Implement more extra_traits

[Susurrus]

Finishing the implementation of rust-lang/rust#57715 now that all platforms are tested in CI. I plan to expand this CI to all platforms that need this treatment (solarish, empscripten, freebsd, and netbsd), but thought I'd get the part I've already started running in CI since I can't test this changes locally.

[rust-highfive]

r? @gnzlbg

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

[gnzlbg]

Cool! Thanks! Ping me when CI is green!

[Susurrus]

So 1.24 is the cutoff for using unsafe around packed struct instances, but older versions are tested in CI. I think we need to do the same macro trick that you used elsewhere to get this code compiling on all CI targets.

[gnzlbg]

The problem is that, if creating a reference to a packed struct field creates an unaligned reference, then the behavior is undefined.

That's the case irrespectively of whether one uses an unsafe block post Rust 1.24, or no unsafe block in older Rust versions.

[gnzlbg]

You could just add an attribute to those methods to silence the unused unsafe warning (e.g. "#[allow(unused_unsafe)]"), but that does not make the behavior of these operations defined.

EDIT: you also cannot create an unaligned reference, and then cast it to a pointer, and then check whether it is aligned or not, because the UB happens before the check .

[gnzlbg]

Instead of using references, you are probably better off copying the struct fields out of the struct into the stack frame, and then passing those (or references to those) to the formatter.

cc @Centril @RalfJung

[gnzlbg]

Alternatively, one can take a reference to the struct, cast it to a pointer, and use pointer offsets to read from the struct fields. One'd need to compute the offsets manually, but the structs are repr(C), so this is possible (the rules are in the C standard). If these structs change, then all offset would need to be manually updated though.

[RalfJung]

you are probably better off copying the struct fields out of the struct into the stack frame, and then passing those (or references to those) to the formatter.

Yes, that is the best approach IMO. You need a reference to call .field, so it needs to be aligned, so you need to make a copy.

[Susurrus]

Remaining failures are not due to this code change and for targets that are allowed to fail. I think this is all good now.

[gnzlbg]

@bors: r+

[bors]

📌 Commit d795b07 has been approved by gnzlbg

[gnzlbg]

LGTM!

[bors]

⌛ Testing commit d795b07 with merge 66f00d2...

[bors]

☀️ Test successful - checks-cirrus, checks-travis, status-appveyor
Approved by: gnzlbg
Pushing 66f00d2 to master...