Be resilient against failed runs

tools/profiling/microbenchmarks/bm_diff.py

@@ -204,7 +204,10 @@ def eintr_be_gone(fn):
 
 
 def read_json(filename):
-  with open(filename) as f: return json.loads(f.read())
+  try:
+    with open(filename) as f: return json.loads(f.read())
+  except ValueError, e:
+    return None
 
 
 def finalize():
@@ -217,16 +220,18 @@ def finalize():
       js_old_ctr = read_json('%s.counters.old.%d.json' % (bm, loop))
       js_old_opt = read_json('%s.opt.old.%d.json' % (bm, loop))
 
-      for row in bm_json.expand_json(js_new_ctr, js_new_opt):
-        print row
-        name = row['cpp_name']
-        if name.endswith('_mean') or name.endswith('_stddev'): continue
-        benchmarks[name].add_sample(row, True)
-      for row in bm_json.expand_json(js_old_ctr, js_old_opt):
-        print row
-        name = row['cpp_name']
-        if name.endswith('_mean') or name.endswith('_stddev'): continue
-        benchmarks[name].add_sample(row, False)
+      if js_new_ctr:
+        for row in bm_json.expand_json(js_new_ctr, js_new_opt):
+          print row
+          name = row['cpp_name']
+          if name.endswith('_mean') or name.endswith('_stddev'): continue
+          benchmarks[name].add_sample(row, True)
+      if js_old_ctr:
+        for row in bm_json.expand_json(js_old_ctr, js_old_opt):
+          print row
+          name = row['cpp_name']
+          if name.endswith('_mean') or name.endswith('_stddev'): continue
+          benchmarks[name].add_sample(row, False)
 
   really_interesting = set()
   for name, bm in benchmarks.items():