1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
|
from rpython.rtyper.lltypesystem.lltype import DelayedPointer
from rpython.tool.algo.unionfind import UnionFind
class GraphAnalyzer(object):
verbose = False
def __init__(self, translator):
self.translator = translator
self._analyzed_calls = UnionFind(lambda graph: Dependency(self))
# method overridden by subclasses
@staticmethod
def bottom_result():
raise NotImplementedError("abstract base class")
@staticmethod
def top_result():
raise NotImplementedError("abstract base class")
@staticmethod
def is_top_result(result):
# only an optimization, safe to always return False
return False
@staticmethod
def result_builder():
raise NotImplementedError("abstract base class")
@staticmethod
def add_to_result(result, other):
raise NotImplementedError("abstract base class")
@staticmethod
def finalize_builder(result):
raise NotImplementedError("abstract base class")
@staticmethod
def join_two_results(result1, result2):
raise NotImplementedError("abstract base class")
def analyze_simple_operation(self, op, graphinfo=None):
raise NotImplementedError("abstract base class")
# some sensible default methods, can also be overridden
def analyze_exceptblock(self, block, seen=None):
return self.bottom_result()
def analyze_exceptblock_in_graph(self, graph, block, seen=None):
return self.analyze_exceptblock(block, seen)
def analyze_startblock(self, block, seen=None):
return self.bottom_result()
def analyze_external_call(self, funcobj, seen=None):
result = self.bottom_result()
if hasattr(funcobj, '_callbacks'):
bk = self.translator.annotator.bookkeeper
for function in funcobj._callbacks.callbacks:
desc = bk.getdesc(function)
for graph in desc.getgraphs():
result = self.join_two_results(
result, self.analyze_direct_call(graph, seen))
return result
def analyze_link(self, graph, link):
return self.bottom_result()
# general methods
def compute_graph_info(self, graph):
return None
def analyze(self, op, seen=None, graphinfo=None):
if op.opname == "direct_call":
try:
funcobj = op.args[0].value._obj
except DelayedPointer:
return self.top_result()
if funcobj is None:
# We encountered a null pointer. Calling it will crash.
# However, the call could be on a dead path, so we return the
# bottom result here.
return self.bottom_result()
if getattr(funcobj, 'external', None) is not None:
x = self.analyze_external_call(funcobj, seen)
if self.verbose and x:
self.dump_info('analyze_external_call %s: %r' % (op, x))
return x
try:
graph = funcobj.graph
except AttributeError:
return self.top_result()
x = self.analyze_direct_call(graph, seen)
if self.verbose and x:
self.dump_info('analyze_direct_call(%s): %r' % (graph, x))
return x
elif op.opname == "indirect_call":
graphs = op.args[-1].value
if graphs is None:
if self.verbose:
self.dump_info('%s to unknown' % (op,))
return self.top_result()
x = self.analyze_indirect_call(graphs, seen)
if self.verbose and x:
self.dump_info('analyze_indirect_call(%s): %r' % (graphs, x))
return x
x = self.analyze_simple_operation(op, graphinfo)
if self.verbose and x:
self.dump_info('%s: %r' % (op, x))
return x
def dump_info(self, info):
print '[%s] %s' % (self.__class__.__name__, info)
def analyze_direct_call(self, graph, seen=None):
if seen is None:
seen = DependencyTracker(self)
if not seen.enter(graph):
return seen.get_cached_result(graph)
result = self.result_builder()
graphinfo = self.compute_graph_info(graph)
for block in graph.iterblocks():
if block is graph.startblock:
result = self.add_to_result(
result,
self.analyze_startblock(block, seen)
)
elif block is graph.exceptblock:
result = self.add_to_result(
result,
self.analyze_exceptblock_in_graph(graph, block, seen)
)
if not self.is_top_result(result):
for op in block.operations:
result = self.add_to_result(
result,
self.analyze(op, seen, graphinfo)
)
if self.is_top_result(result):
break
if not self.is_top_result(result):
for exit in block.exits:
result = self.add_to_result(
result,
self.analyze_link(exit, seen)
)
if self.is_top_result(result):
break
if self.is_top_result(result):
break
result = self.finalize_builder(result)
seen.leave_with(result)
return result
def analyze_indirect_call(self, graphs, seen=None):
result = self.result_builder()
for graph in graphs:
result = self.add_to_result(
result,
self.analyze_direct_call(graph, seen)
)
if self.is_top_result(result):
break
return self.finalize_builder(result)
def analyze_all(self, graphs=None):
if graphs is None:
graphs = self.translator.graphs
for graph in graphs:
for block, op in graph.iterblockops():
self.analyze(op)
class Dependency(object):
def __init__(self, analyzer):
self._analyzer = analyzer
self._result = analyzer.bottom_result()
def merge_with_result(self, result):
self._result = self._analyzer.join_two_results(self._result, result)
def absorb(self, other):
self.merge_with_result(other._result)
class DependencyTracker(object):
"""This tracks the analysis of cyclic call graphs."""
# The point is that GraphAnalyzer works fine if the question we ask
# is about a single graph; but in the case of recursion, it will
# fail if we ask about multiple graphs. The purpose of this
# class is to fix the cache in GraphAnalyzer._analyzed_calls after
# each round, whenever a new set of graphs have been added to it.
# It works by assuming that the following is correct: for any set of
# graphs that can all (indirectly) call each other, all these graphs
# will get the same answer that is the 'join_two_results' of all of
# them.
def __init__(self, analyzer):
self.analyzer = analyzer
# the UnionFind object, which works like a mapping {graph: Dependency}
# (shared with GraphAnalyzer._analyzed_calls)
self.graph_results = analyzer._analyzed_calls
# the current stack of graphs being analyzed
self.current_stack = []
#self.current_stack_set = set()
def enter(self, graph):
if graph not in self.graph_results:
self.current_stack.append(graph)
#self.current_stack_set.add(graph)
self.graph_results.find(graph)
return True
else:
graph = self.graph_results.find_rep(graph)
for j in range(len(self.current_stack)):
othergraph = self.graph_results.find_rep(self.current_stack[j])
if graph is othergraph:
# found a cycle; merge all graphs in that cycle
for i in range(j, len(self.current_stack)):
self.graph_results.union(self.current_stack[i], graph)
# done
break
return False
def leave_with(self, result):
graph = self.current_stack.pop()
#self.current_stack_set.remove(graph)
dep = self.graph_results[graph]
dep.merge_with_result(result)
def get_cached_result(self, graph):
dep = self.graph_results[graph]
return dep._result
class BoolGraphAnalyzer(GraphAnalyzer):
"""generic way to analyze graphs: recursively follow it until the first
operation is found on which self.analyze_simple_operation returns True"""
def bottom_result(self):
return False
def top_result(self):
return True
def is_top_result(self, result):
return result
def result_builder(self):
return False
def add_to_result(self, result, other):
return self.join_two_results(result, other)
def finalize_builder(self, result):
return result
def join_two_results(self, result1, result2):
return result1 or result2
|
