Back in February, Pete Pokrandt posted the following about the infamous
"ALGORITHM FAILURE"s that AREAS occasionally comes up with:
> From: poker@java.meteor.wisc.edu (Pete Pokrandt)
> Subject: Algorithm Failure in ARPRAM?
> Date: Mon, 20 Feb 1995 11:25:25 -0600 (CST)
>
> Hello, fellow NCAR Graphics users!
>
> Has anyone had problems using the color fill utilities from NCAR
> Graphics? The code that I am using works most of the time,
> but occasionally I get the error:
>
> ERROR 5 IN ARPRAM - ALGORITHM FAILURE
>
> after which the program stops. The error can be reproduced using
> the same data on the same machine, but I'm not sure exactly what
> it is about the data that is causing the error.
>
> This is using version 3.2 of NCAR Graphics on an IBM RS/6000 580 running
> AIX 3.2.
Some such problems are due to the fact that AREAS implements an algorithm
that is exact in Euclidean space, but does it in a rational-number space;
what this means is that an occasional problem will occur that just can't
be gotten around. However, I recently had an opportunity to get back into
the guts of AREAS and, in the process, I came across (and fixed) a bug that
would explain some such problems (maybe even most of them). This fix will
be in version 4.0.
Version 4.0 has also been updated to include error recovery in most utilities
(with the notable exception of AUTOGRAPH); it will be possible for the user
to turn on error recovery and at least get control back after an error has
occurred. In the case of "ALGORITHM FAILURE", the best course of action is
probably to skip the case that's "causing" the problem and go on to the next
frame; this is less than desirable, but probably better than just terminating
the program.
So, things have improved, I think ...
Dave Kennison
This archive was generated by hypermail 2b29 : Wed Jun 28 2000 - 09:45:33 MDT