RE: [Flightgear-devel] New texture draw method

[Norman V. <nh...@ca...>]

Curtis L. Olson writes:
>
>Norman Vine writes:
>> This is a problem when the objects live in different 'tiles'.  
>> 
>> But don't all scenery objects in a 'tile' share the same transform ??
>
>In theory, but airports like KSFO and KBOS span the boundaries of
>multiple tiles ... KBOS is on a corner for instance so it hits 4
>tiles.

He he....  gottcha  :-)

However it still seems to me that a float should be adequate
Its been awhile since I have done the math but I think
that with 23 bits of precision < IEEE FLOAT > we should
get a max resoultion of 2**24 divisions per tile
< one extra because the sign bit counts too >

CM_res = pow(2,24) = 16,777,216 divisions per tile

therefore
max tile dimension  for one centimeter accuracy  =
res * .001 * .01 = 167.7 km max tile dimension

I believe that is 'quite' a bit larger then any of our tiles 
greatest dimension so we should be able to get sub 
centimer accuracy by storing our data in float format

To exploit this means we need to determine the furthest 
point in each tile from the tile's center point and then 
scale it to be 2**23
< radal vector 1/2 total distance so lose the extra bit >
then scale all the other points respectively. In effect
this would be an additional local 'tile' transform to get 
the best possible accuracy.

My guess is that would help considerably during
the construction phase, while only complicating things 
a little, and if done there then this transform wouldn't 
be necessary in the final 'product' as the shared neighbor
points will have been placed more acurately.

Now all we need is for someone to test this :-))

Cheers

Norman