Bug 83088 – TransformationMatrix::hasPerspective() should consider the entire bottom row of the matrix

• Status: RESOLVED
• Resolution: WONTFIX
• Component: WebCore Misc.

[Shawn Singh]

TransformationMatrix::hasPerspective() checks only the 4th row, 3rd column to see that it is non-zero.  I think it is more correct to consider the entire bottom row of the matrix when there is a perspective projection.

Take the following example:

  TransformationMatrix A;
  A.makeIdentity()
  A.applyPerspective(1);
  A.translate(50, 0);
  A.rotate3d(0, 90, 0);

The resulting matrix does have perspective, but the 4th row, 3rd colum element is zero.  Specifically that matrix has a bottom row of  (1, 0, 0, 1).  Technically, its -0.0 on the element, and hasPerspective() is likely to work, but it is probably not a good idea to rely on IEEE floating-point mechanisms to get this correct by luck.

Should we be testing the entire bottom row of elements, instead?  i.e. anything that may cause w != 1 should be considered a perspective projection.

Please let me know if you all agree, and I'll submit a quick two-line patch to fix this.

[Simon Fraser (smfr)]

When does this matter?

[Shawn Singh]

I have not seen any issue with this in real-world bugs.  But given that css will allow the web developer to enter any arbitrary matrix for an element, it seems appropriate that we should get this correct.

[Simon Fraser (smfr)]

Perhaps we should rename hasPerspective() to something more general if we change this.

[Shawn Singh]

Sure I could do that.  But in my opinion hasPerspective is still a very appropriate name.  What would you suggest instead?

other options are:  hasSkewAlongW() ...  willCausePerspectiveDivision() ... isBottomRowModified()

[Shawn Singh]

Also, I'll try to make a layout test that fails with existing code.  If I can't then this patch probably isn't worth doing.

[Shawn Singh]

This bug has been sitting here for a while;  I think it's WONTFIX for WebKit.  Thanks!