[Scons-users] Performance of version 2.5.0 vs 2.3.0 on Windows host dropped significantly

[William Blevins]

Bill,

Since the memoization of path_func resolved this particular performance
issue, I don't think it is tree related for Henrik's case, which is why I
am curious. "I compared some Scons versions for our reference project (C++,
161 C/C++ files eg targets) which is a real-world gcc cross-compilation
project not some odd test case." This may imply that path_func needs to be
optimized if possible. Maybe this is just doing a lot of subst calls in
this case, but I am only guessing.

V/R,
William

On Sun, Aug 14, 2016 at 3:04 AM, Bill Deegan <bill at baddogconsulting.com>
wrote:

> It would also be useful to see that tree pre-2.5 and with 2.5, even
> obfuscated.
> -Bill
>
> On Sat, Aug 13, 2016 at 7:03 PM, Bill Deegan <bill at baddogconsulting.com>
> wrote:
>
>> I have to think that given the output of  --tree=prune, even obfuscated,
>> we could create a tree of files and SConstruct/SConscripts to mimic it.
>> And then use that as a benchmark.
>> -Bill
>>
>> On Sat, Aug 13, 2016 at 5:56 PM, William Blevins <wblevins001 at gmail.com>
>> wrote:
>>
>>> Henrik (or anyone else suffering from this specific issue),
>>>
>>> Sorry that I missed an optimization issue in 2.5.0. It hasn't seemed to
>>> affect many projects, so it appears to be dependent on particular project
>>> properties outside of my prediction and which I can only guess about. I was
>>> curious if I had forgotten or neglected to run the scons_testsuite
>>> <https://bitbucket.org/dirkbaechle/scons_testsuite> like a good
>>> developer. Although I am not building all 6 with this machine/setup, of the
>>> 3 that build with my current tool setup, there is very little performance
>>> difference (1-2% tops).
>>>
>>> Is your project source available? In order to try to minimize future
>>> risk, it may be wise for the SCons team to add some diversity to the sample
>>> builds for future profiles.
>>>
>>> V/R,
>>> William
>>>
>>> On Sun, Aug 14, 2016 at 12:47 AM, William Blevins <wblevins001 at gmail.com
>>> > wrote:
>>>
>>>> Henrik,
>>>>
>>>> It has been merged into the trunk and will be available in the future.
>>>>
>>>> V/R,
>>>> William
>>>>
>>>> On Wed, Aug 10, 2016 at 7:45 AM, <hmnews at proconx.com> wrote:
>>>>
>>>>> William,
>>>>>
>>>>> Here are the stats for the full build (after clean):
>>>>>
>>>>> 2.3.0: 12.8s
>>>>> 2.3.5: 13.3s
>>>>> 2.3.6: 13.3s
>>>>> 2.5.0: 19.3s
>>>>> 2.5.0 with pull request applied: 13.42 !!!
>>>>>
>>>>> There is no measurable difference between 2.3.5, 2.3.6 and 2.5.0 with
>>>>> PR applied. Differences in the sub-second level are influenced by
>>>>> measurement noise. 2.3.0 appears slightly faster, I never managed to get
>>>>> below 13s with the other versions.
>>>>>
>>>>> All measurements, these full build times and previous no-change times
>>>>> were using num_jobs set to 8 which is my default.
>>>>>
>>>>> So I definitely suggest to roll your PR into the next release.
>>>>>
>>>>> Henrik
>>>>>
>>>>> William Blevins wrote:
>>>>>
>>>>>> Henrik,
>>>>>>
>>>>>> As a side note, the performance differences from previous versions is
>>>>>> useful. Do you also happen to have performance for full builds for my
>>>>>> reference? Some of those patches may have traded base overhead (slower
>>>>>> already up-to-date) for faster runtime (full or partial build time).
>>>>>>
>>>>>> V/R,
>>>>>> William
>>>>>>
>>>>>> _______________________________________________
>>>>> Scons-users mailing list
>>>>> Scons-users at scons.org
>>>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>>>
>>>>
>>>>
>>>
>>> _______________________________________________
>>> Scons-users mailing list
>>> Scons-users at scons.org
>>> https://pairlist4.pair.net/mailman/listinfo/scons-users
>>>
>>>
>>
>
> _______________________________________________
> Scons-users mailing list
> Scons-users at scons.org
> https://pairlist4.pair.net/mailman/listinfo/scons-users
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://pairlist4.pair.net/pipermail/scons-users/attachments/20160814/cce88303/attachment.html>