I believe we need to return an error instead of silently ignoring write. we should help developer uncover his/her errors (not executing Set/Delete on batch), not ignoring them.

Then why did you approve #67? 🙂 This can only be nil if we write a closed or empty batch, in which case we need to silently ignore the write to satisfy the semantics tested by #67. If we instead set b.batch to a new, empty batch on close we will have a memory leak.

This can only be nil if we write a closed or empty batch, in which case we need to silently ignore the write to satisfy the semantics tested by #67.

can't you modify #67 if you agree with my logic?

Sure, I'll submit a new PR. It will be a bit more breaking, because we have to change the Batch interface to return errors etc, but I agree it's a better solution.