Prevent riakc_set:new/2 when the context is undefined (or patch to_op)

[macintux]

At a glance, creating a set from scratch with riakc_set:new(List, undefined) seemed like a reasonable approach, but riakc_set:to_op/1 doesn't anticipate this case, thus making any set created and stored only contain any elements added after the new/2 invocation.

I'd suggest returning an error when context is undefined, or...

It's also possible to make to_op smart enough to handle this scenario; I haven't conducted any more than minimal tests, so it's possible that the combination of new/2 plus the smarter to_op breaks a half dozen other things.

https://gist.github.com/macintux/1f34d0443fdf029161c6

[macintux]

As an argument for allowing new/2 to be invoked for entirely new sets, here's what it takes to create a set from a list of values today, unless I'm missing something:

     lists:foldl(fun(X, Set) -> riakc_set:add_element(X, Set) end,
                 riakc_set:new(), ListOfValues)}.

Perhaps that's more of an argument for creating riakc_set:add_list than for allowing new/2 to be used this way.

[seancribbs]

@macintux Yes, the new/2 is only intended to be invoked from loading a datatype from the server. However, I don't see any reason an add_all, from_list, or union function couldn't be added.

The main difficulty is that these datatypes are only shadows of the ones on the Riak cluster, so their goal on the client-side is to encapsulate mutations, not to act like regular Erlang data structures.

[seancribbs]

Another option might be to rename new/2 to load/2. At least then it would be obvious that it isn't intended for users to create.

[macintux]

What's the intended use case for new/2 besides the EQC test?

[seancribbs]

Yes, the new/2 is only intended to be invoked from loading a datatype from the server.

^^

[macintux]

Ok, so my reading skills aren't top notch. I'm still a bit confused as to when it would be used, though. riakc_pb_socket:fetch_type returns a populated set record. The only reason to use it that jumps to mind is to overwrite an existing set with an entirely new set of values.

[macintux]

Oh

[macintux]

Yeah, I'm a little slow. How about load_from_server?

[seancribbs]

shrug At the very least, documentation is in order.

[macintux]

Documentation for data types is available via docs.basho.com and the README. Closing.