Glitch export native linear gradient to EMF

[David Mathog]

About half of the exports of native linear gradient fills in rectangles to EMF were failing.

The attached test file grads.svg has many such rectangles. (Ignore the triangles.)

This glitch was breaking round trip testing of the libUEMF test_libuemf_ref.emf, which made it difficult to find any other new problems that code changes might introduce.

[David Mathog (mathog)]

The attached patch seems to resolve this bug. Once it is applied not only are the gradients exported as expected, but round trip
testing for EMF files works again, such that the first and third files, dumped via reademf, differ only in the expected places:

diff test1_emf.txt test3_emf.txt
1c1
< U_EMR_HEADER record: 0 type:1 offset: 0 rsize: 228 crc32:54AA5416
---
> U_EMR_HEADER record: 0 type:1 offset: 0 rsize: 228 crc32:D30A7175
13c13
< Desc. B: test1.emf
---
> Desc. B: test3.emf

[David Mathog (mathog)]

Patch committed as revision 14620.

[David Mathog (mathog)]

The mode of failure was that the rectangle record would be emitted, but the gradient fill was not. So the net result was a rectangle which was neither stroked nor filled, and so was invisible.

[su_v (suv-lp)]

On OS X 10.7.5:
- not reproduced with Inkscape 0.91 r13725
- not reproduced with 0.91+devel rev <= 14224
- reproduced with 0.91+devel 14228 <= rev <= 14619
- not reproduced with 0.91+devel r14620

Regression was in trunk only, and not in a stable release -> closing as 'Fix Released' without milestone.