# -*- coding: utf-8 -*- """ debug.py - Functions to aid in debugging Copyright 2010 Luke Campagnola Distributed under MIT/X11 license. See license.txt for more infomation. """ import sys, traceback, time, gc, re, types, weakref, inspect, os, cProfile from . import ptime from numpy import ndarray from .Qt import QtCore, QtGui __ftraceDepth = 0 def ftrace(func): """Decorator used for marking the beginning and end of function calls. Automatically indents nested calls. """ def w(*args, **kargs): global __ftraceDepth pfx = " " * __ftraceDepth print(pfx + func.__name__ + " start") __ftraceDepth += 1 try: rv = func(*args, **kargs) finally: __ftraceDepth -= 1 print(pfx + func.__name__ + " done") return rv return w def getExc(indent=4, prefix='| '): tb = traceback.format_exc() lines = [] for l in tb.split('\n'): lines.append(" "*indent + prefix + l) return '\n'.join(lines) def printExc(msg='', indent=4, prefix='|'): """Print an error message followed by an indented exception backtrace (This function is intended to be called within except: blocks)""" exc = getExc(indent, prefix + ' ') print("[%s] %s\n" % (time.strftime("%H:%M:%S"), msg)) print(" "*indent + prefix + '='*30 + '>>') print(exc) print(" "*indent + prefix + '='*30 + '<<') def printTrace(msg='', indent=4, prefix='|'): """Print an error message followed by an indented stack trace""" trace = backtrace(1) #exc = getExc(indent, prefix + ' ') print("[%s] %s\n" % (time.strftime("%H:%M:%S"), msg)) print(" "*indent + prefix + '='*30 + '>>') for line in trace.split('\n'): print(" "*indent + prefix + " " + line) print(" "*indent + prefix + '='*30 + '<<') def backtrace(skip=0): return ''.join(traceback.format_stack()[:-(skip+1)]) def listObjs(regex='Q', typ=None): """List all objects managed by python gc with class name matching regex. Finds 'Q...' classes by default.""" if typ is not None: return [x for x in gc.get_objects() if isinstance(x, typ)] else: return [x for x in gc.get_objects() if re.match(regex, type(x).__name__)] def findRefPath(startObj, endObj, maxLen=8, restart=True, seen={}, path=None, ignore=None): """Determine all paths of object references from startObj to endObj""" refs = [] if path is None: path = [endObj] if ignore is None: ignore = {} ignore[id(sys._getframe())] = None ignore[id(path)] = None ignore[id(seen)] = None prefix = " "*(8-maxLen) #print prefix + str(map(type, path)) prefix += " " if restart: #gc.collect() seen.clear() gc.collect() newRefs = [r for r in gc.get_referrers(endObj) if id(r) not in ignore] ignore[id(newRefs)] = None #fo = allFrameObjs() #newRefs = [] #for r in gc.get_referrers(endObj): #try: #if r not in fo: #newRefs.append(r) #except: #newRefs.append(r) for r in newRefs: #print prefix+"->"+str(type(r)) if type(r).__name__ in ['frame', 'function', 'listiterator']: #print prefix+" FRAME" continue try: if any([r is x for x in path]): #print prefix+" LOOP", objChainString([r]+path) continue except: print(r) print(path) raise if r is startObj: refs.append([r]) print(refPathString([startObj]+path)) continue if maxLen == 0: #print prefix+" END:", objChainString([r]+path) continue ## See if we have already searched this node. ## If not, recurse. tree = None try: cache = seen[id(r)] if cache[0] >= maxLen: tree = cache[1] for p in tree: print(refPathString(p+path)) except KeyError: pass ignore[id(tree)] = None if tree is None: tree = findRefPath(startObj, r, maxLen-1, restart=False, path=[r]+path, ignore=ignore) seen[id(r)] = [maxLen, tree] ## integrate any returned results if len(tree) == 0: #print prefix+" EMPTY TREE" continue else: for p in tree: refs.append(p+[r]) #seen[id(r)] = [maxLen, refs] return refs def objString(obj): """Return a short but descriptive string for any object""" try: if type(obj) in [int, float]: return str(obj) elif isinstance(obj, dict): if len(obj) > 5: return "" % (",".join(list(obj.keys())[:5])) else: return "" % (",".join(list(obj.keys()))) elif isinstance(obj, str): if len(obj) > 50: return '"%s..."' % obj[:50] else: return obj[:] elif isinstance(obj, ndarray): return "" % (str(obj.dtype), str(obj.shape)) elif hasattr(obj, '__len__'): if len(obj) > 5: return "<%s [%s,...]