A few examples, including objectives, expected and observed results, along with a description of your operating environment (OS & word size) might be a start, even if we can't individually reproduce those results.
Here is a release note from J6.02 which announces (potentially relevant) optimizations in this area: http://www.jsoftware.com/help/release/underai.htm To determine if this change is responsible for the symptoms, you might try some experiments in a pre-J6.02 interpreter. If it is responsible, it's a good bet that the optimization is implemented in the &. conjunction, so we could start out investigation in the source code of that primitive. -Dan Please excuse typos; composed on a handheld device. On Apr 20, 2013, at 9:00 AM, Raul Miller <[email protected]> wrote: > I have been seeing inconsistent results from the dyad > 22 b.&.(a.&i.) > > I do not have a repeatable test - the variations seem to be path > dependent (same data gives different results depending on how the data > was generated). However, replacing the above dyad with a more verbose > alternative apparently prevented the inconsistencies from occurring. > > This smells like buggy special code but it might be some external > dependence (perhaps an uninitialized value?) > > Unfortunately, I do not currently have any good ideas for isolating > this problem. > > FYI, > > -- > Raul > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
