import lldb import lldb.formatters.Logger # libcxx STL formatters for LLDB # These formatters are based upon the implementation of libc++ that # ships with current releases of OS X - They will not work for other implementations # of the standard C++ library - and they are bound to use the libc++-specific namespace # the std::string summary is just an example for your convenience # the actual summary that LLDB uses is C++ code inside the debugger's own core # this could probably be made more efficient but since it only reads a handful of bytes at a time # we probably don't need to worry too much about this for the time being def make_string(F,L): strval = '' G = F.GetData().uint8 for X in range(L): V = G[X] if V == 0: break strval = strval + chr(V % 256) return '"' + strval + '"' # if we ever care about big-endian, these two functions might need to change def is_short_string(value): return True if (value & 1) == 0 else False def extract_short_size(value): return ((value >> 1) % 256) # some of the members of libc++ std::string are anonymous or have internal names that convey # no external significance - we access them by index since this saves a name lookup that would add # no information for readers of the code, but when possible try to use meaningful variable names def stdstring_SummaryProvider(valobj,dict): logger = lldb.formatters.Logger.Logger() r = valobj.GetChildAtIndex(0) B = r.GetChildAtIndex(0) first = B.GetChildAtIndex(0) D = first.GetChildAtIndex(0) l = D.GetChildAtIndex(0) s = D.GetChildAtIndex(1) D20 = s.GetChildAtIndex(0) size_mode = D20.GetChildAtIndex(0).GetValueAsUnsigned(0) if is_short_string(size_mode): size = extract_short_size(size_mode) return make_string(s.GetChildAtIndex(1),size) else: data_ptr = l.GetChildAtIndex(2) size_vo = l.GetChildAtIndex(1) size = size_vo.GetValueAsUnsigned(0)+1 # the NULL terminator must be accounted for if size <= 1 or size == None: # should never be the case return '""' try: data = data_ptr.GetPointeeData(0,size) except: return '""' error = lldb.SBError() strval = data.GetString(error,0) if error.Fail(): return '' else: return '"' + strval + '"' class stdvector_SynthProvider: def __init__(self, valobj, dict): logger = lldb.formatters.Logger.Logger() self.valobj = valobj; def num_children(self): logger = lldb.formatters.Logger.Logger() try: start_val = self.start.GetValueAsUnsigned(0) finish_val = self.finish.GetValueAsUnsigned(0) # Before a vector has been constructed, it will contain bad values # so we really need to be careful about the length we return since # unitialized data can cause us to return a huge number. We need # to also check for any of the start, finish or end of storage values # being zero (NULL). If any are, then this vector has not been # initialized yet and we should return zero # Make sure nothing is NULL if start_val == 0 or finish_val == 0: return 0 # Make sure start is less than finish if start_val >= finish_val: return 0 num_children = (finish_val-start_val) if (num_children % self.data_size) != 0: return 0 else: num_children = num_children/self.data_size return num_children except: return 0; def get_child_index(self,name): logger = lldb.formatters.Logger.Logger() try: return int(name.lstrip('[').rstrip(']')) except: return -1 def get_child_at_index(self,index): logger = lldb.formatters.Logger.Logger() logger >> "Retrieving child " + str(index) if index < 0: return None; if index >= self.num_children(): return None; try: offset = index * self.data_size return self.start.CreateChildAtOffset('['+str(index)+']',offset,self.data_type) except: return None def update(self): logger = lldb.formatters.Logger.Logger() try: self.start = self.valobj.GetChildMemberWithName('__begin_') self.finish = self.valobj.GetChildMemberWithName('__end_') # the purpose of this field is unclear, but it is the only field whose type is clearly T* for a vector # if this ends up not being correct, we can use the APIs to get at template arguments data_type_finder = self.valobj.GetChildMemberWithName('__end_cap_').GetChildMemberWithName('__first_') self.data_type = data_type_finder.GetType().GetPointeeType() self.data_size = self.data_type.GetByteSize() except: pass def has_children(self): return True # Just an example: the actual summary is produced by a summary string: size=${svar%#} def stdvector_SummaryProvider(valobj,dict): prov = stdvector_SynthProvider(valobj,None) return 'size=' + str(prov.num_children()) class stdlist_entry: def __init__(self,entry): logger = lldb.formatters.Logger.Logger() self.entry = entry def _next_impl(self): logger = lldb.formatters.Logger.Logger() return stdlist_entry(self.entry.GetChildMemberWithName('__next_')) def _prev_impl(self): logger = lldb.formatters.Logger.Logger() return stdlist_entry(self.entry.GetChildMemberWithName('__prev_')) def _value_impl(self): logger = lldb.formatters.Logger.Logger() return self.entry.GetValueAsUnsigned(0) def _isnull_impl(self): logger = lldb.formatters.Logger.Logger() return self._value_impl() == 0 def _sbvalue_impl(self): logger = lldb.formatters.Logger.Logger() return self.entry next = property(_next_impl,None) value = property(_value_impl,None) is_null = property(_isnull_impl,None) sbvalue = property(_sbvalue_impl,None) class stdlist_iterator: def increment_node(self,node): logger = lldb.formatters.Logger.Logger() if node.is_null: return None return node.next def __init__(self,node): logger = lldb.formatters.Logger.Logger() self.node = stdlist_entry(node) # we convert the SBValue to an internal node object on entry def value(self): logger = lldb.formatters.Logger.Logger() return self.node.sbvalue # and return the SBValue back on exit def next(self): logger = lldb.formatters.Logger.Logger() node = self.increment_node(self.node) if node != None and node.sbvalue.IsValid() and not(node.is_null): self.node = node return self.value() else: return None def advance(self,N): logger = lldb.formatters.Logger.Logger() if N < 0: return None if N == 0: return self.value() if N == 1: return self.next() while N > 0: self.next() N = N - 1 return self.value() class stdlist_SynthProvider: def __init__(self, valobj, dict): logger = lldb.formatters.Logger.Logger() self.valobj = valobj self.count = None def next_node(self,node): logger = lldb.formatters.Logger.Logger() return node.GetChildMemberWithName('__next_') def value(self,node): logger = lldb.formatters.Logger.Logger() return node.GetValueAsUnsigned() # Floyd's cyle-finding algorithm # try to detect if this list has a loop def has_loop(self): global _list_uses_loop_detector logger = lldb.formatters.Logger.Logger() if _list_uses_loop_detector == False: logger >> "Asked not to use loop detection" return False slow = stdlist_entry(self.head) fast1 = stdlist_entry(self.head) fast2 = stdlist_entry(self.head) while slow.next.value != self.node_address: slow_value = slow.value fast1 = fast2.next fast2 = fast1.next if fast1.value == slow_value or fast2.value == slow_value: return True slow = slow.next return False def num_children(self): global _list_capping_size logger = lldb.formatters.Logger.Logger() if self.count == None: self.count = self.num_children_impl() if self.count > _list_capping_size: self.count = _list_capping_size return self.count def num_children_impl(self): global _list_capping_size logger = lldb.formatters.Logger.Logger() try: next_val = self.head.GetValueAsUnsigned(0) prev_val = self.tail.GetValueAsUnsigned(0) # After a std::list has been initialized, both next and prev will be non-NULL if next_val == 0 or prev_val == 0: return 0 if next_val == self.node_address: return 0 if next_val == prev_val: return 1 if self.has_loop(): return 0 size = 2 current = stdlist_entry(self.head) while current.next.value != self.node_address: size = size + 1 current = current.next if size > _list_capping_size: return _list_capping_size return (size - 1) except: return 0; def get_child_index(self,name): logger = lldb.formatters.Logger.Logger() try: return int(name.lstrip('[').rstrip(']')) except: return -1 def get_child_at_index(self,index): logger = lldb.formatters.Logger.Logger() logger >> "Fetching child " + str(index) if index < 0: return None; if index >= self.num_children(): return None; try: current = stdlist_iterator(self.head) current = current.advance(index) # we do not return __value_ because then all our children would be named __value_ # we need to make a copy of __value__ with the right name - unfortunate obj = current.GetChildMemberWithName('__value_') obj_data = obj.GetData() return self.valobj.CreateValueFromData('[' + str(index) + ']',obj_data,self.data_type) except: return None def extract_type(self): logger = lldb.formatters.Logger.Logger() list_type = self.valobj.GetType().GetUnqualifiedType() if list_type.IsReferenceType(): list_type = list_type.GetDereferencedType() if list_type.GetNumberOfTemplateArguments() > 0: data_type = list_type.GetTemplateArgumentType(0) else: data_type = None return data_type def update(self): logger = lldb.formatters.Logger.Logger() self.count = None try: impl = self.valobj.GetChildMemberWithName('__end_') self.node_address = self.valobj.AddressOf().GetValueAsUnsigned(0) self.head = impl.GetChildMemberWithName('__next_') self.tail = impl.GetChildMemberWithName('__prev_') self.data_type = self.extract_type() self.data_size = self.data_type.GetByteSize() except: pass def has_children(self): return True # Just an example: the actual summary is produced by a summary string: size=${svar%#} def stdlist_SummaryProvider(valobj,dict): prov = stdlist_SynthProvider(valobj,None) return 'size=' + str(prov.num_children()) # a tree node - this class makes the syntax in the actual iterator nicer to read and maintain class stdmap_iterator_node: def _left_impl(self): logger = lldb.formatters.Logger.Logger() return stdmap_iterator_node(self.node.GetChildMemberWithName("__left_")) def _right_impl(self): logger = lldb.formatters.Logger.Logger() return stdmap_iterator_node(self.node.GetChildMemberWithName("__right_")) def _parent_impl(self): logger = lldb.formatters.Logger.Logger() return stdmap_iterator_node(self.node.GetChildMemberWithName("__parent_")) def _value_impl(self): logger = lldb.formatters.Logger.Logger() return self.node.GetValueAsUnsigned(0) def _sbvalue_impl(self): logger = lldb.formatters.Logger.Logger() return self.node def _null_impl(self): logger = lldb.formatters.Logger.Logger() return self.value == 0 def __init__(self,node): logger = lldb.formatters.Logger.Logger() self.node = node left = property(_left_impl,None) right = property(_right_impl,None) parent = property(_parent_impl,None) value = property(_value_impl,None) is_null = property(_null_impl,None) sbvalue = property(_sbvalue_impl,None) # a Python implementation of the tree iterator used by libc++ class stdmap_iterator: def tree_min(self,x): logger = lldb.formatters.Logger.Logger() steps = 0 if x.is_null: return None while (not x.left.is_null): x = x.left steps += 1 if steps > self.max_count: logger >> "Returning None - we overflowed" return None return x def tree_max(self,x): logger = lldb.formatters.Logger.Logger() if x.is_null: return None while (not x.right.is_null): x = x.right return x def tree_is_left_child(self,x): logger = lldb.formatters.Logger.Logger() if x.is_null: return None return True if x.value == x.parent.left.value else False def increment_node(self,node): logger = lldb.formatters.Logger.Logger() if node.is_null: return None if not node.right.is_null: return self.tree_min(node.right) steps = 0 while (not self.tree_is_left_child(node)): steps += 1 if steps > self.max_count: logger >> "Returning None - we overflowed" return None node = node.parent return node.parent def __init__(self,node,max_count=0): logger = lldb.formatters.Logger.Logger() self.node = stdmap_iterator_node(node) # we convert the SBValue to an internal node object on entry self.max_count = max_count def value(self): logger = lldb.formatters.Logger.Logger() return self.node.sbvalue # and return the SBValue back on exit def next(self): logger = lldb.formatters.Logger.Logger() node = self.increment_node(self.node) if node != None and node.sbvalue.IsValid() and not(node.is_null): self.node = node return self.value() else: return None def advance(self,N): logger = lldb.formatters.Logger.Logger() if N < 0: return None if N == 0: return self.value() if N == 1: return self.next() while N > 0: if