Matthew Parkinson

>  > > @mjp41 Is 7.1.0 available in our pipeline？ Not convinced. You can try, maybe we can build a docker image with it in?

@aghosn does this example illustrate what you are after?

I had a look at LL/SC on Arm, but couldn't work out how to do it. The ABA primitive is designed to be able to use it, I just didn't...

It is enabled with ``` SNMALLOC_CHECK_CLIENT ``` I will update the documentation. That does not enable the memcpy protection. On Windows the performance of the memcpy protection is not great...

Thanks for running this benchmark. The CHECK_CLIENT version will use more memory. It needs to keep enough spare space for randomisation to have enough entropy. There are a few cases...

Yeah, it is to support allocations from outside snmalloc. The key issue is "absolute vast majority", if one isn't and we crash because of it, then customers will get very...

I have a PR ready for > Enforce power of two alignment on large allocations It is just waiting for the refactors in #120 to land on master.

The nearest bit is https://github.com/microsoft/snmalloc/blob/b1da339b3e57fe8fdf5dc2a3c0cb30fb2ea993c2/src/mem/chunkallocator.h#L147-L178 This on a timer pushes unused chunks of memory back into central pools and decommits them. I want to get LowMemory notifications to call a...

This isn't actually required by the process_sandbox usage. I think it is still nice to pull apart, so I have left the PR up. But it can be dropped.

So I like the principle of making a more principled error message from failed allocation. My big concern is that it needs to not affect the fast path. As such...