Improvements to uses of base::SmallMap

storage/browser/blob/blob_memory_controller.cc

Index: storage/browser/blob/blob_memory_controller.cc
diff --git a/storage/browser/blob/blob_memory_controller.cc b/storage/browser/blob/blob_memory_controller.cc
index 298037cc5eed9b4bb240bb431ec80f2f770cf2a9..eae2ac4e6c2dd5b023c12e812894362e87cfb72b 100644
--- a/storage/browser/blob/blob_memory_controller.cc
+++ b/storage/browser/blob/blob_memory_controller.cc
@@ -245,7 +245,7 @@ uint64_t GetTotalSizeAndFileSizes(
         unreserved_file_items,
     std::vector<uint64_t>* file_sizes_output) {
   uint64_t total_size_output = 0;
-  base::SmallMap<std::map<uint64_t, uint64_t>> file_id_to_sizes;
+  base::small_map<std::map<uint64_t, uint64_t>> file_id_to_sizes;

[danakj, 2017/04/18 21:35:50]

Aside.. but this is interesting. What it really wants is something like

vector<pair<int, int>> sizes;
for (each item i)
  sizes.push_back(file_id, offset+length);
sort(sizes, by_pair_first);  // group up same ids together
accumulate(sizes);  // reduce consecutive things with same first by adding their
second.

That could give much better performance I think than a map. I wonder if such
helpful STL tools exist.

[brettw, 2017/04/19 16:29:48]

An STL nerd would say yes, but my guess is it would be 76 lines of unreadable
std::for_each code :(

   for (const auto& item : unreserved_file_items) {
     const DataElement& element = item->item()->data_element();
     uint64_t file_id = BlobDataBuilder::GetFutureFileID(element);