self.next() == None: return None N = N - 1 return self.value() class stdmap_SynthProvider: def __init__(self, valobj, dict): logger = lldb.formatters.Logger.Logger() self.valobj = valobj; self.pointer_size = self.valobj.GetProcess().GetAddressByteSize() self.count = None def update(self): logger = lldb.formatters.Logger.Logger() self.count = None try: # we will set this to True if we find out that discovering a node in the map takes more steps than the overall size of the RB tree # if this gets set to True, then we will merrily return None for any child from that moment on self.garbage = False self.tree = self.valobj.GetChildMemberWithName('__tree_') self.root_node = self.tree.GetChildMemberWithName('__begin_node_') # this data is either lazily-calculated, or cannot be inferred at this moment # we still need to mark it as None, meaning "please set me ASAP" self.data_type = None self.data_size = None self.skip_size = None except: pass def num_children(self): global _map_capping_size logger = lldb.formatters.Logger.Logger() if self.count == None: self.count = self.num_children_impl() if self.count > _map_capping_size: self.count = _map_capping_size return self.count def num_children_impl(self): logger = lldb.formatters.Logger.Logger() try: return self.valobj.GetChildMemberWithName('__tree_').GetChildMemberWithName('__pair3_').GetChildMemberWithName('__first_').GetValueAsUnsigned() except: return 0; def has_children(self): return True def get_data_type(self): logger = lldb.formatters.Logger.Logger() if self.data_type == None or self.data_size == None: if self.num_children() == 0: return False deref = self.root_node.Dereference() if not(deref.IsValid()): return False value = deref.GetChildMemberWithName('__value_') if not(value.IsValid()): return False self.data_type = value.GetType() self.data_size = self.data_type.GetByteSize() self.skip_size = None return True else: return True def get_value_offset(self,node): logger = lldb.formatters.Logger.Logger() if self.skip_size == None: node_type = node.GetType() fields_count = node_type.GetNumberOfFields() for i in range(fields_count): field = node_type.GetFieldAtIndex(i) if field.GetName() == '__value_': self.skip_size = field.GetOffsetInBytes() break return (self.skip_size != None) def get_child_index(self,name): logger = lldb.formatters.Logger.Logger() try: return int(name.lstrip('[').rstrip(']')) except: return -1 def get_child_at_index(self,index): logger = lldb.formatters.Logger.Logger() logger >> "Retrieving child " + str(index) if index < 0: return None if index >= self.num_children(): return None; if self.garbage: logger >> "Returning None since this tree is garbage" return None try: iterator = stdmap_iterator(self.root_node,max_count=self.num_children()) # the debug info for libc++ std::map is such that __begin_node_ has a very nice and useful type # out of which we can grab the information we need - every other node has a less informative # type which omits all value information and only contains housekeeping information for the RB tree # hence, we need to know if we are at a node != 0, so that we can still get at the data need_to_skip = (index > 0) current = iterator.advance(index) if current == None: logger >> "Tree is garbage - returning None" self.garbage = True return None if self.get_data_type(): if not(need_to_skip): current = current.Dereference() obj = current.GetChildMemberWithName('__value_') obj_data = obj.GetData() self.get_value_offset(current) # make sure we have a valid offset for the next items # we do not return __value_ because then we would end up with a child named # __value_ instead of [0] return self.valobj.CreateValueFromData('[' + str(index) + ']',obj_data,self.data_type) else: # FIXME we need to have accessed item 0 before accessing any other item! if self.skip_size == None: logger >> "You asked for item > 0 before asking for item == 0, I will fetch 0 now then retry" if self.get_child_at_index(0): return self.get_child_at_index(index) else: logger >> "item == 0 could not be found. sorry, nothing can be done here." return None return current.CreateChildAtOffset('[' + str(index) + ']',self.skip_size,self.data_type) else: