On the technical side it's all implemented; however, I'm not at all happy with the new methods name ncall (new call); it seems pretty dumb, since it most likely doesn't matter to a user which one is "new".

Another name I considered was call_nowrap, but that also doesn't really explain itself well unless you know what call does, so it's also confusing if you don't care about the protocol and just want your tarantool-integration to work.

Another option would be to add an option to the creation of the object, something like

db = tarantool:new {
   user = 'foo'
   -- ...
   wrap_call_results = false;
}

where wrap_call_results defaults to true, with a big disclaimer that it will at some point be changed to false and people should fix their code.

Either way, my opinion is that this should be the default behavior for call; it's just more intuitive and better for performance, so I think the goal should be to some some day™ switch them out.

I just changed it so now there's only call, but also added a new configuration value to the tarantool:new method: call_semantics, which defaults to 'old'. (values can be 'old' or 'new', everything else errors)

I propose adding it to the readme example with a warning that the default will change in the future, and it should be set manually to 'old' in projects that need the old call result.

I think this is the least annoying way of implementing it and it seems to work pretty well for me. Some additional testing would of course be a good idea, but for me it seemed to work without bugs.

Also, for now I added a warning that the default value of "old" may change at some point; if you already know whether you actually want to change that, you could change it to will (or just leave it the way it is)

I will merge it and test it myself this week. Danke.