[Buildroot] [PATCH 1/2] Add BR2_CMAKE_USE_NINJA_BACKEND option

Cédric Marie cedric.marie at openmailbox.org
Wed Feb 1 17:01:31 UTC 2017


Hi,

Le 2017-01-30 10:23, Thomas Petazzoni a écrit :
> OK, understood. But do you have numbers showing that "make foo-rebuild"
> is actually faster with Ninja ?
> 
> In the tests done by Romain, it's only saving 1-2 seconds on the
> total build of a package. So I would suspect that the savings on a
> partial build are even smaller.
> 
> Even if your CMake packages are private, can you give us some numbers
> that show the time benefits of Ninja?

You're right, it is a private package, that's why I had to test on other 
packages I'm not particularly using, to demonstrate.

The benefit is rather small for my package too.

Make:
real	1m1.517s
user	2m23.996s
sys	0m36.197s

Ninja:
real	0m56.312s
user	2m19.078s
sys	0m29.119s

NB: It includes the whole "time make foo-rebuild" command, because time 
output is "strange" when inserted in pkg-cmake.mk (... time 
$$($$(PKG)_MAKE) ...), don't know why...
Looks like:
130.50user 19.88system 0:42.32elapsed 355%CPU (0avgtext+0avgdata 
24680maxresident)k
0inputs+50416outputs (0major+7602195minor)pagefaults 0swaps

To be honest, when I started to add ninja possibility, I expected much 
bigger savings :)
Yet I have the option, so even if the benefit is small, I keep using it.

But in the end, I don't know whether it's worth pushing upstream or 
not...

Besides Buildroot, I also switched another package from CMake/Make to 
Meson (based on Ninja), and the benefit was much more interesting - 
although I have not kept any measure to give here.
I expected the improvement to be caused by Ninja, not Meson. But it 
seems that Meson makes the difference, rather than Make vs Ninja.


-- 
Cédric



More information about the buildroot mailing list