AGG - Polygon "outline" is drawn when outline is not specified

[havatv]

When composing polygon symbols using a background colour and a foreground pattern in a different colour using AGG, there will be a 1 pixel outline in the background colour also where the border passes through the foreground pattern. This is not correct behaviour - outlines shall be specified explicitly.

Example (only rectangular polygons, but the same effect can be seen for all polygons): ​http://gis.umb.no/cgi-bin/mapserv_5.2dev1?map=/users/ikf/ikfht/mapserv/tests/polyfill_whitespace.map (you have to zoom out a bit to see all the borders...)

Here is an example of how this is specified in LAYER-> CLASS (white pattern on yellow background):

     STYLE
        COLOR 255 205 114
      END # style
      STYLE
	symbol "halvapent"
	SIZE 30
        COLOR 255 255 255
      END # style

[tbonfort]

see also #2749

[tbonfort]

removed the outline drawing in r7974, (with all my apologies to the coast of Norway of #2749 :) )

this seems the correct way to do this, and users can add the outline back in by adding an outlinecolor + width to achieve the same effect

I'm wondering how many "my polygons are disjoint" messages we'll be getting on the mailing-lists :)

[tbonfort]

sample content rendered before the patch

[tbonfort]

sample content rendered after the patch

[havatv]

Map file for the example pattern

[havatv]

I tried to test this with the nightly build now, but the colours went away for some reason. I might have some problems with my mapfile - I have included it as an attachment.

Here is the same example with the nightly version:

​http://gis.umb.no/cgi-bin/mapserv_5.3dev?map=/users/ikf/ikfht/mapserv/tests/polyfill_whitespace.map

It seems as if the outlines are still there (in very light gray).

[tbonfort]

the color disappearing is a side effect of the SIZE/WIDTH parameters passing to doubles (along with using symbolscale)

as for the outlines you see, that's what I was referring to in

I'm wondering how many "my polygons are disjoint" messages we'll be getting on the mailing-lists :)

see ​http://thread.gmane.org/gmane.comp.graphics.agg/2359/focus=2367 for an explanation

[havatv]

I think I understand the effect of anti-aliasing that you describe.

My problem here is that I do not see the fill colour (the polygon "inside" - should be yellow in my case) when using the binary from the nightly (neither with AGG or GD)... I am using Firefox 3.0. 5.2 gives me the fill.

You can try the links provided above and see for yourself (AGG output).

[tbonfort]

Replying to havatv:

My problem here is that I do not see the fill colour (the polygon "inside" - should be yellow in my case) when using the binary from the nightly (neither with AGG or GD)... I am using Firefox 3.0. 5.2 gives me the fill.

that's the part that i was referring to for the size/width bug when passed to doubles... will fix tonight

[brage]

Replying to tbonfort:

removed the outline drawing in r7974, (with all my apologies to the coast of Norway of #2749 :) )

this seems the correct way to do this, and users can add the outline back in by adding an outlinecolor + width to achieve the same effect

I'm wondering how many "my polygons are disjoint" messages we'll be getting on the mailing-lists :)

I belive the problem is that mapserver is rendering the shapes on by one, so that the antialiasing is against the background instead of against the neighbouring shapes within the same layer. From what I can judge from the rather sparse agg documentation and examples, a possible solution might be to render all the shapes in a layer collectively with agg::render_scanlines_compound rather than individually with agg::render_scanlines. Compare the output of flash_rasterizer.cpp vs flash_rasterizer2.cpp on ​http://www.antigrain.com/demo/. However - I have no experience with agg, so I cannot say if this really is the solution for mapserver.

[tbonfort]

the problem with the compund rasterizer is that it doesn't fit well with the mapserver way of drawing features:

we loop through all geometries, each time rendering it with its style(s)

this is incompatible with the compound rasterizer, which requires you to accumulate geometries, and then render the entire set with a *common* style

[havatv]

Replying to tbonfort:

OK, I will test again in a couple of days.

Here comes some wild thoughts on the polygon rendering "problems" by someone who is not very much into rendering...

I guess there is not a way to turn off anti-aliasing in AGG for polygon boundaries (no anti-aliasing would normally be preferable for polygon networks, but perhaps not for isolated polygons). Could it be possible to achieve no anti-aliasing for polygons by buffering the polygon by some small distance before rendering, and then clip using the original polygon after rendering (similar to techniques for continuous tile rendering)?

Or, would it be possible to select AGG or GD rendering on a layer-by-layer basis? Then, AGG could be used for lines, points (and isolated polygons), while GD could be used for polygon networks...