code is exclusively imported into a
+# Python module.
+# 
+# INFOSEEK CORPORATION DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS
+# SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
+# FITNESS. IN NO EVENT SHALL INFOSEEK CORPORATION BE LIABLE FOR ANY
+# SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER
+# RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF
+# CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
+# CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+
+
+import os
+import time
+import string
+import marshal
+import regex
+
+#**************************************************************************
+# Class Stats documentation
+#**************************************************************************
+# This class is used for creating reports from data generated by the
+# Profile class.  It is a "friend" of that class, and imports data either
+# by direct access to members of Profile class, or by reading in a dictionary
+# that was emitted (via marshal) from the Profile class.
+#
+# The big change from the previous Profiler (in terms of raw functionality)
+# is that an "add()" method has been provided to combine Stats from
+# several distinct profile runs.  Both the constructor and the add()
+# method now take arbitrarilly many file names as arguments.
+#
+# All the print methods now take an argument that indicats how many lines
+# to print.  If the arg is a floating point number between 0 and 1.0, then
+# it is taken as a decimal percentage of the availabel lines to be printed
+# (e.g., .1 means print 10% of all available lines).  If it is an integer,
+# it is taken to mean the number of lines of data that you wish to have
+# printed.
+#
+# The sort_stats() method now processes some additionaly options (i.e., in
+# addition to the old -1, 0, 1, or 2).  It takes an arbitrary number of quoted
+# strings to select the sort order.  For example sort_stats('time', 'name')
+# sorts on the major key of "internal function time", and on the minor
+# key of 'the name of the function'.  Look at the two tables in sort_stats()
+# and get_sort_arg_defs(self) for more examples.
+#
+# All methods now return "self",  so you can string together commands like:
+#    Stats('foo', 'goo').strip_dirs().sort_stats('calls').\
+#                               print_stats(5).print_callers(5)
+# 
+#**************************************************************************
+import fpformat
+
+class Stats:
+       def __init__(self, *args):
+               if not len(args):
+                       arg = None
+               else:
+                       arg = args[0]
+                       args = args[1:]
+               self.init(arg)
+               apply(self.add, args).ignore()
+                       
+       def init(self, arg):
+               self.all_callees = None  # calc only if needed
+               self.files = []
+               self.fcn_list = None
+               self.total_tt = 0
+               self.total_calls = 0
+               self.prim_calls = 0
+               self.max_name_len = 0
+               self.top_level = {}
+               self.stats = {}
+               self.sort_arg_dict = {}
+               self.load_stats(arg)
+               trouble = 1
+               try:
+                       self.get_top_level_stats()
+                       trouble = 0
+               finally:
+                       if trouble:
+                               print "Invalid timing data",
+                               if self.files: print self.files[-1],
+                               print
+
+
+       def load_stats(self, arg):
+               if not arg:  self.stats = {}
+               elif type(arg) == type(""):
+                       f = open(arg, 'r')
+                       self.stats = marshal.load(f)
+                       f.close()
+                       try:
+                               file_stats = os.stat(arg)
+                               arg = time.ctime(file_stats[8]) + "    " + arg
+                       except:  # in case this is not unix
+                               pass
+                       self.files = [ arg ]
+               elif type(arg) == type(self):
+                       try:
+                               arg.create_stats()
+                               self.stats = arg.stats
+                               arg.stats = {}
+                       except:
+                               pass
+               if not self.stats:
+                       raise TypeError,  "Cannot create or construct a " \
+                                 + `self.__class__` \
+                                 + " object from '" + `arg` + "'"
+               return
+
+       def get_top_level_stats(self):
+               for func in self.stats.keys():
+                       cc, nc, tt, ct, callers = self.stats[func]
+                       self.total_calls = self.total_calls + nc
+                       self.prim_calls  = self.prim_calls  + cc
+                       self.total_tt    = self.total_tt    + tt
+                       if callers.has_key(("jprofile", 0, "profiler")):
+                               self.top_level[func] = None
+                       if len(func_std_string(func)) > self.max_name_len:
+                               self.max_name_len = len(func_std_string(func))
+                                       
+       def add(self, *arg_list):
+               if not arg_list: return self
+               if len(arg_list) > 1: apply(self.add, arg_list[1:])
