[IPython-dev] [QUAR] Re: Qt SVG clipping bug (it's NOT a matplotlib bug)- Qt won't fix...

[Mark Voorhies]

On Thursday, October 14, 2010 08:27:10 am Michael Droettboom wrote:
> On 10/14/2010 10:52 AM, Robert Kern wrote:
> > On 10/14/10 3:55 AM, Hans Meine wrote:
> > It might be worthwhile for matplotlib to only use the SVG 1.2 Tiny standard for
> > greater compatibility. Tiny is much easier to implement, so I suspect there are
> > now more Tiny renderers now than there are Full renderers.
> >    
> This is true -- and we can probably remove some of the simpler-to-remove 
> parts that are missing from SVG Tiny.  However, to support clipping, we 
> either have to either a) write the routines to do clipping in vector 
> space or b) use rasterized fallbacks (as we do for some alpha issues in 
> Postscript, for instance).  a) is non-trivial in the general case, 
> particularly when accounting for line thicknesses, and b) is a hack.

Just to clarify what's happening on the iPython Qt console side:

* We receive SVG (with clipping path) from Matplotlib

* We wrap the SVG in Qt's rasterizer
  (via rich_ipython_widget._process_execute_payload
   calling svg.svg_to_image) and drop that as a widget on the
  console canvas.

+ As noted, the Qt rasterizer doesn't implement clipping, so
  we get unclipped plots drawn on the console.

+ This strategy also results in rasterized plots in the PDF export.

* For the context menu "Save Image As" and "Export HTML"
  functions, we call save(filename, "PNG") on the Qt
  rasterizer, so we get the same lack-of-clipping artifact.

* For the context menu "Save SVG As" and "Export XHTML"
  we work from Matplotlib's original SVG, so the clipping path
  is retained (and we get correct clipping in Inkscape, Firefox,
  and WebKit).

* Saving as PNG via Matplotlib's GUI as launched from iPython
  gives a correctly clipped PNG.  I'm not sure if Matplotlib is
  rasterizing directly, or drawing to the GUI canvas and then
  asking the GUI to generate a PNG.  Also, not sure which
  GUI backend I'm looking at...

Would it be reasonable to bypass the Qt issue by asking Matplotlib
for a PNG at the time that we receive the SVG and put that on
our console canvas?  It seems like this would be a good general
strategy for supporting non-Qt frontends (e.g., James Gao might
want something like this for the HTML frontend to support older
browsers).

--Mark