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 th
brucedawson
2016/12/22 21:35:00
How about symbols_1/symbols_2 and symbols_A/symbol
|
+ 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]) |
+ |
+ |
+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
stanisc
2016/12/22 21:48:21
What I meant is to replace two "if" statements, th
brucedawson
2016/12/22 22:03:29
I don't think that works because then we don't kno
|
+ 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
brucedawson
2016/12/22 21:35:00
Good catch. Done.
|
+ |
+ |
+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()) |