>" % (type(obj).__name__, ",".join([type(o).__name__ for o in obj[:5]])) else: return "<%s [%s]>" % (type(obj).__name__, ",".join([type(o).__name__ for o in obj])) else: return "<%s %s>" % (type(obj).__name__, obj.__class__.__name__) except: return str(type(obj)) def refPathString(chain): """Given a list of adjacent objects in a reference path, print the 'natural' path names (ie, attribute names, keys, and indexes) that follow from one object to the next .""" s = objString(chain[0]) i = 0 while i < len(chain)-1: #print " -> ", i i += 1 o1 = chain[i-1] o2 = chain[i] cont = False if isinstance(o1, list) or isinstance(o1, tuple): if any([o2 is x for x in o1]): s += "[%d]" % o1.index(o2) continue #print " not list" if isinstance(o2, dict) and hasattr(o1, '__dict__') and o2 == o1.__dict__: i += 1 if i >= len(chain): s += ".__dict__" continue o3 = chain[i] for k in o2: if o2[k] is o3: s += '.%s' % k cont = True continue #print " not __dict__" if isinstance(o1, dict): try: if o2 in o1: s += "[key:%s]" % objString(o2) continue except TypeError: pass for k in o1: if o1[k] is o2: s += "[%s]" % objString(k) cont = True continue #print " not dict" #for k in dir(o1): ## Not safe to request attributes like this. #if getattr(o1, k) is o2: #s += ".%s" % k #cont = True #continue #print " not attr" if cont: continue s += " ? " sys.stdout.flush() return s def objectSize(obj, ignore=None, verbose=False, depth=0, recursive=False): """Guess how much memory an object is using""" ignoreTypes = [types.MethodType, types.UnboundMethodType, types.BuiltinMethodType, types.FunctionType, types.BuiltinFunctionType] ignoreRegex = re.compile('(method-wrapper|Flag|ItemChange|Option|Mode)') if ignore is None: ignore = {} indent = ' '*depth try: hash(obj) hsh = obj except: hsh = "%s:%d" % (str(type(obj)), id(obj)) if hsh in ignore: return 0 ignore[hsh] = 1 try: size = sys.getsizeof(obj) except TypeError: size = 0 if isinstance(obj, ndarray): try: size += len(obj.data) except: pass if recursive: if type(obj) in [list, tuple]: if verbose: print(indent+"list:") for o in obj: s = objectSize(o, ignore=ignore, verbose=verbose, depth=depth+1) if verbose: print(indent+' +', s) size += s elif isinstance(obj, dict): if verbose: print(indent+"list:") for k in obj: s = objectSize(obj[k], ignore=ignore, verbose=verbose, depth=depth+1) if verbose: print(indent+' +', k, s) size += s #elif isinstance(obj, QtCore.QObject): #try: #childs = obj.children() #if verbose: #print indent+"Qt children:" #for ch in childs: #s = objectSize(obj, ignore=ignore, verbose=verbose, depth=depth+1) #size += s #if verbose: #print indent + ' +', ch.objectName(), s #except: #pass #if isinstance(obj, types.InstanceType): gc.collect() if verbose: print(indent+'attrs:') for k in dir(obj): if k in ['__dict__']: continue o = getattr(obj, k) if type(o) in ignoreTypes: continue strtyp = str(type(o)) if ignoreRegex.search(strtyp): continue #if isinstance(o, types.ObjectType) and strtyp == "": #continue #if verbose: #print indent, k, '?' refs = [r for r in gc.get_referrers(o) if type(r) != types.FrameType] if len(refs) == 1: s = objectSize(o, ignore=ignore, verbose=verbose, depth=depth+1) size += s if verbose: print(indent + " +", k, s) #else: #if verbose: #print indent + ' -', k, len(refs) return size class GarbageWatcher(object): """ Convenient dictionary for holding weak references to objects. Mainly used to check whether the objects have been collect yet or not. Example: gw = GarbageWatcher() gw['objName'] = obj gw['objName2'] = obj2 gw.check() """ def __init__(self): self.objs = weakref.WeakValueDictionary() self.allNames = [] def add(self, obj, name): self.objs[name] = obj self.allNames.append(name) def __setitem__(self, name, obj): self.add(obj, name) def check(self): """Print a list of all watched objects and whether they have been collected.""" gc.collect() dead = self.allNames[:] alive = [] for k in self.objs: dead.remove(k) alive.append(k) print("Deleted objects:", dead) print("Live objects:", alive) def __getitem__(self, item): return self.objs[item] class Profiler: """Simple profiler allowing measurement of multiple time intervals. Arguments: msg: message to print at start and finish of profiling disabled: If true, profiler does nothing (so you can leave it in place) delayed: If true, all messages are printed after call to finish() (this can result in more accurate time step measurements) globalDelay: if True, all nested profilers delay printing until the top level finishes Example: prof = Profiler('Function') ... do stuff ... prof.mark('did stuff') ... do other stuff ... prof.mark('did other stuff') prof.finish() """ depth = 0 msgs = [] def __init__(self, msg="Profiler", disabled=False, delayed=True, globalDelay=True): self.disabled = disabled if disabled: return self.markCount = 0 self.finished = False self.depth = Profiler.depth Profiler.depth += 1 if not globalDelay: self.msgs = [] self.delayed = delayed self.msg = " "*self.depth + msg msg2 = self.msg + " >>> Started" if self.delayed: self.msgs.append(msg2) else: print(msg2) self.t0 = ptime.time() self.t1 = self.t0 def mark(self, msg=None): if self.disabled: return if msg is None: msg = str(self.markCount) self.markCount += 1 t1 = ptime.time() msg2 = " "+self.msg+" "+msg+" "+"%gms" % ((t1-self.t1)*1000) if self.delayed: self.msgs.append(msg2) else: print(msg2) self.t1 = ptime.time() ## don't measure time it took to print def finish(self, msg=None): if self.disabled or self.finished: return if msg is not None: self.mark(msg) t1 = ptime.time() msg = self.msg + ' <<< Finished, total time: %gms' % ((t1-self.t0)*1000) if self.delayed: self.msgs.append(msg) if self.depth == 0: for line in self.msgs: print(line) Profiler.msgs = [] else: print(msg) Profiler.depth = self.depth self.finished = True def profile(code, name='profile_run', sort='cumulative', num=30): """Common-use for cProfile""" cProfile.run(code, name) stats = pstats.Stats(name) stats.sort_stats(sort) stats.print_stats(num) return stats #### Code for listing (nearly) all objects in the known universe #### http://utcc.utoronto.ca/~cks/space/blog/python/GetAllObjects # Recursively expand slist's objects # into olist, using seen to track # already processed objects. def _getr(slist, olist, first=True): i = 0 for e in slist: oid = id(e) typ = type(e) if oid in olist or typ is int: ## or e in olist: ## since we're excluding all ints, there is no longer a need to check for olist keys continue olist[oid] = e if first and (i%1000) == 0: gc.collect() tl = gc.get_referents(e) if tl: _getr(tl, olist, first=False) i += 1 # The public function. def get_all_objects(): """Return a list of all live Python objects (excluding int and long), not including the list itself.""" gc.collect() gcl = gc.get_objects() olist = {} _getr(gcl, olist) del olist[id(olist)] del olist[id(gcl)] del olist[id(sys._getframe())] return olist def lookup(oid, objects=None): """Return an object given its ID, if it exists.""" if objects is None: objects = get_all_objects() return objects[oid] class ObjTracker(object): """ Tracks all objects under the sun, reporting the changes between snapshots: what objects are created, deleted, and persistent. This class is very useful for tracking memory leaks. The class goes to great (but not heroic) lengths to avoid tracking its own internal objects. Example: ot = ObjTracker() # takes snapshot of currently existing objects ... do stuff ... ot.diff() # prints lists of objects created and deleted since ot