Supplemental fix for issue 14735 - Prevent self-recursive match

[9rnsr]

The slice expression s[] will match to T[n] by using compile-time length deduction.

Required by: dlang/dmd#4779

[schuetzm]

The slice expression s[] will match to T[n] by using compile-time length deduction.

Huh? Why's that? typeof(s[]) is T[], why should it match T[n]? Is the length deduction you're talking about documented somewhere?

[9rnsr]

Sorry it's not well documented in the spec. It's only in the changelog text:
http://dlang.org/changelog#implicitarraycast

Yes, typeof(s[]) is T[]. but s is a static array T[n] so it's whole slice length can be known at compile time.

[schuetzm]

What a surprise! I always assumed that things like this only worked inside one expression (like VRP)...

[schuetzm]

What a surprise! I always assumed that things like this only worked inside one expression (like VRP)...

Ok, I see it does only work inside one expression.

But on further thought, I don't like this, at least not if it requires hacks such as this one. T[] -> T[n] is an implicit conversion, so it should is already a worse match if there is an overload that takes T[] directly. But this apparently didn't help in this case, because it is still preferred to overloads with template constraints. I feel this is too bug-prone...

[WalterBright]

Regardless of whether this is a misfeature or not, I'm going to pull this because that debate should go elsewhere.

[WalterBright]

Auto-merge toggled on