parallel runs

[John Hall]

Scott and Stephen:
It turns out that I had already noticed that I needed to use 
DistributedTag instead of ReplicatedTag in my field and layout 
creation routines, so the error in FieldTour1.cpp was not my problem 
(unfortunately). When we made the suggested change to Fieldtour1.cpp 
of course we ran that test correctly along with all the others in the 
NewField test directory. We do not seem to have a StencilTest in that 
directory using the standard checkout.

We also tried last night running with no guard cells, 
--pooma-nocompress and changing CompressibleBrick to simply Brick. 
None of these changes had any effect on the final result, which is 
hints of patches of correct behavior surrounded by gibberish.

I have a series of further tests in mind today including running 
under the debugger (totalview, which has finally gotten to the point 
where it is an excellent tool even with KCC) and determining whether 
the data is correct and maybe the problem lies simply in the printing 
(we can all dream, right?). I am also going to start coding some of 
our tests directly in Pooma, although I am already running in the 
Poomalote regression test layer which is essentially straight Pooma 
anyhow since it is our lowest layer.

Anyhow, I will finally get it down to exactly when we can run OK and 
when we can't since I now have some working reliable examples. Until 
then we will have to keep saying that "Blanca/Pooma is without 
parallel without exception" since performance is dramatically 
improved without exceptions.
Thanks,
John