[postgis-devel] on parallel and recursive make

[rmrodriguez at carto.com]

So I decided to test the whole build, using `make -11 && make check`
in master/HEAD, and using `make -j11 check` in a previous commit where
it did work. The results:
- With ccache:
  - Before the changes (7c05b5cdd8a9cf2b9f80666a0be2cac7cf06cd16): 4m9.364s
  - After the changes (ef88e7443188275f14727a33997f43ab118c6bdf): 4m8.678s
  - Diff: -0.686 seconds (improvement)
- Without ccache:
  - Before the changes (7c05b5cdd8a9cf2b9f80666a0be2cac7cf06cd16): 4m20.767s
  - After the changes (ef88e7443188275f14727a33997f43ab118c6bdf): 4m30.805s
  - Diff: +10.038 seconds (slowdown)

So there isn't any noticeable speed up when every compiled object is
cached, and there is a 10 second slowdown when they aren't cached. I
guess this happens because the test objects are built sequentially,
where before they were parallel, but I don't know if there has been
any further changes in the build system between those 2 commits that
might have affected this.

On Wed, Nov 4, 2020 at 2:21 PM <rmrodriguez at carto.com> wrote:
>
> On Wed, Nov 4, 2020 at 1:48 PM Sandro Santilli <strk at kbt.io> wrote:
>
> > The recent change allowed to run all tests at once, so we need to
> > build all components at once, and if each of the components is built
> > in parallel, they end up building liblwgeom in parallel.
>
> And what's the benefit of running all tests at once? The CI builds
> seem to be taking the same time.
>
> Is this change worth all the troubles that it's causing? Maybe I'm not
> seeing how it improves things.
>
>
> --
> Raúl Marín Rodríguez
> carto.com



-- 
Raúl Marín Rodríguez
carto.com