indy-based string-switch proposal

[Japris Pogrammer]

Now starting to experiment with switches as part of your pattern-runtime
branch.
Should I follow the current isea behind current SwitchBootstraps (branch
IDs being defined implicitly, only API-based CalSites and so on) or am I
free to try other approaches now including other bootstrap configurations,
other CallSite implementations etc?


вт, 5 нояб. 2019 г. в 19:59, Paul Sandoz <paul.sandoz at oracle.com>:

>
>
> > On Nov 4, 2019, at 3:09 PM, John Rose <john.r.rose at oracle.com> wrote:
> >
> > On Nov 4, 2019, at 2:27 PM, Paul Sandoz <paul.sandoz at oracle.com <mailto:
> paul.sandoz at oracle.com>> wrote:
> >>
> >> repeating structure using candy
> >
> > Such as an apartment building canvassed by trick-or-treaters?
> > (Boo, spell-check!)
> >
>
> I am glad it afforded some light relief :-) it could have been worse: “… a
> marzipan increase in encoding confectionary …”.
>
>
> > Condy can be good for factoring out repeats.
> > But watch the per-condy overhead.
> >
> > Sometimes there’s a good answer from cooking up a
> > simple stringy DSL as with the string concatenation work.
> >
>
> Yes, good point.
>
> Paul.
>
> > When I evaluate alternative encoding schemes I usually
> > ask what is the expected number of constant pool indexes
> > burned, for an average use case.  Sometimes class file
> > size (in bytes) comes into it also, but CP slots are a uniquely
> > limited resource, since the whole class file has to make
> > do with 2^16-1 of them.
> >
> > The important goal, IMO, is to make each constant
> > in an indy/condy construct carry its weight, or else
> > somehow maintain an expectation that the constant
> > is probably already in use somewhere else in the
> > class file.
> >
> > — John
> >
> > P.S. There are incremental ways to lift the 2^16-1 restriction,
> > if we ever need to go there.  See notes on CONSTANT_Group
> > in https://bugs.openjdk.java.net/browse/JDK-8161256 <
> https://bugs.openjdk.java.net/browse/JDK-8161256>
> >
> >
>
>