On Fri, Feb 11, 2011 at 8:10 PM, Glenn Knickerbocker <[email protected]> wrote: > I'm looking up items in a series of files. For each file, I have a REXX > stage building an ELASTIC feedback loop to chase down chains of > references, and unresolved references get passed on to look up in the > next file. Duplicate references are weeded out with LOOKUP AUTOADD. > Since the later files in the chain might refer back to the earlier ones, > I figured I'd just feed the last unresolved output right back up to the > top with another ELASTIC to start over in the first file. But it never > got there! The whole mess terminated before anything could be fed back > to the top.
Somehow it sounds like you could do with your detail and master reversed and feed the references back in as masters instead, but your description is vague enough that one does the design on a beer coaster ;-) The case where I had to feed back into lookup was with a hierarchical dataset name authorisation. So when the full DSN did not match a master, I stripped the last qualifier and tried again. The feedback uses "fanintwo" because that has a preference for the secondary input. I suppose that today I would use the floor option... > I guess that's slightly better than the alternative, which would be to > stall right away because the second ELASTIC in the chain can't read > anything on its alternate until EOF on the first one. If I replace all > the ELASTICs with ELASTIC|FANINANY, I actually do get the output I want, > but then of course nobody ever sees EOF and the whole thing stalls. If > I save the output and run the whole pipeline again, I lose what's saved > in LOOKUP AUTOADD each time and wind up chasing down a lot of duplicate > references that I then just discard. Have I missed some way to get this > right? The LRUCACHE in my Lookup paper does that. After a number of master deletions, the remaining table is stored in an affixed pipeline and passed to the next incarnation of lookup. I even tried pictures of it in my presentation. http://www.rvdheij.nl/Presentations/2001-V110.pdf (pg 47-) And indeed, eof propagation in lookup is not trivial. We got a serious headache trying to picture it for all streams :-) | Rob -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/
