Re: [Automake-NG] [Automake-ng] [PATCH 0/4] Minor refactoring and cleanup for the dist-related targets

[Stefano Lattarini]

On 05/10/2012 07:08 PM, Jim Meyering wrote:
> Stefano Lattarini wrote:
> 
>> Hi Jim, automakers.
>>
>> On 05/06/2012 08:25 PM, Jim Meyering wrote:
>>> Stefano Lattarini wrote:
>>>> These changes don't offer great simplifications nor optimizations, but
>>>> move some logic from automake execution time and recipe execution
>>>> time to make runtime, and clean up some old namespace pollution and API
>>>> leakiness -- which is the general direction Automake-NG should be moving
>>>> along IMHO.
>>>>
>>>> So, ACK for this?  I will push in 72 hours if there is no objection, but
>>>> I'd like a review before then if possible.
>>>
>>> Hi Stefano,
>>>
>>> I like the ideas and skimmed through the patches (they look fine), but
>>> did not apply or test.
>>>
>> I've now made some simple performance tests on the series.
>>
>> I have configured a tree with 5000 files in EXTRA_DIST at the top-level,
>> and 5000 more files in EXTRA_DIST in a $(SUBDIR).  On my slowish Debian
>> system:
>>
>>   * "make all" run 100 times in a row
>>      (before the series): ~ 25 seconds
>>      (before the series): ~ 26 seconds
>>
Oops, it was "after the series" in the second line.  Ditto for the timings
below.

>>   * "make distdir --dry-run" run 100 times in a row
>>      (before the series): ~ 1 minute
>>      (before the series): ~ 1 minute and a half
>>
>>   * "make distdir" run 10 times in a row
>>      (before the series): ~ 5 minutes and 50 seconds
>>      (before the series): ~ 6 minutes and 15 seconds
>>
>> All in all, good enough I think.  I will push by this evening if there is
>> no further comment.
> 
> 33% speed-up for "make distdir --dry-run"
>  7% speed-up for "make distdir"
>  4% speed-up for "make all"
>
Unfortunately, you got it backwards: it's no measurable change for "make all",
a 7% slow down for "make dist", and a 50% slow down for "make distdir -n" :-(
The plus is that "make dist -n" is now more "faithful", since it displays the
actual list of files that get distributed and of sudirectories that get
created in the distribution directory.

If we could manage to use ':=' to define the Automake new internal variables
(am__dist_files, am__dist_files_cooked, etc), *then* we would get a speed-up
of 8-10% in "make distdir -n" case, and possible a 2-3% in the "make dist"
case.  But alas, we still lacks the scaffolding for that.

Regards,
  Stefano