[Gtkradiant] Reached milestone in q3map2 math precision fixes

[Nerius Landys]

Hi guys, I've committed some code back to trunk pertaining to q3map2 fixes.
See commit r416.  The changes are not dangerous at all because almost all of
the code is inactive due to #defines in q3map2.h.  So you have have to turn
the changes on yourself by setting these to 1 in q3map2.h:

#define EXPERIMENTAL_HIGH_PRECISION_MATH_Q3MAP2_FIXES   1
#define EXPERIMENTAL_SNAP_NORMAL_FIX                    1
#define EXPERIMENTAL_SNAP_PLANE_FIX                     1

Right now all 3 of these are set to 0 in SVN (disable).

Here are some results from running regression tests that are in trunk:

"r417 w/ defines" means you turn on
EXPERIMENTAL_HIGH_PRECISION_MATH_Q3MAP2_FIXES, EXPERIMENTAL_SNAP_NORMAL_FIX,
and EXPERIMENTAL_SNAP_PLANE_FIX in q3map2.h.

regression test         | r417                    | r417 w/ defines
------------------------+-------------------------+------------------
base_winding            | working                 | working
coarse_snap_normal      | broken                  | working
degenerate_winding      | working                 | working
disappearing_sliver     | working                 | working
disappearing_sliver2    | working                 | working
disappearing_sliver3    | broken                  | working
duplicate_plane         | working                 | working
plane_aliasing          | working                 | working
segmentation_fault      | working                 | working
snap_plane              | broken                  | working
sparkly_seam            | working                 | working


Also, I was able to compile Icy Fantasy map (a very complex map with lots of
thin triangles) without any disappearing triangles.  I used to have to use
various tricks such as deleting brushes and replacing them with individual
path mesh triangles.



Bottom line: If you've had problems with disappearing terrain or triangles
in your maps, give these changes a try.  I'd like to get some feedback.

The majority of the improvements have to do with the brush winding
calculations.

Maybe if we feel confident enough that these changes don't break anything we
can flip the #defines to "on" by default sometime in the future.

If you have any other bugs in q3map2, I think we should get a regression
test for it.  Since q3map2 code is somewhat heuristic, the only way to be
confident that it works properly is to have regression tests.

- Rambetter
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://zerowing.idsoftware.com/pipermail/gtkradiant/attachments/20110109/246fe5cc/attachment.htm