logger >> "Unable to infer data-type - returning None (should mark tree as garbage here?)" return None except Exception as err: logger >> "Hit an exception: " + str(err) return None # Just an example: the actual summary is produced by a summary string: size=${svar%#} def stdmap_SummaryProvider(valobj,dict): prov = stdmap_SynthProvider(valobj,None) return 'size=' + str(prov.num_children()) class stddeque_SynthProvider: def __init__(self, valobj, d): logger = lldb.formatters.Logger.Logger() logger.write("init") self.valobj = valobj self.pointer_size = self.valobj.GetProcess().GetAddressByteSize() self.count = None try: self.find_block_size() except: self.block_size = -1 self.element_size = -1 logger.write("block_size=%d, element_size=%d" % (self.block_size, self.element_size)) def find_block_size(self): # in order to use the deque we must have the block size, or else # it's impossible to know what memory addresses are valid self.element_type = self.valobj.GetType().GetTemplateArgumentType(0) self.element_size = self.element_type.GetByteSize() # The code says this, but there must be a better way: # template # class __deque_base { # static const difference_type __block_size = sizeof(value_type) < 256 ? 4096 / sizeof(value_type) : 16; # } if self.element_size < 256: self.block_size = 4096 / self.element_size else: self.block_size = 16 def num_children(self): global _deque_capping_size logger = lldb.formatters.Logger.Logger() if self.count is None: return 0 return min(self.count, _deque_capping_size) def has_children(self): return True def get_child_index(self,name): logger = lldb.formatters.Logger.Logger() try: return int(name.lstrip('[').rstrip(']')) except: return -1 def get_child_at_index(self,index): logger = lldb.formatters.Logger.Logger() logger.write("Fetching child " + str(index)) if index < 0 or self.count is None: return None; if index >= self.num_children(): return None; try: i, j = divmod(self.start+index, self.block_size) return self.first.CreateValueFromExpression('[' + str(index) + ']', '*(*(%s + %d) + %d)' % (self.first.get_expr_path(), i, j)) except: return None def update(self): logger = lldb.formatters.Logger.Logger() try: # A deque is effectively a two-dim array, with fixed width. # 'map' contains pointers to the rows of this array. The # full memory area allocated by the deque is delimited # by 'first' and 'end_cap'. However, only a subset of this # memory contains valid data since a deque may have some slack # at the front and back in order to have O(1) insertion at # both ends. The rows in active use are delimited by # 'begin' and 'end'. # # To find the elements that are actually constructed, the 'start' # variable tells which element in this NxM array is the 0th # one, and the 'size' element gives the number of elements # in the deque. count = self.valobj.GetChildMemberWithName('__size_').GetChildMemberWithName('__first_').GetValueAsUnsigned(0) # give up now if we cant access memory reliably if self.block_size < 0: logger.write("block_size < 0") return map_ = self.valobj.GetChildMemberWithName('__map_') start = self.valobj.GetChildMemberWithName('__start_').GetValueAsUnsigned(0) first = map_.GetChildMemberWithName('__first_') map_first = first.GetValueAsUnsigned(0) map_begin = map_.GetChildMemberWithName('__begin_').GetValueAsUnsigned(0) map_end = map_.GetChildMemberWithName('__end_').GetValueAsUnsigned(0) map_endcap= map_.GetChildMemberWithName('__end_cap_').GetChildMemberWithName('__first_').GetValueAsUnsigned(0) # check consistency if not map_first <= map_begin <= map_end <= map_endcap: logger.write("map pointers are not monotonic") return total_rows, junk = divmod(map_endcap - map_first, self.pointer_size) if junk: logger.write("endcap-first doesnt align correctly") return active_rows, junk = divmod(map_end - map_begin, self.pointer_size) if junk: logger.write("end-begin doesnt align correctly") return start_row, junk = divmod(map_begin - map_first, self.pointer_size) if junk: logger.write("begin-first doesnt align correctly") return if not start_row*self.block_size <= start < (start_row+1)*self.block_size: logger.write("0th element must be in the 'begin' row") return end_row = start_row + active_rows if not count: if active_rows: logger.write("empty deque but begin!