+               other = arg_list[0]
+               if type(self) != type(other) or \
+                         self.__class__ != other.__class__:
+                       other = Stats(other)
+               self.files = self.files + other.files
+               self.total_calls = self.total_calls + other.total_calls
+               self.prim_calls = self.prim_calls + other.prim_calls
+               self.total_tt = self.total_tt + other.total_tt
+               for func in other.top_level.keys():
+                       self.top_level[func] = None
+
+               if self.max_name_len < other.max_name_len:
+                       self.max_name_len = other.max_name_len
+
+               self.fcn_list = None
+
+               for func in other.stats.keys():
+                       if self.stats.has_key(func):
+                               old_func_stat = self.stats[func]
+                       else:
+                               old_func_stat = (0, 0, 0, 0, {},)
+                       self.stats[func] = add_func_stats(old_func_stat, \
+                                 other.stats[func])
+               return self
+                       
+
+
+       # list the tuple indicies and directions for sorting,
+       # along with some printable description
+       sort_arg_dict_default = {\
+                 "calls"     : (((1,-1),              ), "call count"),\
+                 "cumulative": (((3,-1),              ), "cumulative time"),\
+                 "file"      : (((4, 1),              ), "file name"),\
+                 "line"      : (((5, 1),              ), "line number"),\
+                 "module"    : (((4, 1),              ), "file name"),\
+                 "name"      : (((6, 1),              ), "function name"),\
+                 "nfl"       : (((6, 1),(4, 1),(5, 1),), "name/file/line"), \
+                 "pcalls"    : (((0,-1),              ), "call count"),\
+                 "stdname"   : (((7, 1),              ), "standard name"),\
+                 "time"      : (((2,-1),              ), "internal time"),\
+                 }
+
+       # Expand all abbreviations that are unique
+       def get_sort_arg_defs(self):
+               if not self.sort_arg_dict:
+                       self.sort_arg_dict = dict = {}
+                       std_list = dict.keys()
+                       bad_list = {}
+                       for word in self.sort_arg_dict_default.keys():
+                               fragment = word
+                               while fragment:
+                                       if not fragment:
+                                               break
+                                       if dict.has_key(fragment):
+                                               bad_list[fragment] = 0
+                                               break
+                                       dict[fragment] = self. \
+                                                 sort_arg_dict_default[word]
+                                       fragment = fragment[:-1]
+                       for word in bad_list.keys():
+                               del dict[word]
+               return self.sort_arg_dict
+                       
+
+       def sort_stats(self, *field):
+               if not field:
+                       self.fcn_list = 0
+                       return self
+               if len(field) == 1 and type(field[0]) == type(1):
+                       # Be compatible with old profiler
+                       field = [ {-1: "stdname", \
+                                 0:"calls", \
+                                 1:"time", \
+                                 2: "cumulative" }  [ field[0] ] ]
+
+               sort_arg_defs = self.get_sort_arg_defs()
+               sort_tuple = ()
+               self.sort_type = ""
+               connector = ""
+               for word in field:
+                       sort_tuple = sort_tuple + sort_arg_defs[word][0]
+                       self.sort_type = self.sort_type + connector + \
+                                 sort_arg_defs[word][1]
+                       connector = ", "
+                                       
+               stats_list = []
+               for func in self.stats.keys():
+                       cc, nc, tt, ct, callers = self.stats[func]
+                       stats_list.append((cc, nc, tt, ct) + func_split(func) \
+                                          + (func_std_string(func), func,)  )
+
+               stats_list.sort(TupleComp(sort_tuple).compare)
+
+               self.fcn_list = fcn_list = []
+               for tuple in stats_list:
+                       fcn_list.append(tuple[-1])
+               return self
+
+
+       def reverse_order(self):
+               if self.fcn_list: self.fcn_list.reverse()
+               return self
+
+       def strip_dirs(self):
+               oldstats = self.stats
+               self.stats = newstats = {}
+               max_name_len = 0
+               for func in oldstats.keys():
+                       cc, nc, tt, ct, callers = oldstats[func]
+                       newfunc = func_strip_path(func)
+                       if len(func_std_string(newfunc)) > max_name_len:
+                               max_name_len = len(func_std_string(newfunc))
+                       newcallers = {}
+                       for func2 in callers.keys():
+                               newcallers[func_strip_path(func2)] = \
+                                         callers[func2]
+
+                       if newstats.has_key(newfunc):
