Consider mangling ID's in rustdoc-json

[aDotInTheVoid]

Currently JSON ID's are (approximatly) the underling DefId's which means they are not realy opaque, despite what we've said in the RFC.

They happen to be the compiler internal DefId for that item, but in the JSON blob they should be treated as opaque as they aren't guaranteed to be stable across compiler invocations

The problem with this is that their is no obligation to treet the ID's as opaque (for a concreet example see https://github.com/scrabsha/scrabsha.github.io/pull/2). If we want to garentee our ability to change the ID format, we should make it so that no one relies on the specific values of ID's by making them not cary information. Eg 0:0 lets the user know things, wheras a3aefbcbf31fa0bc12b9a73bc9d67976 (md5 hash) doesnt.

Potential Drawbacks

• Non determinism
• Harder to debug
• Slower to build
• Larger output

Which of these are actualy a problem, or just unfounded concerns I'm not sure:

If you want to change this: heres the places to look:

rust/src/librustdoc/json/mod.rs

Line 219 in 0fb1c37

root: types::Id(String::from("0:0")),

rust/src/librustdoc/json/conversions.rs

Line 183 in 0fb1c37

ItemId::DefId(did) => Id(format!("{}", DisplayDefId(did))),

I'm not sure if this is a good idea, and would love feedback

@rustbot modify labels: +C-discussion +A-rustdoc-json

[camelid]

MD5 hashes should be deterministic, so non-determinism would only be introduced if we used a random number generator.

[CraftSpider]

I think MD5 hashes would be okay, though I would also support simply encoding the output in a less 'human readable' way. I'm okay with information being conveyed, if it's more obvious that the user isn't meant to access/rely on that information (think base64 or such)

[aDotInTheVoid]

Found an example of someone parsing the ID format in the wild:

https://github.com/zjp-CN/term-rustdoc/blob/6559ededa253e822c946fa6a2123ad55de7f82a8/tests/parse-json-docs/main.rs#L329-L368

[aDotInTheVoid]

CC @zjp-CN. If there's information that's only availible by parsing the (supposed-to-be) opaque identifiers, I'd love to hear that. Otherwise just be aware that you're relying on an implementation detail, and it's liable to break without notice.

[zjp-CN]

Hi, I did try to parse an id to learn about what it means in the testing, but haven't depended on it much in the app.

There is missing information on parsing type: rustdoc_types::Impl doesn't tell if it's an auto impl, whereas an id starting with a can just tell it is.

[zjp-CN]

Looks like this is not a problem now, since #130078 has landed.