=end") return elif not (end_row-1)*self.block_size <= start+count < end_row*self.block_size: logger.write("nth element must be before the 'end' row") return logger.write("update success: count=%r, start=%r, first=%r" % (count,start,first)) # if consistent, save all we really need: self.count = count self.start = start self.first = first except: self.count = None self.start = None self.map_first = None self.map_begin = None class stdsharedptr_SynthProvider: def __init__(self, valobj, d): logger = lldb.formatters.Logger.Logger() logger.write("init") self.valobj = valobj #self.element_ptr_type = self.valobj.GetType().GetTemplateArgumentType(0).GetPointerType() self.ptr = None self.cntrl = None process = valobj.GetProcess() self.endianness = process.GetByteOrder() self.pointer_size = process.GetAddressByteSize() self.count_type = valobj.GetType().GetBasicType(lldb.eBasicTypeUnsignedLong) def num_children(self): return 1 def has_children(self): return True def get_child_index(self,name): if name=="__ptr_": return 0 if name=="count": return 1 if name=="weak_count": return 2 return -1 def get_child_at_index(self,index): if index == 0: return self.ptr if index == 1: if self.cntrl == None: count = 0 else: count = 1 + self.cntrl.GetChildMemberWithName('__shared_owners_').GetValueAsSigned() return self.valobj.CreateValueFromData("count", lldb.SBData.CreateDataFromUInt64Array(self.endianness, self.pointer_size, [count]), self.count_type) if index == 2: if self.cntrl == None: count = 0 else: count = 1 + self.cntrl.GetChildMemberWithName('__shared_weak_owners_').GetValueAsSigned() return self.valobj.CreateValueFromData("weak_count", lldb.SBData.CreateDataFromUInt64Array(self.endianness, self.pointer_size, [count]), self.count_type) return None def update(self): logger = lldb.formatters.Logger.Logger() self.ptr = self.valobj.GetChildMemberWithName('__ptr_')#.Cast(self.element_ptr_type) cntrl = self.valobj.GetChildMemberWithName('__cntrl_') if cntrl.GetValueAsUnsigned(0): self.cntrl = cntrl.Dereference() else: self.cntrl = None # we can use two different categories for old and new formatters - type names are different enough that we should make no confusion # talking with libc++ developer: "std::__1::class_name is set in stone until we decide to change the ABI. That shouldn't happen within a 5 year time frame" def __lldb_init_module(debugger,dict): debugger.HandleCommand('type summary add -F libcxx.stdstring_SummaryProvider "std::__1::string" -w libcxx') debugger.HandleCommand('type summary add -F libcxx.stdstring_SummaryProvider "std::__1::basic_string, class std::__1::allocator >" -w libcxx') debugger.HandleCommand('type synthetic add -l libcxx.stdvector_SynthProvider -x "^(std::__1::)vector<.+>$" -w libcxx') debugger.HandleCommand('type summary add -F libcxx.stdvector_SummaryProvider -e -x "^(std::__1::)vector<.+>$" -w libcxx') debugger.HandleCommand('type synthetic add -l libcxx.stdlist_SynthProvider -x "^(std::__1::)list<.+>$" -w libcxx') debugger.HandleCommand('type summary add -F libcxx.stdlist_SummaryProvider -e -x "^(std::__1::)list<.+>$" -w libcxx') debugger.HandleCommand('type synthetic add -l libcxx.stdmap_SynthProvider -x "^(std::__1::)map<.+> >$" -w libcxx') debugger.HandleCommand('type summary add -F libcxx.stdmap_SummaryProvider -e -x "^(std::__1::)map<.+> >$" -w libcxx') debugger.HandleCommand("type category enable libcxx") debugger.HandleCommand('type synthetic add -l libcxx.stddeque_SynthProvider -x "^(std::__1::)deque<.+>$" -w libcxx') debugger.HandleCommand('type synthetic add -l libcxx.stdsharedptr_SynthProvider -x "^(std::__1::)shared_ptr<.+>$" -w libcxx') # turns out the structs look the same, so weak_ptr can be handled the same! debugger.HandleCommand('type synthetic add -l libcxx.stdsharedptr_SynthProvider -x "^(std::__1::)weak_ptr<.+>$" -w libcxx') _map_capping_size = 255 _list_capping_size = 255 _list_uses_loop_detector = True _deque_capping_size = 255