Sébastien Doeraene
Sébastien Doeraene
It's meant to be in this PR. Otherwise the `@noinline` that we add to the methods of `RandomGenerator` cause (slight) regressions.
FTR, Node.js 25.2 did not in fact upgrade to a version of V8 that has the latest spec for `configureAll`. However, that's moot anyway because [the opcodes are changing](https://github.com/WebAssembly/custom-descriptors/pull/85). We'll...
@gzm0 This is not yet ready for a code review. However, it is a good time to get your opinion on the overall new compilation scheme and architectural changes. WDYT?
Some results from the `bounce` benchmark, where I have replaced the PRNG by a `ju.SplittableRandom`. Our `SplittableRandom` was not specially crafted to optimize its `Long` usages (unlike `ju.Random`), so it...
> * I agree with the sentiment that the "pack" operation should remain linker internal. As an alternative to the desugarer, have you considered patching the IR in `injectedIRFiles` of...
Yes, I think it would make sense. Paradoxically, that's harder to do than it looks, so I haven't done it yet.
@gzm0 I managed to implement the variant where results are boxed in a `$Long` instance. It's the same internal class used for actual boxes in `Any`. Now it works even...
`LongOpOfExpression` is a bit difficult to grasp, I agree. It means the operation is not splittable at the top level, but all its arguments are expressions (hence, the latter must...
And now the TODO's are resolved as well.
@gzm0 So that I can plan a bit ahead: what's your status on the second round of review, here? I'm starting to pile up several things on top of this...