buffer: zero-sized slices have .buffer but not .parent

[bnoordhuis]

Something that tripped me up earlier today. With master:

$ git rev-parse HEAD
7f1bb8044a3cb3cf8bcf525bcb4dd7d924e05fe3

$ out/Release/node
> Buffer(1).slice(0,1).parent  // one-byte slice has .parent
ArrayBuffer { byteLength: 8192 }

> Buffer(1).slice(0,0).buffer  // zero-sized slice has .buffer
ArrayBuffer { byteLength: 8192 }

> Buffer(1).slice(0,0).parent  // zero-sized slice doesn't have .parent 
undefined

[addaleax]

This seems weirdly intentional … @trevnorris ?

[trevnorris]

@bnoordhuis That is intentional. It's backwards compatible with the pre-typed array implementation. Check it against v0.12.

[bnoordhuis]

Huh, you're right. It's weird because v0.10 doesn't work like that.

Was that an intentional change in v0.12? I can't think of a plausible reason why someone would want it to work like that, it goes against the principle of least surprise.

[jasnell]

Yeah, I've also found this to be a bit odd. @trevnorris does it still have to be this way? What breaks otherwise?

[trevnorris]

@bnoordhuis Change happened b/c of an optimization I implemented, and while writing the change making sure a zero-slice was pointing to its parent was neither documented nor tested. So must have slipped through.

@jasnell Since this change has been around for so long and this is the first bug report based on the change, and also because the use of .parent is pretty much deprecated in favor of using .buffer, I'd say this could change w/o anyone noticing.