Improve handling of <canvas> path operations on an empty path.

[Dean McNamee]

There was originally some work on this:

https://bugs.webkit.org/show_bug.cgi?id=27187

This patch further does a few more things:
  - Don't emit a line or curve, just set the current point (moveTo).
  - For curves, emit the end point and not the control point.
  - Update the Skia backend to implement hasCurrentPoint.

[Dean McNamee]

Created attachment 32772 [details]
Patch

[Oliver Hunt]

Comment on attachment 32772 [details]
Patch

In general this seems okay to me -- how does the output compare to firefox?

[Dean McNamee]

Differs from Firefox, looks like the control point vs endpoint, but I really don't see how control point could ever make any sense.  I'll look at their code today, and maybe file a bug with them.

(In reply to comment #2)
> (From update of attachment 32772 [details])
> In general this seems okay to me -- how does the output compare to firefox?

[Dean McNamee]

(In reply to comment #3)
> Differs from Firefox, looks like the control point vs endpoint, but I really
> don't see how control point could ever make any sense.  I'll look at their code
> today, and maybe file a bug with them.
> 
> (In reply to comment #2)
> > (From update of attachment 32772 [details] [details])
> > In general this seems okay to me -- how does the output compare to firefox?

Ok, so Firefox just passes this directly to cairo:

162 void
163 gfxContext::LineTo(const gfxPoint& pt)
164 {
165     cairo_line_to(mCairo, pt.x, pt.y);
166 }

And Cairo has the control point behavior:

http://cairographics.org/manual/cairo-paths.html#cairo-curve-to

If there is no current point before the call to cairo_curve_to() this function will behave as if preceded by a call to cairo_move_to(cr, x1, y1).

I think they should cairo_has_current_point() and cairo_move_to the endpoints if not.  I'll file a bug with Mozilla.  I wonder why Cairo has this behavior.

[Dean McNamee]

Discussion of the cairo behavior is here:

http://lists.freedesktop.org/archives/cairo/2008-June/014262.html

I guess the reason cairo behaves like it does, is because it still wants to emit a curve.  I think the current behavior for browser should be to not emit anything, and just do the moveTo.  I will follow up with Mozilla.

(In reply to comment #4)
> (In reply to comment #3)
> > Differs from Firefox, looks like the control point vs endpoint, but I really
> > don't see how control point could ever make any sense.  I'll look at their code
> > today, and maybe file a bug with them.
> > 
> > (In reply to comment #2)
> > > (From update of attachment 32772 [details] [details] [details])
> > > In general this seems okay to me -- how does the output compare to firefox?
> 
> Ok, so Firefox just passes this directly to cairo:
> 
> 162 void
> 163 gfxContext::LineTo(const gfxPoint& pt)
> 164 {
> 165     cairo_line_to(mCairo, pt.x, pt.y);
> 166 }
> 
> And Cairo has the control point behavior:
> 
> http://cairographics.org/manual/cairo-paths.html#cairo-curve-to
> 
> If there is no current point before the call to cairo_curve_to() this function
> will behave as if preceded by a call to cairo_move_to(cr, x1, y1).
> 
> I think they should cairo_has_current_point() and cairo_move_to the endpoints
> if not.  I'll file a bug with Mozilla.  I wonder why Cairo has this behavior.

[Dimitri Glazkov (Google)]

Landed as http://trac.webkit.org/changeset/45973.