Improve error handling of the transport security state generator.

net/tools/transport_security_state_generator/trie/trie_bit_buffer.cc

Index: net/tools/transport_security_state_generator/trie/trie_bit_buffer.cc
diff --git a/net/tools/transport_security_state_generator/trie/trie_bit_buffer.cc b/net/tools/transport_security_state_generator/trie/trie_bit_buffer.cc
index 5add4ead22c9fb2ba3fe3512674c17dcc494e519..e9e8f4b9a684e0d9599930e2db2dc4650d3cd7b4 100644
--- a/net/tools/transport_security_state_generator/trie/trie_bit_buffer.cc
+++ b/net/tools/transport_security_state_generator/trie/trie_bit_buffer.cc
@@ -71,16 +71,19 @@ uint8_t TrieBitBuffer::BitLength(uint32_t input) const {
   return number_of_bits;
 }
 
-void TrieBitBuffer::WriteChar(uint8_t byte,
+bool TrieBitBuffer::WriteChar(uint8_t byte,
                               const HuffmanRepresentationTable& table,
                               HuffmanBuilder* huffman_builder) {
   HuffmanRepresentationTable::const_iterator item;
   item = table.find(byte);
-  DCHECK(item != table.end());
+  if (item == table.end()) {
+    return false;
+  }

[Ryan Sleevi, 2017/02/14 20:53:17]

Why the change from DCHECK to being a soft-check? What's the motivation?

[martijnc, 2017/02/15 22:07:59]

Not crashing on incorrect inputs. But looking at this again the DCHECK was
probably more appropriate because the HuffmanRepresentationTable is generated by
the code based on the (user) input. Missing characters thus indicate an error in
the code and should never happen, even with incorrect input.

Reverted to the old behavior.

   if (huffman_builder) {
     huffman_builder->RecordUsage(byte);
   }
   WriteBits(item->second.bits, item->second.number_of_bits);
+  return true;
 }
 
 void TrieBitBuffer::AppendBitsElement(uint8_t bits, uint8_t number_of_bits) {