Forester plants trees where probability to grow is nearly zero

It's probably better to add an attribute to the mountain terrains so that the worker's findspace program can filter for it.

See also https://bugs.launchpad.net/widelands/+bug/536155

The current consensus on the forum is to rank the fields by the highest probability to grow in the findspace program, e.g. by adding a parameter "grows:sapling". This should be configurable, e.g. in case we will want terrain affinity for fields in the future, and for modders.

If the player elects to build a forester to places where only marginally usable land is present, the player could be in desperate need of wood.

Should the forester tune the threshold dynamically: If there are both good and poor spots around, the forester would mostly use the good soil. If all land is bad, he would still do something instead of nothing.

I could do this, if skipping Release 20 is not a major upset to anybody. I might have time next week (or 6 months from now).

Nobody is working on this right now, so we'd be glad if you took care of this bug whenever you find the time :)

Would it be okay to assume that the tree growth probability of each spot remains constant over time?

The Probability to grow is calculated for each node from the values defined for each type of terrain. So they don't change over time (e.g. during a game, if you mean that).

At least as long the current implementation don't change.

Not true - terrains can be changed by scripting, e.g. the Atlantean scenario when the water rises, or the lava in Trident of Fire. While these 2 cases don't matter, the engine does allow any other terrain change, so we need to support it.

Didn't thought of scripting...

I shouldn't post to any tree related things :D

Is there a single piece of C++ that gets called when the terrain changes?

I now have a working implementation (if terrain changes are neglected). The code is so lousy that a complete rewrite is needed, though. In short: goal exists, and is easy to reach (but far away).

Attached a branch with a fix. Appears to work, but is not yet well tested even on my scale yet.

In the original bug description, a threshold was asked-for. Well, the fix is different. In the attached branch, the foresters prefers good soil, but occasionally also tries his luck with the less good one. In more details, the forester takes n free random spots from its work area, and uses the best of those.

There was a request in one of the duplicate tickets for non-equal performance between tribes, which is implemented here. Currently, Atlantean foresters are most careful and barbarian lumberjacks least. (technically: one can specify the n separately for each tribe).

The constants are not well-thought-of. I put a simple cache in there, as else the number of nested loops would have been big in my taste. As result, it is okay to use large values of n, without making planting too heavy. If any of you are in desperate need of more playing, try it out and give feedback here. Else, wait for the merge request button push.

Can you make a merge request? Then we will get a appveyor windows build. Then i will post link in the forum thread and our users using windows can test also your branch :-)

Now the appveyor builds are done.

No feedback on forum -> force-feeding to beta testers ;-)

Fixed in build20-rc1