der: add AnyCow

[dishmaker]

AnyRef, that allocates as Any with alloc feature

[tarcieri]

I would prefer not to have any of these types that try to special case Cow.

We can likely do something similar to what I proposed in #1795 for Any/AnyRef using tagged pointers to store the tag for AnyRef

[tarcieri]

Hmm, one problem with tagged pointers is that the Tag type can now need up to 48-bits to represent, which can't be stored in e.g. a 32-bit tagged pointer

[dishmaker]

Yes, the Tag alongside &'a [u8] is a problem.

One parallel solution is:

pub struct AnyCow<'a> {
    tag: Tag,
    value: Cow<'a, BytesRef>,
}

but that wouldn't be alloc-gated on allocating variant.

[tarcieri]

Upstream Cow should now work for OctetStringRef/OctetString.

Are there any other cases you think are immediately important?

[dishmaker]

Since you've reworked Utf8StringRef, it's perfect. I have to play with newest der some day, and test egui ASN.1 editor to see if ergonomics/performance increased.

For minimal support of x509-cert with Cow, i think BitStringRef/BitString need similar treatment, but that wouldn't work with regular Cow.

formats/x509-cert/src/certificate.rs

Line 162 in 4db4c09

pub(crate) issuer_unique_id: Option<BitString>,

Then, Any (aliased AttributeValue) needs some Cows:

formats/x509-cert/src/attr.rs

Lines 90 to 92 in 4db4c09

	pub struct AttributeTypeAndValue {
	pub oid: AttributeType,
	pub value: AttributeValue,

[tarcieri]

BitString/BitStringRef could work with Cow, but it would probably involve some hacks like using tagged pointers, and storing the owned BitString as a tagged pointer.

I don't think I can't do the same with Any/AnyRef though now that tags have been changed from u8, because there's too much data to put in the pointer