Tim Burgess

Ah, yeah - backward compatibility... One option to kick off the process might be to start raising a `DeprecationWarning`/`FutureWarning` when the modules are imported? Will happily submit a patch for...

Yeah, definitely It's worth noting that the physical memory usage doesn't increase as both processes refer to the same physical memory, but if there is anything in place that limits...

Yeah, I've just been running into this lately as well. Seems like it might be a good idea to switch the compiler/mpi/gpu sniffs to use pytools. Note that along the...

Since we have been discussing #320 a bit lately, and the lack of first touch is proving to be a killer for KNL, it's worth pointing out that first touch...

> the lack of first touch is proving to be a killer for KNL I should probably elaborate here. There is typically no NUMA locality issue (at least not one...

``` devito.Operator(Eq(array.forward, array)) ``` Well, that copies the first timestep to the later ones; it presumably still relies on the first timestep being initialised to zero. This may already be...

I think there might be a case to be made that not calling `__array_finalize__` at unpickling time might be considered a numpy bug, since this document https://docs.scipy.org/doc/numpy/user/basics.subclassing.html seems to claim...

Well, that might be true, but it's relatively easy to end up with standalone `Data` objects. e.g. this returns a `Data`: ``` def f(): # create devito stuff # run...

Sorry, just getting back to this. It was the simplest way to make a reproducer, but I'm not 100% sure that pickling is the only source of the issue. At...

Probably a very similar fix if so, but does this potentially apply to `_dist_scatter` before going to C also?