update spec for asInstanceOf: it's weaker than pattern matching

[adriaanm]

Just now in the SIP meeting, @odersky says pattern matching is the safe variant, while asInstanceOf does the minimal thing to make bytecode pass (although the spec can't really refer to that)

[SethTisue]

(maybe this is settled/uncontroversial, but just in case:)

informally in the community, I always hear asInstanceOf described (as @odersky also did during the SIP meeting just now) as telling the compiler to get the hell out of your way. I think almost everyone would be rather surprised to discover that it's spec'ed in terms of pattern matching. if anything, it's conceived of as exactly the other way around

I don't think it would be hard to craft language that says that some additional platform-dependent checking may occur at runtime (JVM yes, Scala.js no afaik), and it would even be fair to briefly present the JVM behavior as an example

[sjrd]

Scala.js performs additional checking under fastOptJS for Scala types (i.e., not for JS types, which are <: js.Any), and throws UndefinedBehaviorErrors where the JVM would throw a ClassCastException. Under fullOptJS, the checks are removed.

In any case, the asInstanceOfs also obey the rule that says "do the minimum so that the IR checker [~= the bytecode verifier] is happy".