printing figures with development version

[Ben Abbott]

On Apr 1, 2009, at 5:16 PM, Rik wrote:

>>
>> ------------------------------------------------------------------------
>>
>> Subject:
>> Re: printing figures with development version
>> From:
>> Ben Abbott <bpabbott at mac.com>
>> Date:
>> Sat, 28 Mar 2009 13:38:38 -0400
>> To:
>> "John W. Eaton" <jwe at octave.org>
>>
>> To:
>> "John W. Eaton" <jwe at octave.org>
>> CC:
>> octave-maintainers at octave.org
>>
>> On Mar 28, 2009, at 12:17 PM, John W. Eaton wrote:
>>
>>> On 27-Mar-2009, Ben Abbott wrote:
>>>
>>> | What you see is intended. You'll need to adjust the figure's paper
>>> properties. Specifically, "papersize" and "paperpositon"
>>> |
>>> | If you'd like an 6.4in x 4.8in eps file
>>> |
>>> | figure (1)
>>> | clf
>>> | plot (randn (50, 1))
>>> | set (gcf, "paperunits", "inches"); # the default is "inches"
>>> | paper_size = [6.4, 4.8];
>>> | set (gcf, "papersize", paper_size)
>>> | set (gcf, "papertype", "<custom>")
>>> | set (gcf, "paperposition", [0, 0, paper_size])
>>> | print (gcf, strcat ("figtest", version, ".eps"), "-depsc")
>>> |
>>> | This will be much simpler once the all listeners for these
>>> properties are in place.
>>>
>>> If I print to a .eps file, I expect that the bounding box of the
>>> figure will be fairly tight around the figure so that it may be
>>> included directly in another document without a lot of extra
>>> whitespace surrounding it.  I think that is most useful.  I also  
>>> don't
>>> expect to have to fiddle with paper properties to make that happen
>>> (how would I know that 6.4x4.8 inches are good values?).
>>>
>>> Earlier versions of Octave did not add extra whitespace around the
>>> figure, so I think we will see a lot of complaints about this change
>>> if it escapes into a stable version.
>>>
>>> What is the motivation for the change?  Is there something good  
>>> about
>>> this design choice that I'm missing?  If the change is purely for
>>> Matlab compatibility, then I think this is one of those things  
>>> that we
>>> should not copy.
>>>
>>> jwe
>>
>>
>> The current implementation produces compatible behavior ... except
>> that the BoundingBox defined by gnuplot is not tight, and the one
>> produced by Matlab is.
>>
>> The current implementation may be modified to obtain the desired
>> result by changing the papersize for the eps output to ...
>>
>>    papersize = paperposition + [50, 50, 100, 100]/72;
>>
>> Where the paperunits are implied to be "inches". The 50pt border is a
>> gnuplot feature.
>>
>> I'll prepare a changeset.
>>
>> Ben
>>
> Per John's observation, I think this will generate a lot of complaints
> from people used to the old style.
>
> Of more immediate concern to me, there are 22 figures in the Octave
> documentation that are auto-generated by Octave code.  The output  
> format
> is png for the HTML documentation and pdf for the pdf documentation
> (Duh!).  Under the old code these figures slipped into the text nicely
> but they now have about 1/3 of a blank page above and below them.   
> If it
> is certain that this is the new format for the print command I will
> re-code the figure generation scripts to include all of the
> papersize-setting commands above.  If there is an option to flip
> Matlab-compatability on or off I would prefer to use that and not  
> change
> the scripts.
>
> Cheers,
> Rik

Rik,

I certainly get all the credit for this problem. I was focused on the  
goal of compatibility with Matlab and didn't consider the documentation.

In any event, I think the best solution would be one that is  
compatible with Matlab, and is simple to implement.

My first thought was to run the lines below prior to producing the  
figures for the documentation.

	image_size = [6.4, 4.8]; % in inches
	border = 0; % For postscript use 50/72
	set (0, "defaultfigurepapertype", "<custom>")
	set (0, "defaultfigurepapersize", image_size + 2*border)
	set (0, "defaultfigurepaperposition", [border, border, image_size])

This *should* produce pdf files at 6.4x4.8 inches.

Unfortunately, these defaults are not passed onto the figures :-(

Even so, as the defaults not set for the figure is a bug, I recommend  
this solution.

Rik, do you know enough c++ to take care of setting the figure's  
default properities? (I don't).

Ben