News for hackers. Stuff that matters.
Re:Libgraal: GraalVM compiler as a precompiled GraalVM nativ... by singularity2001 on Tuesday July 16, 2019 @07:50AM (#20449184)
can it be run in wasm and output native code (instead of teavm bytecode)?
Re:Libgraal: GraalVM compiler as a precompiled GraalVM nativ... by tom_mellior on Tuesday July 16, 2019 @07:56AM (#20449213)
The GraalVM compiler emits native code. It consumes JVM bytecode, or intermediate representations of other programming languages implemented using the Truffle framework.
Re:Libgraal: GraalVM compiler as a precompiled GraalVM nativ... by pron on Tuesday July 16, 2019 @10:30AM (#20450367)
I assume that a WASM backend could be created for Graal Native Image, so that it could compile Java byteocde to WASM ahead of time, but I wonder what benefits that would have over compiling bytecode to JS (as TeaVM does).
Re:Libgraal: GraalVM compiler as a precompiled GraalVM nativ... by singularity2001 on Tuesday July 16, 2019 @11:21AM (#20450900)
There is now a pure wasm version of TeaVM. My idea was that graal could run in wasm and compile to wasm, the (only) benefits would be faster and smaller code.
Re:Libgraal: GraalVM compiler as a precompiled GraalVM nativ... by pron on Tuesday July 16, 2019 @05:25PM (#20454044)
I'm not sure it would be either. JS has a good JIT that Java code can significantly benefit from (much more than, say, C) -- WASM doesn't, so compiling to JS could actually be faster; JS also has a GC and WASM doesn't, which means that the image must contain a GC, making it larger, not smaller. Either way, it's speculation, but I don't think it's obvious that compiling Java bytecode to WASM is preferable to compiling it to JS, but maybe the TeaVM maintainers can share their experience.