Re: Benchmarks for graphics

[Andreas Weber]

Am 11.03.2015 um 16:27 schrieb Rik:
> On 03/11/2015 03:24 AM, address@hidden wrote:
>> Le 10/03/2015 01:43, Ben Abbott a écrit :
>>>> On Mar 9, 2015, at 6:08 PM, Pantxo Diribarne
>>>> <address@hidden> wrote:
>>>>
>>>> Le 09/03/2015 22:57, Pantxo Diribarne a écrit :
>>>>> Le 09/03/2015 20:54, Andreas Weber a écrit :
>>>>>
>>>>>> Am 09.03.2015 um 09:13 schrieb Pantxo Diribarne
>>>>>>
>>>>>>> During the last months I found that this page [1] was really helpful
>>>>>>>
>>>>>>> in order to compare plots with older versions of Octave. Could we run
>>>>>>>
>>>>>>> the script that generated those plots  and make a 4.0 benchmark. I'd
>>>>>>>
>>>>>>> be happy to help on this (I have Matlab 2013a at work) but need
>>>>>>>
>>>>>>> instructions: where do I find the necessary scripts? How do I run them?
>>>>>>>
>>>>>>> @Andy: will you be able to host the resulting plots on the same page?
>>>>>>>
>>>>>> Hi Pantxo,
>>>>>>
>>>>>> the used scripts are in octave-src/scripts/testfun/private/
>>>>>>
>>>>>> dump_demos creates a script with the demo blocks which can be run with
>>>>>>
>>>>>> FLTK, gnuplot and Matlab,
>>>>>>
>>>>>> html_compare_plot_demos fills the result to the html template.
>>>>>>
>>>>>> Btw, do you know http://wiki.octave.org/Plot_comparison?
>>>>>>
>>>>> Yes, I even edited it once :-) but my memory is more or less in my
>>>>> bookmarks table, in which I forgot to put that page.
>>>>>
>>>>>> I can run the scripts for FLTK and gnuplot on 4.0-rc1 but haven't access
>>>>>>
>>>>>> to Matlab (I'm even unsure if the matlab scripts would run after the
>>>>>>
>>>>>> last changes which removed for example sombrero) and yes, I can host
>>>>>>
>>>>>> them on tech-chat.de
>>>>>>
>>>>> Thanks, I generated the script and will try to run it on Matlab soon.
>>>>>
>>>>>> -- Andy
>>>>> Pantxo
>>>>>
>>>> I forgot, if we are to generate the plots separately we should probably
>>>> use the very same script. I attached the script generated from the
>>>> current dev although it should be identical with 4.0-rc1.
>>>>
>>>> Pantxo
>>>>
>>>> <dump_plot_demos.m>
>>> Wow!. Lots of regressions since the last time this was run on Matlab.
>>>
>>> Some simple things need to be fixed.
>>>
>>> (1) In Matlab "!' is an alias for "system".
>> I'll take care of those.
> I won't finish it for this release, but I was in the process of writing a
> very simple oct2mat function that is to be part of the dump_demos script. 
> Replacing '!=' with '~=', 'endif' with 'end', etc.  is very simple with
> regular expressions and would allow us to write all of our demos in Octave
> syntax.  Currently, you just have to know that demos written in the plot
> directory should be written in Matlab format.  Anybody who is not a core
> dev usually misses this fact and so the demos have become less Matlab
> compatible over time.
Looks like there is a regression which stopped replacing the " with ' on
default. I'll dig into this later

>
>>> (2) Matlab recently introduced the "assert()" function, which is not
>>> compatible with Octave's. In general tests should be avoided in demo
>>> scripts.
>> Agreed, those should be turned into proper tests or removed
> Agreed.
It's more than a year since we agreed to remove asserts from demos.
Should we start doing it or is there anything new we should consider first?
Before b8934a57e13e we added an empty assert (which does nothing).

>>> (4) Matlab does not have a sombrero() function. The demos should avoid it.
>> Many demos use it and it's like Octave's second logo after all :-). I'd
>> vote for keeping them.
>> @Andy: you talked about sombrero being removed. Is there a reason for not
>> having a Matlab compatible version with the other utility scripts?
The part which copied and modified sombrero was removed in b8934a57e13e
but I don't know the reason
> Let's keep it.  There's no need for us to twist our code into knots just to
> occasionally run plot comparisons.  The sombrero function is very small and
> disk space is cheap anyways so I don't see an issue with keeping a Matlab
> utility version around.
>
>

-- Andy