Re: plot issues -> favor pdfcairo [new changeset]

[Ben Abbott]

On Tuesday, June 09, 2009, at 09:26AM, "Benjamin Lindner" <address@hidden> 
wrote:
>Ben Abbott wrote:
>> 
>> On Jun 7, 2009, at 3:13 PM, Benjamin Lindner wrote:
>> 
>>> Ben Abbott wrote:
>>>> On Jun 6, 2009, at 10:38 AM, Ben Abbott wrote:
>>>>> On Jun 6, 2009, at 7:49 AM, Benjamin Lindner wrote:
>>>>>
>>>>>> Ben Abbott wrote:
>>>>>>> On Jun 5, 2009, at 4:08 PM, Ethan Merritt wrote:
>>>>>>>> On Friday 05 June 2009 11:12:52 Benjamin Lindner wrote:
>>>>>>>>> Ethan Merritt wrote:
>>>>>>>>>> On Friday 05 June 2009, Ben Abbott wrote:
>>>>>>>>>>
>>>>>>>>>>>>> Petr, what are the benefits of the pdfcairo and pngcairo 
>>>>>>>>>>>>> terminals
>>>>>>>>>>>>> over the pdf and png terminals?
>>>>>>>>>>
>>>>>>>>>> Aside from licensing issues for PDFLib, using cairo to generate 
>>>>>>>>>> the plots
>>>>>>>>>> allows antialiasing, transparency, and UTF-8 support.
>>>>>>>>>
>>>>>>>>> This may be a sutpid question, but what should a vector-based 
>>>>>>>>> graphics
>>>>>>>>> description support antialiasing for?
>>>>>>>>
>>>>>>>> Ben asked about both pdfcairo and pngcairo.
>>>>>>>> The anti-aliasing is an issue for png, not for pdf.
>>>>>>>> Conversely, the transparency support is an issue for pdf but not 
>>>>>>>> for png.
>>>>>>>>
>>>>>>>>> pdfcairo might be superior if you require transparency and UTF-8,
>>>>>>>>> granted, but the quality of the generated output is disappointing
>>>>>>>>> compared to pdf via postscript.
>>>>>>>>> I do a lot of image plots and found that the resulting file 
>>>>>>>>> sizes with
>>>>>>>>> the pdfcairo terminal are 4-8 times larger than a ps->pdf 
>>>>>>>>> output. Also
>>>>>>>>> you don't have good control over font selection, which is IMO a
>>>>>>>>> knock-out criteria when doing high-quality plots for e.g. latex 
>>>>>>>>> inclusion.
>>>>>>>>
>>>>>>>> All I can say is that I have had the opposite experience.
>>>>>>>> Maybe that's because I work in a UTF environment and need support
>>>>>>>> for CJK character sets.  PostScript is basically hopeless for those.
>>>>>>>> There are some very fragile workarounds, but they are so 
>>>>>>>> installation-
>>>>>>>> specific that it doesn't work to build scripts or work flow 
>>>>>>>> around them.
>>>>>>>>
>>>>>>>> For latex inclusion, ps2pdf or direct PDF generation should be 
>>>>>>>> exactly
>>>>>>>> the same, and subject to the same limitations of whatever converted
>>>>>>>> Computer Modern fonts you are using.  If that is a primary concern,
>>>>>>>> then using one of the latex-based terminals directly is a better 
>>>>>>>> bet.
>>>>>>>>
>>>>>>>>  Ethan
>>>>>>> It doesn't appear to me that there is one solution that is 
>>>>>>> preferred over the other in all cases.
>>>>>>> I'll propose the following, and encourage all to comment.
>>>>>>> -----------
>>>>>>> 1) if "pdfcairo" is present => use "set term pdfcairo ..."
>>>>>>> 2) if "pdf" is present => use "set term pdf ..."
>>>>>>> 3) if neither => use "set term postscript ..." and then convert 
>>>>>>> using ghostscript
>>>>>>> -----------
>>>>>>
>>>>>> Sounds very reasonable to me.
>>>>>>
>>>>>>> Although I'm a big LaTeX user, I elevated pdfcairo over pdf for 
>>>>>>> the transparency feature.
>>>>>>> Similarly for png
>>>>>>> -----------
>>>>>>> 1) if "pngcairo" is present => use "set term pngcairo ..."
>>>>>>> 2) if "png" is present => use "set term png ..."
>>>>>>> 3) if neither => use "set term postscript eps ..." and then 
>>>>>>> convert using ghostscript
>>>>>>> -----------
>>>>>>
>>>>>> same as above
>>>>>>
>>>>>>> For LaTeX, I will soon be adding support for the Lua/TikZ 
>>>>>>> terminal. Its rendering is a bit slow, but produces excellent 
>>>>>>> results for both latex and pdflatex.
>>>>>>> The solutions above will only work well for gnuplot 4.3+. Prior to 
>>>>>>> that the variable GPVAL_TERMINALS does not exist, and octave has 
>>>>>>> no manner to check for the existence of specific terminals.
>>>>>>
>>>>>> For windows platform this is OK, since a working console 
>>>>>> application is only possible with 4.3+
>>>>>> Also interactive zooming with piped gnuplot works only with 4.3+
>>>>>>
>>>>>> benjamin
>>>>>
>>>>> Great!
>>>>>
>>>>> I've attached a changeset for the pdf part. It is combined with a 
>>>>> changeset for forcing mono rendering (there is some problem with 
>>>>> rgb2gray that has delayed me in pushing that change).
>>>>>
>>>>> In any event, I'd appreciated confirmation form linux and window's 
>>>>> users that this chageset works correctly.
>>>>>
>>>>> Ben
>>>> I've rolled up all the changesets for these plot issues. The attached 
>>>> must be applied to the current sources. With this change applied, the 
>>>> print command will favor the cairo terminals and will properly render 
>>>> a gray-scale image when the -mono option is specified.
>>>> I'd appreciate some testing to make sure there are no surprises.
>>>> I also  noticed we've been on bug list. I've moved this over to the 
>>>> maintainers list (I should have done that many emails ago).
>>>
>>> Hmm, I tried to import your changeset, but it fails for all hunks.
>>> My tip is 9305:52b4d82e5b4f from http://www.octave.org/hg/octave
>>>
>>> benjamin
>> 
>> I did find a problem with the prior changeset (new bug), but I don't 
>> understand why it wouldn't apply for you. In any event, try the attached 
>> version.
>
>Argh, stupid CRLF line endings mess.
>Got the changeset to apply.
>However, it does not work as expected.
>
>I have a gnuplot version, which has the pdfcairo and pngcairo terminals 
>and has the png terminal.
>Printing to pdf as "print -dpdf" now still invokes ghostscript.
>I believe I can guess where the problem is.
>
>       available_terminals = __gnuplot_get_var__ (gcf, "GPVAL_TERMINALS");
>       available_terminals = regexp (available_terminals, "\\b\\w+\\b", 
>"match");
>       gnuplot_supports_term = any (strcmp (available_terminals, termn));
>
>This yields for the gnuplot under test a cell array available_terminals 
>which includes a terminal "pdfcairo" but does *not* include a terminal 
>"pdf". Thus a "print -dpdf" will yield gnuplot_supports_term=false, and 
>then the check for the cairo terminals is never done.
>
>I guess the same will happen if gnuplot supports pngcairo but does not 
>support the png terminal.
>
>The decision logic is the wrong way round.
>
>benjamin
>

Thanks for the feedback. I'll fix this later today.

Ben