Re: feedback for the xlsx import improvements

Hey,

On Tue, Sep 6, 2016 at 9:10 PM, Maarten Bosmans <mkbosmans@gmail.com> wrote:

> 2016-09-01 2:59 GMT+02:00 Markus Mohrhard <markus.mohrhard@googlemail.com
> > :
> > For the bug fix this would be  a simple test document that can be used in
> > one of the automated tests.
>
> I'm looking into making a test case for the two bugfixes.

Great.


> > For the performance improvement it would be a
> > document showing the problem but not taking too long. The perf documents
> are
> > run under callgrind which means they take around 50 to 100 times the
> normal
> > import time during the test. So ideally the document already shows the
> > problem but still opens in less than 30 seconds.
>
> For the performance, I used the document attached to the bugreport
> (tdf#100709) and converted using `soffice --convert-to ods`.
> On my computer conversion took about 6 seconds, of which roughly 40%
> import, 40% export and 20% startup/other. After the perf improvement,
> import time is only a couple of percent, and total time is reduced to
> 3-4s.
> Would this document be suitable as a performance regression test as
> is, or should I try to make it smaller?

A similar file created from scratch. Taking a document that you have not
created is normally not correctly licensed. However a document that just
takes 6 seconds is already good and a good template that you can use to
design your document.


> Also, it would be nice to be able to only measure import, without the
> export, as that more clearly shows the gain of over 10x, instead of
> just the 2x reduction in runtime of the whole conversion. Is this
> possible within the performance testing framework of LibreOffice?


Yes. Of course. For the performance tests we only need the document and
I'll add it to the list. The files are in another repository and the
configuration for defining how the files are handled is not that obvious.
The results will appear on perf.libreoffice.org after adding the file.

Markus

> Maarten