Handle zero-size allocations properly in heap profiler

base/trace_event/heap_profiler_heap_dump_writer.cc

 // Copyright 2015 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.
 
 #include "base/trace_event/heap_profiler_heap_dump_writer.h"
 
 #include <stdint.h>
 
 #include <algorithm>
 #include <iterator>
@@ skipping 117 matching lines @@
   // no bucket is discarded is doing a heap sort, which is O(n log n).
   std::make_heap(buckets.begin(), buckets.end());
 
   // Keep including buckets until adding one would increase the number of
   // bytes accounted for by less than 0.8 percent. This simple heuristic works
   // quite well. The large buckets end up in [it, end()), [begin(), it) is the
   // part that contains the max-heap of small buckets.
   size_t accounted_for = 0;
   std::vector<Bucket>::iterator it;
   for (it = buckets.end(); it != buckets.begin(); --it) {
-    // Compute the contribution to the number of bytes accounted for as a
-    // fraction of 125 (in increments of 0.8 percent). Anything less than 1/125
-    // is rounded down to 0 due to integer division. Buckets are iterated by
-    // descending size, so later buckets cannot have a larger contribution than
-    // this one.
+    // Compute the total accounted amount of bytes
+    // If next bucket increase this amount by less then 0.8% just skip the tail

[Primiano Tucci (use gerrit), 2016/04/06 08:20:44]

I'd add a "(1/125)" after 0.8% otherwise It's not clear where it comes from.
Also, nit: missing trailing dots.

     accounted_for += buckets.front().size;
-    size_t contribution = buckets.front().size * 125 / accounted_for;
-    if (contribution == 0)
+    if (buckets.front().size < (accounted_for / 125))
       break;
 
     // Put the largest bucket in [begin, it) at |it - 1| and max-heapify
     // [begin, it - 1). This puts the next largest bucket at |buckets.front()|.
     std::pop_heap(buckets.begin(), it);
   }
 
   // At this point, |buckets| looks like this (numbers are bucket sizes):
   //
   // <-- max-heap of small buckets --->
@@ skipping 71 matching lines @@
       BreakDown(subbucket);
 }
 
 const std::set<Entry>& HeapDumpWriter::Summarize(
     const hash_map<AllocationContext, size_t>& bytes_by_context) {
   // Start with one bucket that represents the entire heap. Iterate by
   // reference, because the allocation contexts are going to point to allocation
   // contexts stored in |bytes_by_context|.
   Bucket root_bucket;
   for (const auto& context_and_size : bytes_by_context) {
+    DCHECK_GT(context_and_size.second, 0u);
     const AllocationContext* context = &context_and_size.first;
     const size_t size = context_and_size.second;
     root_bucket.bytes_by_context.push_back(std::make_pair(context, size));
     root_bucket.size += size;
   }
 
   AddEntryForBucket(root_bucket);
 
   // Recursively break down the heap and fill |entries_| with entries to dump.
   BreakDown(root_bucket);
@@ skipping 46 matching lines @@
     const hash_map<AllocationContext, size_t>& bytes_by_size,
     StackFrameDeduplicator* stack_frame_deduplicator,
     TypeNameDeduplicator* type_name_deduplicator) {
   internal::HeapDumpWriter writer(stack_frame_deduplicator,
                                   type_name_deduplicator);
   return Serialize(writer.Summarize(bytes_by_size));
 }
 
 }  // namespace trace_event
 }  // namespace base