Yi Lin

> Here's the API reference for the old MMTk https://www.jikesrvm.org/JavaDoc/org/mmtk/utility/alloc/Allocator.html#getMaximumAlignedSize-int-int-int- I was aware of that method and the comment. However, the comment merely says there is such an assertion but...

> Here's my understanding: `getMaximumAlignedSize` estimates how much memory is needed to allocate an object. The result is then used to pick the appropriate allocator. > The estimation is simple:...

There is one more difference between mmtk-core and Java MMTk: mmtk-core no longer does `checkAllocator()`, as that is the binding's job. So that means, `getMaximumAlignedSize()` would only be used in...

I thought about using the new running script. I think we may need a feature from the running script to do this: abort on failed runs. @caizixian

Bulk zeroing can only be done in mmtk-core. However, if a binding would like to zero each object themselves, we can support this (and avoid double zeroing) by providing an...

One other thing we may want to check is whether the OS gives zeroed memory with mmap. In Linux, the OS gives us zeroed memory, for which we do not...

> Done. Please don't merge until I've done a performance evaluation after rebasing. I only have stale results. Yeah. I won't merge it right now. I will update the CI...

That looks good. I have also updated the CI scripts. We can merge this PR. Update: I will run the binding tests just as a final check.

@k-sareen Can you take a look at the failed test? Just check if it is reproducible, and if it is related with the changes in this pull request.

> `luindex` is known to be broken for the generational plans (at least the new `luindex`). Likely to do with barriers. See: [mmtk/mmtk-openjdk#156](https://github.com/mmtk/mmtk-openjdk/issues/156). I'll investigate and see why this one...