opencascade and Boolean operations

Hi,
Below you will find a nice description of the boolean operations by Ames Arlo. It is related to some questions from an ACIS user regarding the robustness of booleans. Directly it has no connection to OCC but I think it is very usefull. Actually I'm using OCC for a very little period and I don't have a proper opinion about the boolean operations. But, just like Ames, I definitely have a sad opinion about all the booleans I used until now. Just like you said "Sometimes, and sometimes not..."

Here is the quotation:

"ACIS computes booleans by first constructing all the intersections
conditions (surface/surface, curve/curve, ...), then deciding how to limit
the intersections (what portions of the intersection curve are really on the
face), and resolving the model's topology. The inconsistency you're finding
is that ACIS is getting conflicting (inconsistent) information about how the
curve is contained in the model. It can think, for instance, that both
sides of a curve are present in a surface.

Fundamentally, boolean operations are not robust over the entire possible
space of geometric objects you might construct. There really has never been
a complete theory developed, especially where we permit the models to have
gaps at edges. Grazing cases, and models exhibiting inaccuracies, tend to
show such problems.

A variety of hacks exist, once you've trapped the error. You can scale the
parts down (sometimes up) to close up gaps (or to make previously too-close
points distinct). You can translate the parts closer to the origin to get
into the sweet spot of the floating point numbers. You can simplify your
parts so they don't have so much unimportant (and generally destabilizing)
geometry. You can locally adjust the positions of the offending faces/edges
so they aren't close to one another (this approach isn't so outlandish as it
seems -- look at Mulmuley, Computational Geometry for a scholarly treatment
of Randomized Algorithms; hacking the problem is a standard technique).

Understand that the most complete testing on booleans has been done in cases
of DESIGN operations, rather than arbitrary computer-generated operations.
Building CAD systems so designers can add holes, slots, etc is
bread-and-butter for most geometry engine writers, so that works a lot
better than, say, slicing a model with an arbitrarily located set of planes
(say for a rapid prototyping application). So, expect a bit of brittleness
if you're automatically constructing your geometries, and think about little
tweaks that work around the brittleness.

The concern here isn't about ACIS vs other geometry representations.
Everybody has the problem. Some work has been done in trying to implement
absolutely robust booleans, but even good theory on the problem is hard to
come by. Even exactly representing the intersection curves would help, but
runs the computational and development burdens through the roof.

Arlo Ames
alames@sandia.gov"

Best wishes,
Andras

Dear Vivek

1. "I have been experimenting with openCascade boolean operations for about two
weeks and my realization is that it is buggy!"
Ok. May be.
Could you transfer conversation to subject domain? Please attach examples.

2. "...I am wondering if I am doing something wrong...."
Here also we will look.

3 "...many people have had frustrating experiences with boolean operations..."
Ok. But there are a lot of threads among where the reasons of failure was wrong data or
just inept usage. As for me: I have good experience with boolean operations, especialy
with the latest version.

4 "...does anyone know of a good (and robust) boolean library that I can buy?..."
"... ACIS library... it's quite expensive ($20,000 + royalties)..."
It is very sad to know. I'm very sorry.
But in case of wrong data please do not hope that ACIS library will be able
to help you.

Wishes