Re: [Libreoffice-qa] minutes of ESC call ...

Hey Michael, Keny, Eike,

On Thu, Nov 26, 2015 at 5:51 PM, Michael Meeks <michael.meeks@collabora.com>
wrote:

>             + calc s/w interpreter-related patch
>                 [ also got in before the freeze; an 18x speedup on CPU for
>                   some test sheets; currently guarded by a variable - and
>                   subsetting to just some sheets.
>                      + plan to test with crash-testing sheets & enable by
> default
>                   Would like to look at it (Eike)
>                      + dislike env.vars set - avoids the unit-tests
>                         + plan to remove it before ship (Kendy)
>                             + turn into normal config & default ->
> testable.
>                         + would like to turn it on now - default on for
> user-testing (Kendy)
>                             + optional B2 in the release-plan (Cloph)
>                                 + can have an intermediate release.
>                      + goal of the code - vectors of doubles
>                         + formula results (Michael)
>                             + moves big chunk of ptr chasing & branching
> from inner loop.
>                 ]


please add a performance test that ensures that future refactorings won't
regress here. In general any patch claiming performance improvements should
be accompanied by a performance test to make sure that we can still
refactor the code without someone complaining that we introduced huge
performance regression.

Related to the code I wanted to mention that the getenv calls should be
cached in a static variable as they might become expensive when called
repeatedly and that the new virtual method calls are not free so I would
like to see some before and after numbers for both the supposed 18x speedup
case and our normal matrix handling code.

Regards,
Markus