Re: [PATCH v8 4/5] Implement line-history search (git log -L)

From: Thomas Rast <trast@student.ethz.ch>
Date: Mar 1, 2013, 08:49 UTC
Message-ID: <87vc9b1p48.fsf@pctrast.inf.ethz.ch>
In-Reply-To: <7vmwuogjsm.fsf@alter.siamese.dyndns.org>

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

Show 17 quoted lines

> Junio C Hamano <gitster@pobox.com> writes:
>
>> Overall, I like this better than the "log --follow" hack; as the
>> revision traversal is done without any pathspec when being "careful
>> and slow" (aka -M), you do not suffer from the "just use a singleton
>> pathspec globally regardless of what other history paths are being
>> traversed" limitation of "log --follow".
>>
>> The patch series certainly is interesting.
>
> Having said that, I notice that "careful and slow" is just "too slow
> to be usable" even on a small tree like ours.  Try running
>
>     $ git log -M -L:get_name:builtin/describe.c
>
> and see how long you have to wait until you hit the first line of
> output.

I'll dig some more. It *should* be essentially the following times taken together:

  $ time git log --raw -M --topo-order >/dev/null

  real    0m5.448s
  user    0m4.599s
  sys     0m0.794s
  $ time git log -L:get_name:builtin/describe.c >/dev/null

  real    0m0.832s
  user    0m0.796s
  sys     0m0.032s

  $ time git log -L:get_name:builtin-describe.c 81b50f3ce40^ >/dev/null

  real    0m0.489s
  user    0m0.465s
  sys     0m0.022s

So I'm losing a factor of about 4 somewhere, which I can't explain right now.

It could be improved further if --follow was fixed, because then the first step should be much faster than diffing *all* the trees.

-- 
Thomas Rast
trast@{inf,student}.ethz.ch