binary_size_tool: fix for ambiguous addr2line output

build/android/pylib/symbols/elf_symbolizer.py

 # Copyright 2014 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 import collections
 import datetime
 import logging
 import multiprocessing
 import os
 import posixpath
@@ skipping 57 matching lines @@
   |max_queue_size| bound, a new addr2line instance is kicked in.
   In the case of a very eager producer (i.e. all |max_concurrent_jobs| instances
   have a backlog of |max_queue_size|), back-pressure is applied on the caller by
   blocking the SymbolizeAsync method.
 
   This module has been deliberately designed to be dependency free (w.r.t. of
   other modules in this project), to allow easy reuse in external projects.
   """
 
   def __init__(self, elf_file_path, addr2line_path, callback, inlines=False,
-      max_concurrent_jobs=None, addr2line_timeout=30, max_queue_size=50):
+      max_concurrent_jobs=None, addr2line_timeout=30, max_queue_size=50,
+      source_root_path=None, strip_base_path=None):
     """Args:
       elf_file_path: path of the elf file to be symbolized.
       addr2line_path: path of the toolchain's addr2line binary.
       callback: a callback which will be invoked for each resolved symbol with
           the two args (sym_info, callback_arg). The former is an instance of
           |ELFSymbolInfo| and contains the symbol information. The latter is an
           embedder-provided argument which is passed to SymbolizeAsync().
       inlines: when True, the ELFSymbolInfo will contain also the details about
           the outer inlining functions. When False, only the innermost function
           will be provided.
       max_concurrent_jobs: Max number of addr2line instances spawned.
           Parallelize responsibly, addr2line is a memory and I/O monster.
       max_queue_size: Max number of outstanding requests per addr2line instance.
       addr2line_timeout: Max time (in seconds) to wait for a addr2line response.
           After the timeout, the instance will be considered hung and respawned.
+      source_root_path: The path to the library source code is used in

[Andrew Hayden (chromium.org), 2014/06/26 09:00:11]

Grammar here still needs to be cleaned up a bit. I think the original
description I provided is sufficient here, so let's move your example to the end
instead of the beginning of this text. How about this?

In some toolchains only the name of the source
file is output, without any path information; disambiguation searches
through the source directory specified by |source_root_path| argument
for files whose name matches, adding the full path information to the
output. For example, if the toolchain outputs "unicode.cc" and there
is a file called "unicode.cc" located under |source_root_path|/foo/bar,
the tool will replace "unicode.cc" with
"|source_root_path|/foo/bar/unicode.cc". If there are multiple files
with the same name, disambiguation will fail because the tool cannot
determine which of the files was the source of the symbol.

+          the disambiguation process. Disambiguation means to resolve ambiguous
+          source paths, for example turn addr2line output "unicode.cc" into a
+          absolute path. In some toolchains only the name of the source
+          file is output, without any path information; disambiguation searches
+          through the source directory specified by |source_root_path| argument
+          for files whose name matches. If there are multiple files with the
+          same name, disambiguation will fail.
+      strip_base_path: Rebases the symbols source paths onto |source_root_path|
+          (i.e. s/^strip_base_path/source_root_path/).

[Andrew Hayden (chromium.org), 2014/06/26 09:00:11]

This is a nice succinct explanation, though the use of search and replace syntax
here may be confusing for folks who don't use Linux/Mac... but since it's an ELF
symbolizer, I guess that's ok :P :P

Still for clarity and accessibility to a wider audience you might want to
provide an example.

[Primiano Tucci (use gerrit), 2014/06/27 08:53:56]

I find the idea that a developer is not familiar with the regex syntax
terrifying, but I'm fine either ways  ;-) 

     """
     assert(os.path.isfile(addr2line_path)), 'Cannot find ' + addr2line_path
     self.elf_file_path = elf_file_path
     self.addr2line_path = addr2line_path
     self.callback = callback
     self.inlines = inlines
     self.max_concurrent_jobs = (max_concurrent_jobs or
                                 min(multiprocessing.cpu_count(), 4))
     self.max_queue_size = max_queue_size
     self.addr2line_timeout = addr2line_timeout
     self.requests_counter = 0  # For generating monotonic request IDs.
     self._a2l_instances = []  # Up to |max_concurrent_jobs| _Addr2Line inst.
 
+    # If necessary, create disambiguation lookup table
+    self.disambiguate = source_root_path is not None
+    self.disambiguation_table = {}
+    self.strip_base_path = strip_base_path
+    if(self.disambiguate):
+      self.source_root_path = os.path.abspath(source_root_path)
+      self._CreateDisambiguationTable()
+
     # Create one addr2line instance. More instances will be created on demand
     # (up to |max_concurrent_jobs|) depending on the rate of the requests.
     self._CreateNewA2LInstance()
 
   def SymbolizeAsync(self, addr, callback_arg=None):
     """Requests symbolization of a given address.
 
