Junio C Hamano <gitster@pobox.com> writes:

> Justin Tobler <jltobler@gmail.com> writes:
>
>>> +	err = stack_segments_for_compaction(st, &seg);
>>> +	if (err)
>>> +		return err;
>>> +
>>> +	*required = segment_size(&seg) > 0;
>>
>> As mentioned on the previous patch, I wonder if we could just return the
>> number of tables in the compaction segment as part of
>> `stack_segments_for_compaction()`. A negative value could indicate an
>> error. All other values would reflect the number of tables to be
>> compacted.
>>
>> This way callers interested in whether compaction should be performed
>> could just do: stack_segments_for_compaction > 0. We could maybe avoid
>> having a separate function like we do here and just expose
>> `stack_segments_for_compaction()`.
>
> Is the cost of compacting a single table expected to be roughly the
> same across tables?  The number of tables to be compacted would not
> be a useful information to help making a better decision otherwise,
> so I am guessing that it is the underlying assumption the above
> suggestion comes from.

It would be sufficient information to know whether or not we can
compact. But only when in 'git refs optimize --auto' mode.