[ASan] Allow the users of SymbolizationLoop to make use of the -dsym_hint option in llvm-symbolizer

[glider]

Summary

Let the users of SymbolizationLoop define a function that produces the list of .dSYM hints (possible path to the .dSYM bundle) for the given binary.
Because the hints can't be added to an existing llvm-symbolizer process, we spawn a new symbolizer process ones each time a new hint appears.
Those can only appear for binaries that we haven't seen before.

[glider]

Adding llvm-commits

[earthdok]

projects/compiler-rt/lib/asan/scripts/asan_symbolize.py

73

how about "os"?

86

for consistency this should be "--dsym-hint"

357

I think those comments should be moved into the code.

361

Needs a comment on why we don't want to just use the last symbolizer always.

363

...which has the full set of hints.

370

"if dsym_hint and (dsym_hint not in self.dsym_hints)" would be more readable I think

[glider]

Addressed the comments.
Please note that this revision has been abandoned before you've started adding the comments.
I've uploaded a new one: http://reviews.llvm.org/D6310

projects/compiler-rt/lib/asan/scripts/asan_symbolize.py

73

The same property of SymbolizationLoop is already called "system"

86

Done.

357

Not sure they'll become easier to follow then.

361

Done.

363

Done.

370

Done. However this code has changed a bit in http://reviews.llvm.org/D6310

projects/compiler-rt/lib/asan/scripts/asan_symbolize.py

Returns:
         list of strings (one string for each inlined frame) describing
         the code locations for this instruction (that is, function name, file
         name, line and column numbers).
     """
     return None
 
 
 class LLVMSymbolizer(Symbolizer):
-  def __init__(self, symbolizer_path, addr):
+  def __init__(self, symbolizer_path, default_arch, system, dsym_hints=[]):
     super(LLVMSymbolizer, self).__init__()
     self.symbolizer_path = symbolizer_path
-    self.default_arch = guess_arch(addr)
+    self.default_arch = default_arch
+    self.system = system
+    self.dsym_hints = dsym_hints
     self.pipe = self.open_llvm_symbolizer()
 
   def open_llvm_symbolizer(self):
     cmd = [self.symbolizer_path,
            '--use-symbol-table=true',
            '--demangle=%s' % demangle,
            '--functions=short',
            '--inlining=true',
            '--default-arch=%s' % self.default_arch]
+    if self.system == 'Darwin':
+      for hint in self.dsym_hints:
+        cmd.append('-dsym-hint=%s' % hint)
     if DEBUG:
       print ' '.join(cmd)
     try:
       result = subprocess.Popen(cmd, stdin=subprocess.PIPE,
                                 stdout=subprocess.PIPE)
     except OSError:
       result = None
     return result
 
   def symbolize(self, addr, binary, offset):
     """Overrides Symbolizer.symbolize."""
     if not self.pipe:
       return None
     result = []
     try:
-      symbolizer_input = '%s %s' % (binary, offset)
+      symbolizer_input = '"%s" %s' % (binary, offset)
       if DEBUG:
         print symbolizer_input
       print >> self.pipe.stdin, symbolizer_input
       while True:
         function_name = self.pipe.stdout.readline().rstrip()
         if not function_name:
           break
         file_name = self.pipe.stdout.readline().rstrip()
         file_name = fix_filename(file_name)
         if (not function_name.startswith('??') or
             not file_name.startswith('??')):
           # Append only non-trivial frames.
           result.append('%s in %s %s' % (addr, function_name,
                                          file_name))
     except Exception:
       result = []
     if not result:
       result = None
     return result
 
 
-def LLVMSymbolizerFactory(system, addr):
+def LLVMSymbolizerFactory(system, default_arch, dsym_hints=[]):
   symbolizer_path = os.getenv('LLVM_SYMBOLIZER_PATH')
   if not symbolizer_path:
     symbolizer_path = os.getenv('ASAN_SYMBOLIZER_PATH')
     if not symbolizer_path:
       # Assume llvm-symbolizer is in PATH.
       symbolizer_path = 'llvm-symbolizer'
-  return LLVMSymbolizer(symbolizer_path, addr)
+  return LLVMSymbolizer(symbolizer_path, default_arch, system, dsym_hints)
 
 
 class Addr2LineSymbolizer(Symbolizer):
   def __init__(self, binary):
     super(Addr2LineSymbolizer, self).__init__()
     self.binary = binary
     self.pipe = self.open_addr2line()
 
if res:
           addr, function_name, file_name, line_no)]
       print result
       return result
     else:
       return None
 
 
 class SymbolizationLoop(object):
-  def __init__(self, binary_name_filter=None):
+  def __init__(self, binary_name_filter=None, dsym_hint_producer=None):
     # Used by clients who may want to supply a different binary name.
     # E.g. in Chrome several binaries may share a single .dSYM.
     self.binary_name_filter = binary_name_filter
+    self.dsym_hint_producer = dsym_hint_producer
     self.system = os.uname()[0]
     if self.system not in ['Linux', 'Darwin', 'FreeBSD']:
       raise Exception('Unknown system')
-    self.llvm_symbolizer = None
+    self.llvm_symbolizers = {}
+    self.last_llvm_symbolizer = None
+    self.dsym_hints = set([])
     self.frame_no = 0
 
   def symbolize_address(self, addr, binary, offset):
-    # Initialize llvm-symbolizer lazily.
-    if not self.llvm_symbolizer:
-      self.llvm_symbolizer = LLVMSymbolizerFactory(self.system, addr)
+    # On non-Darwin (i.e. on platforms without .dSYM debug info) always use
+    # a single symbolizer binary.
+    # On Darwin, if the dsym hint producer is present:
+    #  1. check whether we've seen this binary already; if so,
+    #     use |llvm_symbolizers[binary]|;
+    #  2. otherwise check if we've seen this hint already; if so,
+    #     reuse |last_llvm_symbolizer|;
+    #  3. otherwise create a new symbolizer and pass all currently known
+    #     .dSYM hints to it.
+    use_last_symbolizer = True
+    if not binary in self.llvm_symbolizers:
+      if self.system == 'Darwin' and self.dsym_hint_producer:
+        dsym_hint = self.dsym_hint_producer(binary)
+        if dsym_hint and not dsym_hint in self.dsym_hints:
+          use_last_symbolizer = False
+        self.dsym_hints.add(dsym_hint)
+      if self.last_llvm_symbolizer and use_last_symbolizer:
+          self.llvm_symbolizers[binary] = self.last_llvm_symbolizer
+      else:
+        self.last_llvm_symbolizer = LLVMSymbolizerFactory(
+            self.system, guess_arch(addr), self.dsym_hints)
+        self.llvm_symbolizers[binary] = self.last_llvm_symbolizer
     # Use the chain of symbolizers:
     # Breakpad symbolizer -> LLVM symbolizer -> addr2line/atos
     # (fall back to next symbolizer if the previous one fails).
     if not binary in symbolizers:
       symbolizers[binary] = ChainSymbolizer(
-          [BreakpadSymbolizerFactory(binary), self.llvm_symbolizer])
+          [BreakpadSymbolizerFactory(binary), self.llvm_symbolizers[binary]])
     result = symbolizers[binary].symbolize(addr, binary, offset)
     if result is None:
       # Initialize system symbolizer only if other symbolizers failed.
       symbolizers[binary].append_symbolizer(
           SystemSymbolizerFactory(self.system, addr, binary))
       result = symbolizers[binary].symbolize(addr, binary, offset)
     # The system symbolizer must produce some result.
     assert result