JEP 348: The wrong way?

[Johannes Kuhn]

"JEP 348: Compiler Intrinsics for Java SE APIs"'s most prominent use case is
String::format.

In this use case, we suspect to gain some performance improvement if the
format
string doesn't have to be parsed on each invocation.

I wonder if this is not something the JIT could do, with some help from the
developers.

In short, the JIT might reuse the result of a @Pure method instead of
calling
it again.

Declaring methods as @Pure and let the JIT figure out the details has
probably
other uses, but for now keep the discussion on String::format.

If the parsed format can be cached by the JIT, existing users of
String::format could benefit, while also avoiding two different
implementations
that have to be keep in sync.

I don't claim that this solution is simpler (it probably is not) or better
(run the numbers), just a suggestion.

JEP 348 doesn't come for free. Once committed to it, the bootstrap
methods have
to stay forever, even if they only return a ConstantCallSite later.

I hope that you will evaluate my suggestion and don't commit to early to an
implementation that turns out to be the wrong way later.

With best regards,
Johannes Kuhn