I don’t think it would be that difficult, now that I think about it. The splitdim() trans hooks could just check ndims in the parent and child before entering their dereferencing loops.
> On Nov 18, 2015, at 8:53 AM, Chris Marshall <[email protected]> wrote: > > Craig- > > How difficult would it be to have it > work for dataflow piddles that have > their dimensions in the right order? > > In that case, the reshape would be > equivalent to a sequence of splitdim() > calls but much less tedious to execute. > > It could still die and give a message > if things were hinky. > > --Chris > > On Wed, Nov 18, 2015 at 10:41 AM, Craig DeForest <[email protected] > <mailto:[email protected]>> wrote: > I “fixed” it by masking it out — attempting to do the thing that tickled the > bug now throws a meaningful error message instead. > > The issue is in the way reshape() is implemented (using setdims()), > interacting with the way mv() is implemented. The mv() code pre-caches the > dimlist offsets for both parent and child, so if some dims disappear from the > parent or child, then in the next dataflow operation the mv() trans code can > dereference a pointer to hyperspace. One way to fix that would be to check > against ndims() each time — but setdims() and reshape() shouldn’t be called > on dataflow-linked PDLs anyhow (they’re mostly for doing things like slurping > in a data file and then formatting it once it is in memory), so it made sense > (to me) to simply mask out the behavior. > > There are some tests to make sure reshape() throws an error in the proper > condition, and doesn’t in other conditions. > > > > > > On Nov 14, 2015, at 8:02 PM, Karl Glazebrook <[email protected] > > <mailto:[email protected]>> wrote: > > > > > > The reshape() bug was fixed? > > > > Is there a test? > > > > Karl > > > > > >> On 15 Nov 2015, at 8:18 am, Chris Marshall <[email protected] > >> <mailto:[email protected]>> wrote: > >> > >> All- > >> > >> ...and should be appearing at a mirror > >> near you soon. This release is the first > >> release candidate for the upcoming > >> PDL-2.015 release. It has fixes for the > >> major problems found in the PDL-2.014 > >> release. > >> > >> Please test for your platformsand with > >> your PDL modules and programs as soon as > >> possible. The PDL-2.015 release will be > >> no later than 01-Dec and by next weekend > >> if things test well! > >> > >> Thanks, > >> Chris > >> >
------------------------------------------------------------------------------
_______________________________________________ pdl-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/pdl-general
