moshi: Moshi slower than Gson reading quoted values.

Ran the following in large loops.

try (com.google.gson.stream.JsonReader reader = new com.google.gson.stream.JsonReader(
    new InputStreamReader(Foo.class.getClassLoader().getResourceAsStream("foo.json"), UTF_8))) {
  reader.setLenient(true);
  reader.beginArray();
  while (reader.hasNext()) {
    reader.nextLong();
  }
  reader.endArray();
}

vs.

try (JsonReader reader = JsonReader.of(
    Okio.buffer(Okio.source(Foo.class.getClassLoader().getResourceAsStream("foo.json"))))) {
  reader.setLenient(true);
  reader.beginArray();
  while (reader.hasNext()) {
    reader.nextLong();
  }
  reader.endArray();
}

With foo.json containing a huge array of single-quoted longs, Gson was about 25% faster. With foo.json containing unquoted longs, Moshi was about as fast as Gson.

I found this performance degradation in other quoted values, like nextString and nextName, too.

I suspect it is the performance of nextQuotedValue and BufferedSource’s indexOfElement. (I will have to learn how to use real measurement tools, but I wanted to file here to remember to get back to this after finding this last Friday.)