Script to compare global variables between two PDBs

tools/win/pdb_compare_globals.py

Index: tools/win/pdb_compare_globals.py
diff --git a/tools/win/pdb_compare_globals.py b/tools/win/pdb_compare_globals.py
new file mode 100644
index 0000000000000000000000000000000000000000..0f9a6071327033061139655ecacfeddfd75da588
--- /dev/null
+++ b/tools/win/pdb_compare_globals.py
@@ -0,0 +1,92 @@
+# Copyright (c) 2016 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.
+
+"""
+This script uses ShowGlobals.exe to compare two PDBs to see what interesting
+globals are present in one but not the other. You can either pass in a .pdb file
+or you can pass in a .txt file that contains the results of calling ShowGlobals.
+This helps when investigating size regressions. Often code-size regressions are
+associated with global variable changes, and those global variables can be
+easier to track and investigate than the code.
+
+Typical output from ShowGlobals.exe is lines like these:
+
+  #Dups   DupSize   Size  Section Symbol name     PDB name
+
+  0       0       122784  2       kBrotliDictionary       chrome.dll.pdb
+  1       1824    0       0       LcidToLocaleNameTable   chrome.dll.pdb
+"""
+
+import os
+import subprocess
+import sys
+
+
+def LoadSymbols(pdb_name):
+  result = {}
+  extension = os.path.splitext(pdb_name)[1].lower()
+  if extension in ['.pdb']:
+    command = 'ShowGlobals.exe "%s"' % pdb_name
+    lines = subprocess.check_output(command).splitlines()
+  elif extension in ['.txt']:
+    lines = open(pdb_name).readlines()
+  else:
+    print 'Unrecognized extension in %s' % pdb_name
+    return result
+  for line in lines:
+    parts = line.split('\t')
+    if len(parts) >= 5 and not line.startswith('#'):
+      # Put the first four columns (the numerical data associated with a symbol)
+      # into a dictionary indexed by the fifth column, which is the symbol name.
+      symbol_name = parts[4]
+      result[symbol_name] = parts[:4]
+  return result
+
+
+def ShowExtras(pdbA, pdbB, nameA, nameB):

[stanisc, 2016/12/22 21:13:20]

pdbA and pdbB names are a bit confusing because these are actually collections
of data extracted from pdb's. It took me some time to realize that.

[brucedawson, 2016/12/22 21:35:00]

How about symbols_1/symbols_2 and symbols_A/symbols_B? symbols_left,
symbols_right?

I'm leaning towards symbols_1/symbols_2 in main and symbols_A/symbols_B in the
other functions.

I could use the same names in main() and in the other functions but I'm worried
that would be confusing, so I need two sets of names, I think.

Any suggestions?

+  print 'Symbols that are in %s but not in %s' % (nameA, nameB)
+  for key in pdbA:
+    if not key in pdbB:
+      # Print all the numerical data, followed by the symbol name, separated by
+      # tabs.
+      print '\t'.join(pdbA[key] + [key])
+  print
+
+
+def ShowDifferences(pdbA, pdbB, nameA, nameB):
+  print 'Symbols that are changed from %s to %s' % (nameA, nameB)
+  for key in pdbA:
+    if key in pdbB:

[stanisc, 2016/12/22 21:13:20]

if key in pdbA and key in pdbB ?

[brucedawson, 2016/12/22 21:35:00]

The key is guaranteed to be in pdbA because we are iterating through pdbA.

[stanisc, 2016/12/22 21:48:21]

What I meant is to replace two "if" statements, this one and the one above, with
a combined one with and operator.

[brucedawson, 2016/12/22 22:03:29]

I don't think that works because then we don't know what key to check. The 'for'
statement means that we go through all of the keys that are in pdbA.

That is, the first statement is not an 'if'.

+      value_a = pdbA[key]
+      value_b = pdbB[key]
+      if value_a != value_b:
+        # Print the symbol name and then the two versions of the numerical data,
+        # indented.
+        print '%s changed from/to:' % key
+        print '\t' + '\t'.join(pdbA[key])
+        print '\t' + '\t'.join(pdbB[key])

[stanisc, 2016/12/22 21:13:20]

replace pdbA[key] with value_a and pdbB[key] with value_b

[brucedawson, 2016/12/22 21:35:00]

Good catch. Done.

+  print
+
+
+def main():
+  pdb1 = LoadSymbols(sys.argv[1])
+  pdb2 = LoadSymbols(sys.argv[2])
+
+  if len(pdb1) == 0:
+    print 'No data found in %s - fastlink?' % sys.argv[1]
+    return
+  if len(pdb2) == 0:
+    print 'No data found in %s - fastlink?' % sys.argv[2]
+    return
+
+  print ('%d interesting globals in %s, %d interesting globals in %s' %
+         (len(pdb1), sys.argv[1], len(pdb2), sys.argv[2]))
+
+  ShowExtras(pdb1, pdb2, sys.argv[1], sys.argv[2])
+  ShowExtras(pdb2, pdb1, sys.argv[2], sys.argv[1])
+  ShowDifferences(pdb1, pdb2, sys.argv[1], sys.argv[2])
+
+
+if __name__ == '__main__':
+  sys.exit(main())