Re: Doubling performance of pkgsrc/mk

[Jonathan Perkin <jperkin%mnx.io@localhost>]

* On 2022-10-20 at 17:47 BST, Jonathan Perkin wrote:

> I got fed up with pkgsrc being slow again, so have spent a bit of time 
> recently improving performance.
>
> Here are some benchmarks for the build of pkgtools/url2pkg (chosen as 
> its "build" time is very short so exposes any delays with the 
> infrastructure):
>
>  +----------+---------+--------+-------+
>  | OS       | SmartOS | NetBSD | macOS |
>  +----------+---------+--------+-------+
>  | Baseline |   12.5s |  10.3s |  3.0s |
>  | Patched  |    5.7s |   5.3s |  1.4s |
>  +----------+---------+--------+-------+

As you'll have seen from pkgsrc-changes@, I've now committed the bulk of 
this work.  Please let me know ASAP if you see any fallout.  Obviously 
I've tested this a huge amount in bulk builds, but there can always be 
corner cases I'm not aware of.

The one change I've not committed yet is this one to gcc.mk, primarily 
because gcc.mk absolutely terrifies me:

  https://github.com/TritonDataCenter/pkgsrc/commit/05cd0dd3ebb79befae673e2d1a09e4cf5d6481ee

I'll probably keep this in my local tree for a while longer, but anyone 
who runs bulk builds may find it helps scan times quite a bit as you're 
no longer running gcc -dumpversion for absolutely no reason ~20,000 
times.  I personally see zero need for CC_VERSION_STRING and a bunch of 
other stuff in that file, but maybe others have strong opinions on that.

Happy compiling at 2x speed,

--
Jonathan Perkin   -   mnx.io   -   pkgsrc.smartos.org
Open Source Complete Cloud   www.tritondatacenter.com