metadata_v2: get rid of entry_count, further reducing size

include/dwarfs/metadata_types.h

@@ -73,6 +73,8 @@ class directory_view
   ::apache::thrift::frozen::View<thrift::metadata::directory> getdir() const;
   ::apache::thrift::frozen::View<thrift::metadata::directory>
   getdir(uint32_t ino) const;
+  uint32_t entry_count(
+      ::apache::thrift::frozen::View<thrift::metadata::directory> self) const;
 
   ::apache::thrift::frozen::MappedFrozen<thrift::metadata::metadata> const*
       meta_;

src/dwarfs/entry.cpp

@@ -205,7 +205,7 @@ void dir::pack(thrift::metadata::metadata& mv2,
   d.parent_inode =
       has_parent() ? std::dynamic_pointer_cast<dir>(parent())->inode_num() : 0;
   d.first_entry = mv2.entries.size();
-  d.entry_count = entries_.size();
+  // d.entry_count = entries_.size();
   mv2.directories.push_back(d);
   for (entry_ptr const& e : entries_) {
     mv2.entry_index.at(e->inode_num()) = mv2.entries.size();

src/dwarfs/metadata_types.cpp

@@ -48,12 +48,18 @@ directory_view::getdir(uint32_t ino) const {
   return meta_->directories()[ino];
 }
 
-uint32_t directory_view::entry_count() const { return getdir().entry_count(); }
+uint32_t directory_view::entry_count() const { return entry_count(getdir()); }
+
+uint32_t directory_view::entry_count(
+    ::apache::thrift::frozen::View<thrift::metadata::directory> self) const {
+  auto next = getdir(inode() + 1);
+  return next.first_entry() - self.first_entry();
+}
 
 boost::integer_range<uint32_t> directory_view::entry_range() const {
   auto d = getdir();
   auto first = d.first_entry();
-  return boost::irange(first, first + d.entry_count());
+  return boost::irange(first, first + entry_count(d));
 }
 
 uint32_t directory_view::first_entry() const { return getdir().first_entry(); }

src/dwarfs/scanner.cpp

@@ -168,12 +168,18 @@ class save_directories_visitor : public visitor_base {
 
   void pack(thrift::metadata::metadata& mv2, global_entry_data& ge_data) {
     for (auto p : directories_) {
-      p->pack(mv2, ge_data);
-
       if (!p->has_parent()) {
         p->pack_entry(mv2, ge_data);
       }
+
+      p->pack(mv2, ge_data);
     }
+
+    thrift::metadata::directory dummy;
+    dummy.parent_inode = 0;
+    dummy.first_entry = mv2.entries.size();
+    // dummy.entry_count = 0;
+    mv2.directories.push_back(dummy);
   }
 
  private:
@@ -451,6 +457,7 @@ void scanner_<LoggerPolicy>::scan(filesystem_writer& fsw,
 
   log_.info() << "saving directories...";
   mv2.entry_index.resize(first_file_inode + im->count());
+  mv2.directories.reserve(first_link_inode + 1);
   save_directories_visitor sdv(first_link_inode);
   root->accept(sdv);
   sdv.pack(mv2, ge_data);

thrift/metadata.thrift

@@ -44,7 +44,6 @@ struct chunk {
 struct directory {
    1: required UInt32 parent_inode,
    2: required UInt32 first_entry,
-   3: required UInt32 entry_count,   // TODO: we can remove this if we sort entries by directory
 }
 
 /**
@@ -97,15 +96,22 @@ struct metadata {
     */
    1: required list<chunk>     chunks,
 
-   // all directories, indexed by inode number
+   /**
+    * All directories, indexed by inode number. There's one extra
+    * dummy directory at the end whose `first_entry` point to the
+    * end of `entries`, so that directory entry lookup work the
+    * same for all directories.
+    */
    2: required list<directory> directories,
 
    // all entries, can be looked up by inode through entry_index
    3: required list<entry>     entries,
 
-   // chunk index, indexed by (inode - chunk_index_offset); this
-   // list has one extra item at the back that points to the end
-   // of `chunks`, so chunk lookups work the same for all inodes
+   /**
+    * Chunk index, indexed by (inode - chunk_index_offset).
+    * There's one extra dummy item at the end that points to the
+    * end of `chunks`, so chunk lookups work the same for all inodes.
+    */
    4: required list<UInt32>    chunk_index,
 
    // entry index, indexed by inode