The "limit" argument to mongoc_collection_find should be used for two purposes:

1. Mark a mongoc_cursor_t as "done" once it has fetched "limit" documents
2. Send limit to the server as the nToReturn field in the OP_QUERY and OP_GETMORE message headers

The driver respects #1, but not #2. Instead, it always sends nToReturn of 0 to the server, meaning, "accept server default" batch size of 101 documents at first, then 4MB of documents after.

This bug incurs a performance penalty: the final batch is always full-sized (about 4MB) even if the cursor only needs part of it.

If mongoc_collection_find's batch_size is non-0 the logic is correct; it's when batch_size is 0 (the default) that limit is incorrectly ignored when formatting OP_QUERY and OP_GETMORE:

static int32_t
_mongoc_n_return (mongoc_cursor_t * cursor)
{
   /* by default, use the batch size */
   int32_t r = cursor->batch_size;

   if (cursor->is_command) {
      /* commands always have n_return of 1 */
      r = 1;
   } else if (cursor->limit) {
      /* calculate remaining */
      uint32_t remaining = cursor->limit - cursor->count;

      /* use min of batch or remaining */
      r = BSON_MIN(r, (int32_t)remaining);
   }

   return r;
}