Improve symbol_extractor.CreateNameToSymbolInfo handling of duplicate symbols.

tools/cygprofile/symbol_extractor.py

Index: tools/cygprofile/symbol_extractor.py
diff --git a/tools/cygprofile/symbol_extractor.py b/tools/cygprofile/symbol_extractor.py
index 81c6e66d8fbca2805ba8b7c673d12eb18a3d8d14..fa8c45065b73a29afcf6f8629a48f8a4a28949d8 100755
--- a/tools/cygprofile/symbol_extractor.py
+++ b/tools/cygprofile/symbol_extractor.py
@@ -6,6 +6,7 @@
 """Utilities to get and manipulate symbols from a binary."""
 
 import collections
+import logging
 import os
 import re
 import subprocess
@@ -116,4 +117,21 @@ def CreateNameToSymbolInfo(symbol_infos):
   Returns:
     a dict {name: symbol_info, ...}
   """
-  return {symbol_info.name: symbol_info for symbol_info in symbol_infos}
+  symbol_infos_by_name = {}

[Benoit L, 2015/02/23 14:09:49]

This function no longer does what it says in its documentation. Can you please
update it?

[azarchs, 2015/02/23 15:39:37]

I'd say it's functionality relative the documentation is unchanged. It was
perhaps underspecified before, so I've improved that.

+  duplicated_symbols = set()

[pasko, 2015/02/23 14:28:13]

They rather 'collided' than were 'duplicated'. So let's refer to it as 'symbol
collissions'?

[azarchs, 2015/02/23 15:39:37]

Done.

+  for symbol_info in symbol_infos:
+    if symbol_info.name in symbol_infos_by_name:
+      # If a symbol appears at multiple offsets, keep the lowest offset.
+      duplicated_symbols.add(symbol_info.name)
+      if symbol_infos_by_name[symbol_info.name].offset > symbol_info.offset:
+        symbol_infos_by_name[symbol_info.name] = symbol_info
+    else:
+      symbol_infos_by_name[symbol_info.name] = symbol_info
+  if len(duplicated_symbols) > 0:
+    logging.warning('%d symbols were found at more than one offset.' %
+                    len(duplicated_symbols))
+    duplicated_symbols_to_show = min(50, len(duplicated_symbols))
+    logging.warning('First %d duplicated symbols:\n%s' % (

[pasko, 2015/02/23 14:28:13]

I would prefer warning level on the level of check_orderfile.py, where we have
_MAX_WARNINGS_TO_PRINT. Exploding logging limits is not very nice.

This function can be implemented as GroupSymbolInfosByName, where for each name
symbol_infos would be sorted by offset. (see about 'Slicing' here:
http://stackoverflow.com/questions/8024571/insert-an-item-into-sorted-list-py...)

Then it's a slight modification in _CountMisorderedSymbols to take the first
item, and limit warnings).

[azarchs, 2015/02/23 15:39:38]

Done.

+        duplicated_symbols_to_show,
+        '\n'.join(list(duplicated_symbols)[:duplicated_symbols_to_show])))
+  return symbol_infos_by_name