Hatch clipping

[pelson]

Fixes a genuine bug with hatch clipping in Agg & tidies up some of the _backend_agg.cpp code.

Note the change from typedef agg::scanline_p8 to typedef agg::scanline_u8 has had no noticeable detrimental impact (p==packed, u==unpacked, see http://www.antigrain.com/doc/scanlines/scanlines.agdoc.html#toc0001).

There are some changes in here which could fall into v1.2.x, but given our targets for v1.3 and the fact that I am actively working on _backend_agg.cpp, I think it is fine for it to go to master only.

[pelson]

I've been using this to measure performance of any changes I'm making (via the backend driver), so thought I would include it here.

[mdboom]

Thanks. I just merged the Travis/inkscape fix into master, so I don't think you have it on your branch. Can you rebase on master so that Travis can run?

From the linked document, I can see how there are different performance implications of packed vs. unpacked, but how was that causing the hatched rendering bug? (Just for my own edification, really, nice to have a solution in any event).

All the cleanups etc. look good.

[pelson]

Note the change from typedef agg::scanline_p8 to typedef agg::scanline_u8 has had no noticeable detrimental impact (p==packed, u==unpacked, see http://www.antigrain.com/doc/scanlines/scanlines.agdoc.html#toc0001).

Sorry, somebody asked me a question as I was typing this... I meant to finish the sentence with "the change was not necessary to fix the hatching bug".

[mdboom]

Ah -- so what is the purpose of the change from packed to unpacked? (Other than it being simpler...)?

[pelson]

Ah -- so what is the purpose of the change from packed to unpacked? (Other than it being simpler...)?

It is necessary to start looking at compound rasterization (which was what I started doing, before spotting this). Its all a little incestuous, but as far as I can tell, these changes also seem to make a minor performance increase.

[mdboom]

Ok, no worries, then.

[pelson]

@mdboom - thoughts?

[mdboom]

There are a bunch of seemingly legitimate test failures on Travis that all seem to have clip paths in common, so I think perhaps something related to clip paths is broken here. Unfortunately, I can't reproduce the problem locally when installing this branch, so it's anybody's guess at this point. (This is where getting the build products from Travis would be extremely useful, but unfortunately, that's not possible).

[pelson]

Ah, ok. I'll try to investigate on my own VM which emulates the travis environment and have a look into this.