Add {size/min_align/pref_align}_of_val intrinsic

[ghost]

Allows DSTs to be sized generically at runtime.

core::mem::{size/align/min_align}_of_val have also been slightly altered
to use the new intrinsics, and to also allow DSTs.

Doesn't work for DST enums and tuples, because I had nothing to test with to ensure it works. Taking a pointer to a DST enum or tuple is currently an ICE(#16812). Would it be better to wait until that ICE is fixed before doing this? Would it be better to add skeleton functionality that should work once the ICE is fixed?

Trait Objects currently seem to only store min_align in their vtables (presuming that this is correct? My tests suggest that the value stored there matched min_align rather than pref_align, at least) so pref_align_of_val will just return that for now. Is that okay?

closes #19063

[rust-highfive]

r? @pcwalton

(rust_highfive has picked a reviewer for you, use r? to override)

[huonw]

trailing '.'

[huonw]

Awesome!

r? @nick29581

[ghost]

I realised shortly after doing this that most of what I did was already in glue.rs, so I made some modifications to the function there so that I could use it instead.

There were a couple of errors in the file that I noticed and changed:

• align was using the lower of the two alignment values, rather than the higher.
• sized_size was using llsize_of_alloc, which includes padding, which is not what is really wanted.

[huonw]

Hey @whataloadofwhat, did you mean to close this PR?

[ghost]

Yes, I was talking with eddyb on the irc and realised that it was probably going to be producing wrong results so I closed it.

I'm still looking for a way to get it to work, but I don't expect it to happen any time soon (if at all, I'm a little out of my depth here to be honest), and I don't want to discourage someone else from giving it a try by there being an open pull request for it.

Sorry about that, but thanks for the interest.