was initialized ... do stuff ... ot.diff() # prints lists of objects created and deleted since last call to ot.diff() # also prints list of items that were created since initialization AND have not been deleted yet # (if done correctly, this list can tell you about objects that were leaked) arrays = ot.findPersistent('ndarray') ## returns all objects matching 'ndarray' (string match, not instance checking) ## that were considered persistent when the last diff() was run describeObj(arrays[0]) ## See if we can determine who has references to this array """ allObjs = {} ## keep track of all objects created and stored within class instances allObjs[id(allObjs)] = None def __init__(self): self.startRefs = {} ## list of objects that exist when the tracker is initialized {oid: weakref} ## (If it is not possible to weakref the object, then the value is None) self.startCount = {} self.newRefs = {} ## list of objects that have been created since initialization self.persistentRefs = {} ## list of objects considered 'persistent' when the last diff() was called self.objTypes = {} ObjTracker.allObjs[id(self)] = None self.objs = [self.__dict__, self.startRefs, self.startCount, self.newRefs, self.persistentRefs, self.objTypes] self.objs.append(self.objs) for v in self.objs: ObjTracker.allObjs[id(v)] = None self.start() def findNew(self, regex): """Return all objects matching regex that were considered 'new' when the last diff() was run.""" return self.findTypes(self.newRefs, regex) def findPersistent(self, regex): """Return all objects matching regex that were considered 'persistent' when the last diff() was run.""" return self.findTypes(self.persistentRefs, regex) def start(self): """ Remember the current set of objects as the comparison for all future calls to diff() Called automatically on init, but can be called manually as well. """ refs, count, objs = self.collect() for r in self.startRefs: self.forgetRef(self.startRefs[r]) self.startRefs.clear() self.startRefs.update(refs) for r in refs: self.rememberRef(r) self.startCount.clear() self.startCount.update(count) #self.newRefs.clear() #self.newRefs.update(refs) def diff(self, **kargs): """ Compute all differences between the current object set and the reference set. Print a set of reports for created, deleted, and persistent objects """ refs, count, objs = self.collect() ## refs contains the list of ALL objects ## Which refs have disappeared since call to start() (these are only displayed once, then forgotten.) delRefs = {} for i in self.startRefs.keys(): if i not in refs: delRefs[i] = self.startRefs[i] del self.startRefs[i] self.forgetRef(delRefs[i]) for i in self.newRefs.keys(): if i not in refs: delRefs[i] = self.newRefs[i] del self.newRefs[i] self.forgetRef(delRefs[i]) #print "deleted:", len(delRefs) ## Which refs have appeared since call to start() or diff() persistentRefs = {} ## created since start(), but before last diff() createRefs = {} ## created since last diff() for o in refs: if o not in self.startRefs: if o not in self.newRefs: createRefs[o] = refs[o] ## object has been created since last diff() else: persistentRefs[o] = refs[o] ## object has been created since start(), but before last diff() (persistent) #print "new:", len(newRefs) ## self.newRefs holds the entire set of objects created since start() for r in self.newRefs: self.forgetRef(self.newRefs[r]) self.newRefs.clear() self.newRefs.update(persistentRefs) self.newRefs.update(createRefs) for r in self.newRefs: self.rememberRef(self.newRefs[r]) #print "created:", len(createRefs) ## self.persistentRefs holds all objects considered persistent. self.persistentRefs.clear() self.persistentRefs.update(persistentRefs) print("----------- Count changes since start: ----------") c1 = count.copy() for k in self.startCount: c1[k] = c1.get(k, 0) - self.startCount[k] typs = list(c1.keys()) typs.sort(lambda a,b: cmp(c1[a], c1[b])) for t in typs: if c1[t] == 0: continue num = "%d" % c1[t] print(" " + num + " "*(10-len(num)) + str(t)) print("----------- %d Deleted since last diff: ------------" % len(delRefs)) self.report(delRefs, objs, **kargs) print("----------- %d Created since last diff: ------------" % len(createRefs)) self.report(createRefs, objs, **kargs) print("----------- %d Created since start (persistent): ------------" % len(persistentRefs)) self.report(persistentRefs, objs, **kargs) def __del__(self): self.startRefs.clear() self.startCount.clear() self.newRefs.clear() self.persistentRefs.clear() del ObjTracker.allObjs[id(self)] for v in self.objs: del ObjTracker.allObjs[id(v)] @classmethod def isObjVar(cls, o): return type(o) is cls or id(o) in cls.allObjs def collect(self): print("Collecting list of all objects...") gc.collect() objs = get_all_objects() frame = sys._getframe() del objs[id(frame)] ## ignore the current frame del objs[id(frame.f_code)] ignoreTypes = [int] refs = {} count = {} for k in objs: o = objs[k] typ = type(o) oid = id(o) if ObjTracker.isObjVar(o) or typ in ignoreTypes: continue try: ref = weakref.ref(obj) except: ref = None refs[oid] = ref typ = type(o) typStr = typeStr(o) self.objTypes[oid] = typStr ObjTracker.allObjs[id(typStr)] = None count[typ] = count.get(typ, 0) + 1 print("All objects: %d Tracked objects: %d" % (len(objs), len(refs))) return refs, count, objs def forgetRef(self, ref): if ref is not None: del ObjTracker.allObjs[id(ref)] def rememberRef(self, ref): ## Record the address of the weakref object so it is not included in future object counts. if ref is not None: ObjTracker.allObjs[id(ref)] = None def lookup(self, oid, ref, objs=None): if ref is None or ref() is None: try: obj = lookup(oid, objects=objs) except: obj = None else: obj = ref() return obj def report(self, refs, allobjs=None, showIDs=False): if allobjs is None: allobjs = get_all_objects() count = {} rev = {} for oid in refs: obj = self.lookup(oid, refs[oid], allobjs) if obj is None: typ = "[del] " + self.objTypes[oid] else: typ = typeStr(obj) if typ not in rev: rev[typ] = [] rev[typ].append(oid) c = count.get(typ, [0,0]) count[typ] = [c[0]+1, c[1]+objectSize(obj)] typs = list(count.keys()) typs.sort(lambda a,b: cmp(count[a][1], count[b][1])) for t in typs: line = " %d\t%d\t%s" % (count[t][0], count[t][1], t) if showIDs: line += "\t"+",".join(map(str,rev[t])) print(line) def findTypes(self, refs, regex): allObjs = get_all_objects() ids = {} objs = [] r = re.compile(regex) for k in refs: if r.search(self.objTypes[k]): objs.append(self.lookup(k, refs[k], allObjs)) return objs def describeObj(obj, depth=4, path=None, ignore=None): """ Trace all reference paths backward, printing a list of different ways this object can be accessed. Attempts to answer the question "who has a reference to this object" """ if path is None: path = [obj] if ignore is None: ignore = {} ## holds IDs of objects used within the function. ignore[id(sys._getframe())] = None ignore[id(path)] = None gc.collect() refs = gc.get_referrers(obj) ignore[id(refs)] = None printed=False for ref in refs: if id(ref) in ignore: continue if id(ref) in list(map(id, path)): print("Cyclic reference: " + refPathString([ref]+path)) printed = True continue newPath = [ref]+path if len(newPath) >= depth: refStr = refPathString(newPath) if '[_]' not in refStr: ## ignore '_' references generated by the interactive shell print(refStr) printed = True else: describeObj(ref, depth, newPath, ignore) printed = True if not printed: print("Dead end: " + refPathString(path)) def typeStr(obj): """Create a more useful type string by making types report their class.""" typ = type(obj) if typ == types.InstanceType: return "" % obj.__class__.