[Plib-devel] Large geometry objects and crash in ssgLOS

[Paolo L. <p.l...@ci...>]

Dear friends,

I use to load CAD-originated 3DS files containing quite large geometry
objects (e.g. #vert 26K, #faces 53K) - no problem in ordinary =
visualization,
yet I get a crash at ssgLOS call showing a problem in accessing =
vertex/index
arrays at signed short int boundary (i.e. after 32767).

Beyond the general recommendation by Steve not to have large objects for =
an
efficient frustum culling, the problem seems, infact, due to a general
unclarity about what the limit for a geometry object actually is. In
ssgXxxArray (Vertex, Normal, Index, ...) get/set methods accept unsigned =
int
arguments, the add method is not limited anyhow, yet e.g. =
ssgIndexArray::get
returns a pointer to short.

In the 3DS loader all the index elements are pointers to unsigned short, =
and
that's what contributes to fill ssgVtxArray's. Don't know about other
loader's policy on the topic.

Currently I locally patched the 3DS loader to split geometies in 10K
triangle chunks, since the 32K limit would strangely not solve the =
problem
yet.

In a next release I think it would worth a large effort to rationalize
various Vtable, VtxTable, and VtxArray in a single geometry leaf node,
optimally dealing with gfx card capability (e.g. has Vertex Arrays, =
VBO's,
...) under user hints (e.g. normally use VBO when available, but if the =
user
says 'use display list' it reverts to Vertex Arrays when available, =
etc.) -
all the loaders and ssgaShapes shoud be reworked a bit, but such a =
unique
geometry node would then be the place where to operate a centralized =
size
check/split policy. This would also favour further developments towards =
a
OGL 2.0-wise Plib 2.0 version, by avoiding coding things in three =
different
places.

Greetings -

Paolo Leoncini

-------------------------------------------------------------------------=

Paolo Leoncini     phone:  +39 (0823) 623134
Visualization & Virtual Reality   fax:    +39 (0823) 623126
CIRA - Italian Center for Aerospace Researches mailto:p.l...@ci...
Via Maiorise - 81043 Capua (CE) Italy  http://www.cira.it