+                               newstats[newfunc] = add_func_stats( \
+                                         newstats[newfunc],\
+                                         (cc, nc, tt, ct, newcallers))
+                       else:
+                               newstats[newfunc] = (cc, nc, tt, ct, newcallers)
+               old_top = self.top_level
+               self.top_level = new_top = {}
+               for func in old_top.keys():
+                       new_top[func_strip_path(func)] = None
+
+               self.max_name_len = max_name_len
+
+               self.fcn_list = None
+               self.all_callees = None
+               return self
+
+
+
+       def calc_callees(self):
+               if self.all_callees: return
+               self.all_callees = all_callees = {}
+               for func in self.stats.keys():
+                       if not all_callees.has_key(func):
+                               all_callees[func] = {}
+                       cc, nc, tt, ct, callers = self.stats[func]
+                       for func2 in callers.keys():
+                               if not all_callees.has_key(func2):
+                                       all_callees[func2] = {}
+                               all_callees[func2][func]  = callers[func2]
+               return
+
+        #******************************************************************
+       # The following functions support actual printing of reports
+        #******************************************************************
+
+       # Optional "amount" is either a line count, or a percentage of lines.
+
+       def eval_print_amount(self, sel, list, msg):
+               new_list = list
+               if type(sel) == type(""):
+                       new_list = []
+                       for func in list:
+                               if 0<=regex.search(sel, func_std_string(func)):
+                                       new_list.append(func)
+               else:
+                       count = len(list)
+                       if type(sel) == type(1.0) and 0.0 <= sel < 1.0:
+                               count = int (count * sel + .5)
+                               new_list = list[:count]
+                       elif type(sel) == type(1) and 0 <= sel < count:
+                               count = sel
+                               new_list = list[:count]
+               if len(list) != len(new_list):
+                       msg = msg + "   List reduced from " + `len(list)` \
+                                 + " to " + `len(new_list)` + \
+                                 " due to restriction <" + `sel` + ">\n"
+                       
+               return new_list, msg
+
+
+
+       def get_print_list(self, sel_list):
+               width = self.max_name_len
+               if self.fcn_list:
+                       list = self.fcn_list[:]
+                       msg = "   Ordered by: " + self.sort_type + '\n'
+               else:
+                       list = self.stats.keys()
+                       msg = "   Random listing order was used\n"
+
+               for selection in sel_list:
+                       list,msg = self.eval_print_amount(selection, list, msg)
+
+               count = len(list)
+
+               if not list:
+                       return 0, list
+               print msg
+               if count < len(self.stats):
+                       width = 0
+                       for func in list:
+                               if  len(func_std_string(func)) > width:
+                                       width = len(func_std_string(func))
+               return width+2, list
+               
+       def print_stats(self, *amount):
+               for filename in self.files:
+                       print filename
+               if self.files: print
+               indent = "        "
+               for func in self.top_level.keys():
+                       print indent, func_get_function_name(func)
+               
+               print  indent, self.total_calls, "function calls",
+               if self.total_calls != self.prim_calls:
+                       print "(" + `self.prim_calls`, "primitive calls)", 
+               print "in", fpformat.fix(self.total_tt, 3), "CPU seconds"
+               print
+               width, list = self.get_print_list(amount)
+               if list:
+                       self.print_title()
+                       for func in list:
+                               self.print_line(func)
+                       print 
+                       print
+               return self
+
+                       
+       def print_callees(self, *amount):
+               width, list = self.get_print_list(amount)
+               if list:
+                       self.calc_callees()
+
+                       self.print_call_heading(width, "called...")
+                       for func in list:
+                               if self.all_callees.has_key(func):
+                                       self.print_call_line(width, \
+                                                 func, self.all_callees[func])
+                               else:
+                                       self.print_call_line(width, func, {})
+                       print
+                       print
+               return self
+
+       def print_callers(self, *amount):
+               width, list = self.get_print_list(amount)
+               if list:
+                       self.print_call_heading(width, "was called by...")