__name__ else: return str(typ) def searchRefs(obj, *args): """Pseudo-interactive function for tracing references backward. Arguments: obj: The initial object from which to start searching args: A set of string or int arguments. each integer selects one of obj's referrers to be the new 'obj' each string indicates an action to take on the current 'obj': t: print the types of obj's referrers l: print the lengths of obj's referrers (if they have __len__) i: print the IDs of obj's referrers o: print obj ro: return obj rr: return list of obj's referrers Examples: searchRefs(obj, 't') ## Print types of all objects referring to obj searchRefs(obj, 't', 0, 't') ## ..then select the first referrer and print the types of its referrers searchRefs(obj, 't', 0, 't', 'l') ## ..also print lengths of the last set of referrers searchRefs(obj, 0, 1, 'ro') ## Select index 0 from obj's referrer, then select index 1 from the next set of referrers, then return that object """ ignore = {id(sys._getframe()): None} gc.collect() refs = gc.get_referrers(obj) ignore[id(refs)] = None refs = [r for r in refs if id(r) not in ignore] for a in args: #fo = allFrameObjs() #refs = [r for r in refs if r not in fo] if type(a) is int: obj = refs[a] gc.collect() refs = gc.get_referrers(obj) ignore[id(refs)] = None refs = [r for r in refs if id(r) not in ignore] elif a == 't': print(list(map(typeStr, refs))) elif a == 'i': print(list(map(id, refs))) elif a == 'l': def slen(o): if hasattr(o, '__len__'): return len(o) else: return None print(list(map(slen, refs))) elif a == 'o': print(obj) elif a == 'ro': return obj elif a == 'rr': return refs def allFrameObjs(): """Return list of frame objects in current stack. Useful if you want to ignore these objects in refernece searches""" f = sys._getframe() objs = [] while f is not None: objs.append(f) objs.append(f.f_code) #objs.append(f.f_locals) #objs.append(f.f_globals) #objs.append(f.f_builtins) f = f.f_back return objs def findObj(regex): """Return a list of objects whose typeStr matches regex""" allObjs = get_all_objects() objs = [] r = re.compile(regex) for i in allObjs: obj = allObjs[i] if r.search(typeStr(obj)): objs.append(obj) return objs def listRedundantModules(): """List modules that have been imported more than once via different paths.""" mods = {} for name, mod in sys.modules.items(): if not hasattr(mod, '__file__'): continue mfile = os.path.abspath(mod.__file__) if mfile[-1] == 'c': mfile = mfile[:-1] if mfile in mods: print("module at %s has 2 names: %s, %s" % (mfile, name, mods[mfile])) else: mods[mfile] = name def walkQObjectTree(obj, counts=None, verbose=False, depth=0): """ Walk through a tree of QObjects, doing nothing to them. The purpose of this function is to find dead objects and generate a crash immediately rather than stumbling upon them later. Prints a count of the objects encountered, for fun. (or is it?) """ if verbose: print(" "*depth + typeStr(obj)) report = False if counts is None: counts = {} report = True typ = str(type(obj)) try: counts[typ] += 1 except KeyError: counts[typ] = 1 for child in obj.children(): walkQObjectTree(child, counts, verbose, depth+1) return counts QObjCache = {} def qObjectReport(verbose=False): """Generate a report counting all QObjects and their types""" global qObjCache count = {} for obj in findObj('PyQt'): if isinstance(obj, QtCore.QObject): oid = id(obj) if oid not in QObjCache: QObjCache[oid] = typeStr(obj) + " " + obj.objectName() try: QObjCache[oid] += " " + obj.parent().objectName() QObjCache[oid] += " " + obj.text() except: pass print("check obj", oid, str(QObjCache[oid])) if obj.parent() is None: walkQObjectTree(obj, count, verbose) typs = list(count.keys()) typs.sort() for t in typs: print(count[t], "\t", t) class PrintDetector(object): def __init__(self): self.stdout = sys.stdout sys.stdout = self def remove(self): sys.stdout = self.stdout def __del__(self): self.remove() def write(self, x): self.stdout.write(x) traceback.print_stack() def flush(self): self.stdout.flush()