[PATCH] Big houses have too small territory radius

[elexis]

On a17 and a18 you could progressively expand your territorry with houses. While this still works with small houses (especially for the celtic ones), it doesn't work anymore with big houses (as of r16847), since the territorry ends somewhere inside of the house or just after the footprint. This can be a big disadvantage in comparison to celtic players. For example on maps where you can only extend with houses but not barracks, like snowflake searocks or some unknown land / unknown nomad maps (where all the wood is in the middle and CCs are surrounded by mountains).

big house in a18:

big house on svn

I'm not sure which revision has changed the territorry computation, maybe it was r16676?

An easy fix / workaround might be to slightly increase the radius for big houses.

[elexis]

Also I noticed that the territory of the houses seems to be misaligned.

Some more pictures showing the territory of unconnected houses:

[Stan]

Just Template Tweaking. I created a big house template, to make it easier in the future.

[Stan]

Radius of 24

[Stan]

Range of 23 as you can see it's messed up for some reason, which probably explains why little houses have less issues. So this is not likely to be a radius bug, but more a computational one.

[elexis]

Here the images uploaded by Stan:

House with radius 23

House with radius 24

We must find a solution that gives less territorry than the 24 solution (although that would still be better than what we currently have).

[elexis]

Fixes the actual cause of the problem.

[elexis]

Stan found out that it was caused by the change in r16751 line 425 of the CCmpTerritoryManager.cpp.

In a18, that radius value was 5, currently it is 2 in svn. By changing the radius in templates you can make it either 2 (way too small) or 3 (way too big), by typing something below 24 or bigger than 24 respectively.

If give that value a floatingpoint type, that value will become 2.5 (notice its exactly half of the previous value) and the territorry influence will be the way it was before!

[elexis]

We can't use float/double in the simulation as the values might slightly differ on different platforms and thus cause out-of-sync. Thus we continue to use u32.

[elexis]

Uses u16 for radius as that value should be smaller than 1000. Thus we can safely multiply it 2^{8 without causing overflows.}

[sanderd17]

Nice find, I'll review the patch with care tomorrow.

[sanderd17]

In 17122:

Introduce less rounding errors in the falloff to allow a more precise territory calculation. Fixes #3334. Based on code by elexis.

[sanderd17]

I went for an easier solution. This also gives the falloff a better precision, while less conversions are needed. Thanks for finding the cause of the issue.