     This method is not guaranteed to return immediately. It generally does, but
     in some scenarios (e.g. all addr2line instances have full queues) it can
     block to create back-pressure.
@@ skipping 37 matching lines @@
     for a2l in self._a2l_instances:
       a2l.WaitForIdle()
       a2l.Terminate()
 
   def _CreateNewA2LInstance(self):
     assert(len(self._a2l_instances) < self.max_concurrent_jobs)
     a2l = ELFSymbolizer.Addr2Line(self)
     self._a2l_instances.append(a2l)
     return a2l
 
+  def _CreateDisambiguationTable(self):
+    """ Non-unique file names will result in None entries"""
+    self.disambiguation_table = {}
+
+    for root, _, filenames in os.walk(self.source_root_path):
+      for f in filenames:
+        self.disambiguation_table[f] = os.path.join(root, f) if (f not in
+                                       self.disambiguation_table) else None
+
 
   class Addr2Line(object):
     """A python wrapper around an addr2line instance.
 
     The communication with the addr2line process looks as follows:
       [STDIN]         [STDOUT]  (from addr2line's viewpoint)
     > f001111
     > f002222
                     < Symbol::Name(foo, bar) for f001111
                     < /path/to/source/file.c:line_number
@@ skipping 131 matching lines @@
         source_line = None
         m = ELFSymbolizer.Addr2Line.SYM_ADDR_RE.match(line2)
         if m:
           if not m.group(1).startswith('?'):
             source_path = m.group(1)
             if not m.group(2).startswith('?'):
               source_line = int(m.group(2))
         else:
           logging.warning('Got invalid symbol path from addr2line: %s' % line2)
 
-        sym_info = ELFSymbolInfo(name, source_path, source_line)
+        # In case disambiguation is on, and needed

[Andrew Hayden (chromium.org), 2014/06/26 09:00:11]

This loop has gotten a lot nicer. I think it's self-documenting now, which is
cool. Nice job.

[Primiano Tucci (use gerrit), 2014/06/26 09:33:45]

Really? is it?

I think that nobody here is really understanding all the cases where the path is
kept as it is vs. made absolute (and w.r.t of which base dir).
What would be self-documenting, instead, would be code in the unittests which
clearly defines the possible input and the output expectations.

Just to be clear, this still LGTM. I just give up trying to follow  the path
logic here. Similarly, as a reviewer, I will not encourage any developer, who
will happen to touch this code in future, to spend any time trying to reverse
engineer the intended behavior here.

This is the very first Chrome dev principle: there is just code which is covered
by tests and code which is broken.

+        was_ambiguous = False
+        disambiguated = False
+        if self._symbolizer.disambiguate:
+          if source_path and not os.path.isabs(source_path):
+            path = self._symbolizer.disambiguation_table.get(source_path)
+            was_ambiguous = True
+            disambiguated = path is not None
+            source_path = path if disambiguated else source_path
+
+          # Use absolute paths (so that paths are consistent, as disambiguation
+          # uses absolute paths)
+          if source_path and not was_ambiguous:
+            source_path = os.path.abspath(source_path)
+
+        if source_path and self._symbolizer.strip_base_path:
+          # Strip the base path
+          source_path = re.sub('^' + self._symbolizer.strip_base_path,
+              self._symbolizer.source_root_path or '', source_path)
+
+        sym_info = ELFSymbolInfo(name, source_path, source_line, was_ambiguous,
+                                 disambiguated)
         if prev_sym_info:
           prev_sym_info.inlined_by = sym_info
         if not innermost_sym_info:
           innermost_sym_info = sym_info
 
       self._processed_symbols_count += 1
       self._symbolizer.callback(innermost_sym_info, callback_arg)
 
     def _RestartAddr2LineProcess(self):
       if self._proc:
@@ skipping 60 matching lines @@
 
     @property
     def first_request_id(self):
       """Returns the request_id of the oldest pending request in the queue."""
       return self._request_queue[0][2] if self._request_queue else 0
 
 
 class ELFSymbolInfo(object):
   """The result of the symbolization passed as first arg. of each callback."""
 
-  def __init__(self, name, source_path, source_line):
+  def __init__(self, name, source_path, source_line, was_ambiguous=False,
+               disambiguated=False):
     """All the fields here can be None (if addr2line replies with '??')."""
     self.name = name
     self.source_path = source_path
     self.source_line = source_line
     # In the case of |inlines|=True, the |inlined_by| points to the outer
     # function inlining the current one (and so on, to form a chain).
     self.inlined_by = None
+    self.disambiguated = disambiguated
+    self.was_ambiguous = was_ambiguous
 
   def __str__(self):
     return '%s [%s:%d]' % (
         self.name or '??', self.source_path or '??', self.source_line or 0)