[foozzie] Improve failure state deduplication

tools/foozzie/v8_suppressions.py

 # Copyright 2016 the V8 project authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 """
 Suppressions for V8 correctness fuzzer failures.
 
 We support three types of suppressions:
 1. Ignore test case by pattern.
 Map a regular expression to a bug entry. A new failure will be reported
@@ skipping 130 matching lines @@
 ]
 
 
 ###############################################################################
 # Implementation - you should not need to change anything below this point.
 
 # Compile regular expressions.
 ALLOWED_LINE_DIFFS = [re.compile(exp) for exp in ALLOWED_LINE_DIFFS]
 IGNORE_LINES = [re.compile(exp) for exp in IGNORE_LINES]
 
+ORIGINAL_SOURCE_PREFIX = 'v8-foozzie source: '
 
 def line_pairs(lines):
   return itertools.izip_longest(
       lines, itertools.islice(lines, 1, None), fillvalue=None)
 
 
 def caret_match(line1, line2):
   if (not line1 or
       not line2 or
       len(line1) > MAX_LINE_LENGTH or
@@ skipping 17 matching lines @@
     match2 = exp.match(line2)
     if match1 and match2:
       # If there are groups in the regexp, ensure the groups matched the same
       # things.
       if match1.groups() == match2.groups():  # tuple comparison
         return True
   return False
 
 
 def diff_output(output1, output2, allowed, ignore1, ignore2):
+  """Returns a tuple (difference, source).
+
+  The difference is None if there's no difference, otherwise a string
+  with a readable diff.
+
+  The source is a string with the last source output within the test case.
+  It is the string 'none' if no such output existed.

[tandrii(chromium), 2017/01/13 06:38:34]

i don't understand why 'none' is a good choice, but I also don't see why it's a
bad one.

[Michael Achenbach, 2017/01/13 07:49:52]

It's a random string. Could be anything (except one of the real filenames). I
don't want to use an empty string as that can later not be differentiated
between running or not running this code at all.

In practice I hope we're not going to see this value at all. I we do, then all
failure states with the none value will be deduped on the first that occurred.
But I can use the output of that one to figure out why it happened, resolve it
and close it again. 

[Michael Achenbach, 2017/01/13 07:52:39]

But there's of course a bug :/. The none value shouldn't get hashed. It should
stay as is. Otherwise it might be mapped together with some other values... hmm.
Maybe I just pull hashing into this function...

+  """
   def useful_line(ignore):
     def fun(line):
       return all(not e.match(line) for e in ignore)
     return fun
 
   lines1 = filter(useful_line(ignore1), output1)
   lines2 = filter(useful_line(ignore2), output2)
 
+  # This keeps track where we are in the original source file of the fuzz
+  # test case.
+  source = 'none'
+
   for ((line1, lookahead1), (line2, lookahead2)) in itertools.izip_longest(
       line_pairs(lines1), line_pairs(lines2), fillvalue=(None, None)):
 
     # Only one of the two iterators should run out.
     assert not (line1 is None and line2 is None)
 
     # One iterator ends earlier.
     if line1 is None:
-      return '+ %s' % short_line_output(line2)
+      return '+ %s' % short_line_output(line2), source
     if line2 is None:
-      return '- %s' % short_line_output(line1)
+      return '- %s' % short_line_output(line1), source
 
     # If lines are equal, no further checks are necessary.
     if line1 == line2:
+      # Instrumented original-source-file output must be equal in both
+      # versions. It only makes sense to update it here when both lines
+      # are equal.
+      if line1.startswith(ORIGINAL_SOURCE_PREFIX):
+        source = line1[len(ORIGINAL_SOURCE_PREFIX):]
       continue
 
     # Look ahead. If next line is a caret, ignore this line.
     if caret_match(lookahead1, lookahead2):
       continue
 
     # Check if a regexp allows these lines to be different.
     if ignore_by_regexp(line1, line2, allowed):
       continue
 
     # Lines are different.
-    return '- %s\n+ %s' % (short_line_output(line1), short_line_output(line2))
+    return (
+        '- %s\n+ %s' % (short_line_output(line1), short_line_output(line2)),
+        source,
+    )
 
   # No difference found.
-  return None
+  return None, source
 
 
 def get_suppression(arch1, config1, arch2, config2):
   return V8Suppression(arch1, config1, arch2, config2)
 
 
 class Suppression(object):
   def diff(self, output1, output2):
     return None
 
@@ skipping 56 matching lines @@
     bug = check(IGNORE_OUTPUT.get(arch, {}))
     if bug:
       return bug
     bug = check(IGNORE_OUTPUT.get(config, {}))
     if bug:
       return bug
     bug = check(IGNORE_OUTPUT.get('%s,%s' % (arch, config), {}))
     if bug:
       return bug
     return None