avoid more tuple allocations in Array constructors

[JeffBezanson]

The problem here is that Array{...}((n,)) passes the tuple to C as Any, requiring it to be heap allocated. This is a quick fix that gets around that by unpacking the tuple from julia first.

Issues like this can potentially be eliminated once and for all by #12447. Another approach to explore is passing the tuple to C as Ptr{Int}, allowing it to remain stack allocated. @vtjnash Is that possible?

[vtjnash]

You can ccall with the & syntax. Eventually Ref should also handle this, but not today.

[JeffBezanson]

I think I'll merge this for now since it's backportable, then later change the C entry points to never need tuples.

[andyferris]

You can ccall with the & syntax. Eventually Ref should also handle this, but not today.

@vtjnash can I ask about this? I didn't realize there was a difference here. Can I pass any isbits type to ccall with & and it won't heap allocate it first? Or just Tuple?

And secondly, is this a pointer to the actual binding or does it make a stack copy? For interacting e.g. with BLAS I would love to be able to "initialize" an immutable in BLAS. However BLAS doesn't return a value, but wants a mutable pointer, and to make this fast, I would like that to be a pointer to my stack frame. I do realize attempting this violates semantics of immutability (to have C mutate an immutable), but possibly it is warranted as a kind of "constructor" for isbits types? (To make this semantically correct, if one could introduce a new uninitialized binding in the ccall that would be cool).

(I had tried this "hack" in the past but for local variables in a function it seemed to be making some kind of copy that you couldn't access after the ccall. I might have been using Ref instead of & since the docs specified & as old syntax, but I can't remember anymore).

[vtjnash]

This isn't the right forum to ask questions, please use the mailing list to avoid driving the issue OT. And no, you can't violate immutability, that's illogical.

[andyferris]

Sorry Jameson (and everyone else). I got overexcited.... but the point is well taken.