Continuing the discussion from Log of breaking changes in 0.2 beta:
We fixed the code for agg.sum so the return type is consistent with what’s listed in the documentation. In particular, inputs of type tbool or tint32 are implicitly converted to tint64 and the return type of agg.sum() is tint64. Likewise, inputs of type tfloat32 are implicitly converted to tfloat64 and the return type of agg.sum() is tfloat64.