This message was deleted.

Yeah so I think that use-case is a little iffy right now, although there are a few ways around it. You should be able to use the dask result builder in the combine= argument, which would force computation + do any joining.

@Jan Hurst yeah the “build_result function” in the graph adapter isn’t called when using a materializers, i.e. when you use

.materialize()

instead of

.execute()

. So the “compute” at end doesn’t apply. I think this is a gap in the docs that we’ll need to address. Basically with materializers we’ve removed the need for specifying a build_result implementation at driver creation time — which I think is the better design, but yeah we need to more explicitly document the interplay materializers have with graph adapters.