Just now I was helping out a new enzo/yt user with a strange bug. After some time looking for a bug in their problem setup, I realized that the real issue was lingering .yt and .harrays files. I've personally been bitten by this on several occasions, wasting time before I remembered that I had forgotten to setup my .yt/config file. I think this behavior is very confusing and should only be enabled explicitly by a user who wants to repeatedly run an expensive calculation on the same dataset.
The consensus at that point seemed to be that serialization is useful for certain use cases, but that it should be an opt-in thing. Once this change is pulled in, it will be straightforward to turn on serialization in your local copy of yt, either by explicitly turning it on in your .yt/config file or using the ytcfg object before importing from yt.mods.
Since there hasn't been much forward progress on this issue since that e-mail thread, I've gone ahead and opened this PR. If you object to this change, please comment below so we can have a discussion about this.
I think this should be accepted. The wins outweigh the losses, it's an optional change, and we have notified affected individuals. I think it is fine to say that on Monday, unless we hear otherwise, this should be accepted.