- drawing face normals was broken (using wrong index).
- drawing flat faces in solid mode would display ugly tesselation normals (as if the model was made of triangles).
- drawing smooth faces in solid mode would show vertex normals based on tesselation (heavily slant in 1 direction).
now the normals are calculated and stored per polygon (will save some memory too for non tri meshes).
- in the case of deform modifiers this is very safe (assuming the mods themselves dont need tessface data), since the DM is freed right after, so this is an easy speedup.
- in the case of bevel and split edge modifiers this should still work out fine since mods that need tessface data will generate it.
also added debug define to help solve invalid values with valgrind's memcheck.
gives noticeable speedup since there was a loop no all faces to set index values on redraw.
- went over every BM_SetIndex call and added comments about its use, if its setting dirty values or correct ones.
- replace loops over all elements _just_ to set index values, with a new function, BM_ElemIndex_Ensure(...), this will eventually only initialize values when they are dirty.
For some cases, this will make dissolve work. For others dissolve will still fail (report unable to dissolve, not crash) in some cases. To get dissolve to work in these cases we need to define what happens when you dissolve a disk that has faces which share more than one edge.
1. Removed pinning, at least until after trunk
reintegration (it's a usefull feature, but
incomplete).
2. Ripped out the subclassing code I originally
wrote so bmesh would be more usable for modifiers
(I ended up improving CDDM instead).
3. Inlined some of the macros in bmesh.h, still need to
do the macros in bmesh_operator_api.h.
4. Removed the BMESH_ITER_*** macros (except for BMESH_ITER_INDEX).
They were too clunkly to be worth it.
Multires interpolation now works on cases like
simple cubes. (though it isn't perfect
when bevelling sharp corners).
Normal flipping is handled correctly, and
multires interpolation now works on
normal-inconsistent meshes (so long as
they are manifold, at least).
Multires interpolation is considerably better
now, though it still has a problem with occasionally
producing little random tangent spikes. Still, it's
far better then it was.
Also fixed a bug in dissolve faces.
Implemented a new "super knife". Activate with k. Holding CTRL
will allow extended cutting ala old lines mode. Confirm with enter
and escape. You cannot cancel, btw, you can only confirm (and undo
later if you want). Hopefully I'll support undo within the tool soon.
* Supports cutting edges, into faces, etc. You can pretty much do whatever
you want. Will snap to vertices too.
* Note that if you cut into a face, it must be valid topologically when
you press enter to confirm.
* It's pretty and graphical :)
* You can only cut visible geometry.
* UVs/vcols are a little buggy still
Now, thou shalt all cease and desist all lack of motivation for
testing! No longer shall users put off testing until "it's cooler"!
:P
Phase 1 of restructuring done. There are now two
distinct subclass systems within the bmesh API;
one is compile-time, and forms the backend of what will
eventually be a "lite" bmesh API for modifiers (the
ones that use bmesh are simply too slow right now).
The other is dynamic, and will be used to implement
multires reprojection.
The idea was to solve as many serious problems with
memory, speed, etc, at once as possible and set up others
to be solved more easily later.
I've also added holes into the data structure, but not
the api; I don't plan to finish implementing that until
after bmesh gets into trunk. I simply wanted to lessen
how much code I'll have to rewrite, since I was doing a
fairly major restructuring anyway.
In addition, I've added iteration support to mempool, to
avoid having to store linked list pointers (though this
has caveats).
Next step: merge in trunk changes. . .oh, what fun :P
