Further Truffle-related clean ups and simplifications.

[thomaswue]

Please review, @chrisseaton @nirvdrum @eregon.

[thomaswue]

We do not need a safepoint if the block is executed only once. In general, doing the safepoint only after the first iteration reduces the number of safepoints compiled/executed.

[eregon]

But the RedoException branch is only taken when using the redo, keyword, no?
And that is quite rare, so there is no more safepoint in a typical loop with this code if I see right (or is it done with instrumentation by the LoopNode?).

[chrisseaton]

Yes redo is unusual - most loops will run this inner while loop just once per actual loop iteration. I think we need to poll the safepoint before the return as well. This duplication of safepoints might be why I put it at the head in the first place.

[thomaswue]

You are right. I will put the safepoints back to the original places.

[chrisseaton]

Perhaps we want an ArrayUtils class in com.oracle.truffle.api.util, that are all amenable to PE. Even if some of the cases like System.arraycopy work fine anyway, it would mean that people don't have to figure out if they're safe or not for every array utility method in the JDK. Maybe the same for BigInteger - a set of helper methods that add @TruffleBoundary on those that need them, but not those that don't.

[eregon]

Can we merge this?