Force push! and unshift! to need at least one item

[jakebolewski]

Is there any reason for this fallback (I'm guessing to support splatting of potentially zero arguments)? Otherwise it can be a source of subtle bugs.

cc @andreasnoack

[tkelman]

if you're going to push! a splatted collection onto something, you should be using append! anyway

Should there be a specific error message suggesting to do so?

[StefanKarpinski]

I.e. should we add a deprecation for this? It's hard to say since if it gets triggered, we may suspect that someone is splatting values into a push! call, but we can't be sure.

[mbauman]

It seems like it'd be a good Lint message. @tonyhffong

julia> lintstr("x=[1,2,3]; push!([], x...)")
0-element Array{Lint.LintMessage,1}

[ivarne]

+1 for the API cleanup.

Depreciation seems to be in order. We're definitely not in a hurry to potentially break peoples code for this.

[IainNZ]

I like this, because I've accidentally done this (1 arg push) and didn't get a method error - I'd have liked one.

[JeffBezanson]

Gotta love "0, 1, or N" problems :)
+1 for a deprecation

[andreasnoack]

@IainNZ That was exactly the motivation. Last night and this morning were spent figuring out why my code was numerically unstable only to realize that I didn't push anything to the vector that should keep track of the error growth.