+                       for func in list:
+                               cc, nc, tt, ct, callers = self.stats[func]
+                               self.print_call_line(width, func, callers)
+                       print
+                       print
+               return self
+
+       def print_call_heading(self, name_size, column_title):
+               print string.ljust("Function ", name_size) + column_title
+
+
+       def print_call_line(self, name_size, source, call_dict):
+               print string.ljust(func_std_string(source), name_size),
+               if not call_dict:
+                       print "--"
+                       return
+               clist = call_dict.keys()
+               clist.sort()
+               name_size = name_size + 1
+               indent = ""
+               for func in clist:
+                       name = func_std_string(func)
+                       print indent*name_size + name + '(' \
+                                 + `call_dict[func]`+')', \
+                                 f8(self.stats[func][3])
+                       indent = " "
+
+
+
+       def print_title(self):
+               print string.rjust('ncalls', 9),
+               print string.rjust('tottime', 8),
+               print string.rjust('percall', 8),
+               print string.rjust('cumtime', 8),
+               print string.rjust('percall', 8),
+               print 'filename:lineno(function)'
+
+
+       def print_line(self, func):  # hack : should print percentages
+               cc, nc, tt, ct, callers = self.stats[func]
+               c = `nc`
+               if nc != cc:
+                       c = c + '/' + `cc`
+               print string.rjust(c, 9),
+               print f8(tt),
+               if nc == 0:
+                       print ' '*8,
+               else:
+                       print f8(tt/nc),
+               print f8(ct),
+               if cc == 0:
+                       print ' '*8,
+               else:
+                       print f8(ct/cc),
+               print func_std_string(func)
+
+
+       def ignore(self):
+               pass # has no return value, so use at end of line :-)
+
+
+#**************************************************************************
+# class TupleComp Documentation
+#**************************************************************************
+# This class provides a generic function for comparing any two tuples.
+# Each instance records a list of tuple-indicies (from most significant
+# to least significant), and sort direction (ascending or decending) for
+# each tuple-index.  The compare functions can then be used as the function
+# argument to the system sort() function when a list of tuples need to be
+# sorted in the instances order.
+#**************************************************************************
+class TupleComp:
+       def __init__(self, comp_select_list):
+               self.comp_select_list = comp_select_list
+
+       def compare (self, left, right):
+               for index, direction in self.comp_select_list:
+                       l = left[index]
+                       r = right[index]
+                       if l < r:
+                               return -direction
+                       if l > r:
+                               return direction
+               return 0
+
+               
+
+#**************************************************************************
+
+def func_strip_path(func_name):
+       file, line, name = func_name
+       return os.path.basename(file), line, name
+
+def func_get_function_name(func):
+       return func[2]
+
+def func_std_string(func_name): # match what old profile produced
+       file, line, name = func_name
+       return file + ":" + `line` + "(" + name + ")"
+
+def func_split(func_name):
+       return func_name
+
+#**************************************************************************
+# The following functions combine statists for pairs functions.
+# The bulk of the processing involves correctly handling "call" lists,
+# such as callers and callees. 
+#**************************************************************************
+
+    # Add together all the stats for two profile entries
+def add_func_stats(target, source):                            
+       cc, nc, tt, ct, callers = source
+       t_cc, t_nc, t_tt, t_ct, t_callers = target
+       return (cc+t_cc, nc+t_nc, tt+t_tt, ct+t_ct, \
+                 add_callers(t_callers, callers))
+
+
+    # Combine two caller lists in a single list.
+def add_callers(target, source):
+       new_callers = {}
+       for func in target.keys():
+               new_callers[func] = target[func]
+       for func in source.keys():
+               if new_callers.has_key(func):
+                       new_callers[func] = source[func] + new_callers[func]
+               else:
+                       new_callers[func] = source[func]
+       return new_callers
+
+     # Sum the caller statistics to get total number of calls recieved
+def count_calls(callers):
+       nc = 0
+       for func in callers.keys():
+               nc = nc + callers[func]
+       return nc
+
+#**************************************************************************
+# The following functions support printing of reports
+#**************************************************************************
+
+def f8(x):
+       return string.rjust(fpformat.fix(x, 3), 8)
+
+