Repository navigation
derive helpers not in attrs field in rustdoc json #157107
Description
Problem
This regression has been noticed when running the integration test of BuFFI after updating the toolchain from Rust 1.95.0 to 1.96.0.
For the functionality of BuFFI we rely on attribute information being provided in the output of rustdoc json which we parse using rustdoc-types. We are aware that this json output is unstable and we usually can adjust our side to keep track, but this time the version of rustdoc-types has not changed, yet it seems that some attribute information that was provided as part of Item objects seems to have vanished completely. After checking it I can confirm that it is not showing up in the rustdoc json output anymore.
Code
We have this enum in the example for BuFFI and we rely on the attributes to generate the correct FFI code.
#[derive(serde::Serialize, serde::Deserialize)]
pub enum RandomEnum {
/// An empty case that is here to make the test simpler
NoValue,
/// A timestamp from chrono that we would like to use in the API
TimeStamp(#[serde(with = "crate::DateTimeHelper")] DateTime<Utc>),
}I have chosen this example because it is the point where the test currently fails. There are more types in the example that show the various ways how we make use of attributes. If necessary I can provide a smaller example where it is also visible that the attributes are not part of the rustdoc json anymore.
If we have a look at the case TimeStamp, we want to have information about the serde attribute.
After parsing the generated json with rustdoc-types and resolving it partially, before we had
...
attrs: [
Other(
"#[serde(with = \"crate::DateTimeHelper\")]",
),
],
...
dbg! output on 1.95.0
[buffi/src/lib.rs:1486:29] &t = Item {
id: Id(
126,
),
crate_id: 0,
name: Some(
"0",
),
span: Some(
Span {
filename: "example/buffi_example/src/lib.rs",
begin: (
69,
56,
),
end: (
69,
69,
),
},
),
visibility: Default,
docs: None,
links: {},
attrs: [
Other(
"#[serde(with = \"crate::DateTimeHelper\")]",
),
],
deprecation: None,
inner: StructField(
ResolvedPath(
Path {
path: "chrono::DateTime",
id: Id(
94,
),
args: Some(
AngleBracketed {
args: [
Type(
ResolvedPath(
Path {
path: "chrono::Utc",
id: Id(
95,
),
args: None,
},
),
),
],
constraints: [],
},
),
},
),
),
}
while now we get
...
attrs: [],
...
dbg! output on 1.96.0 (note the empty `attrs` field)
[buffi/src/lib.rs:1486:29] &t = Item {
id: Id(
126,
),
crate_id: 0,
name: Some(
"0",
),
span: Some(
Span {
filename: "example/buffi_example/src/lib.rs",
begin: (
69,
56,
),
end: (
69,
69,
),
},
),
visibility: Default,
docs: None,
links: {},
attrs: [],
deprecation: None,
inner: StructField(
ResolvedPath(
Path {
path: "chrono::DateTime",
id: Id(
94,
),
args: Some(
AngleBracketed {
args: [
Type(
ResolvedPath(
Path {
path: "chrono::Utc",
id: Id(
95,
),
args: None,
},
),
),
],
constraints: [],
},
),
},
),
),
}
Using cargo-bisect-rustc I was able to narrow down the problem to this commit eb9d3ca.
I think the relevat PR could be #153540.
Version it worked on
It most recently worked on: Rust 1.95.0
Version with regression
rustc --version --verbose:
rustc 1.96.0 (ac68faa20 2026-05-25)
binary: rustc
commit-hash: ac68faa20c58cbccd01ee7208bf3b6e93a7d7f96
commit-date: 2026-05-25
host: aarch64-apple-darwin
release: 1.96.0
LLVM version: 22.1.2
Pinging @aDotInTheVoid (we briefly chatted about BuFFI and its use of rustdoc json before) and @scrabsha (author of the PR listed above) because I think they could be interested in the regression and might be best suited to answer if that is intended or not. I hope that is ok. 😅
@rustbot modify labels: +regression-from-stable-to-stable -regression-untriaged
Activity
scrabsha commentedon May 29, 2026
oh, i did not know derive helpers were exposed in rustdoc-json in the first place. my PR is indeed the culprit.
14 remaining items
JonathanBrouwer commentedon May 30, 2026
Made a PR but we can wait a little bit with merging this in case we can find a better solution
#157150
JonathanBrouwer commentedon May 30, 2026
Actually I'm not sure attributes from re-exports were ever forwarded for rustdoc-json, so maybe this approach works. I'll have more time to play around with this idea tomorrow, feel free to do it if you have time earlier @scrabsha
scrabsha commentedon May 30, 2026
i'd prefer to dedicate my tomorrow to understanding how type inference work, happy to rubber duck around the
actually_rustdocapproach though :)Rollup merge of rust-lang#157150 - JonathanBrouwer:fix-attrs, r=jdons…
obi1kenobi commentedon May 30, 2026
Derive helpers are often part of the public API contract and SemVer-relevant. The following is a SemVer-major breaking change:
#[derive(Default)] pub enum Witness { - #[default] A, + #[default] B, }Nothing we can retroactively about this case unfortunately. But could I be in in the loop for future changes around removing or materially changing how rustdoc JSON data is represented?
In particular I'm curious to understand more about the re-exports case, since cargo-semver-checks does track re-exports.
JonathanBrouwer commentedon May 30, 2026
This was an unintentional change to rustdoc-json which we will be reverting in #157150. I know you as someone who's interested in the format and I appreciate your work so would definitely have pinged you if this was intentional :)
aDotInTheVoid commentedon May 30, 2026
Thanks for the report!
Rustdoc-Json is a unstable feature that's only usable on nightly, not stable. While this is a bug with rustdoc-json, it's not a stable-to-stable regression.
This is fine, rustdoc-json doesn't inline items (unlike rustdoc-html), so we can rely on the fact that we're only generated documentation for local items, not cross-crate items.
[-]`attrs` field empty in rustdoc json on Rust 1.96.0 (stable)[/-][+]derive helpers not in `attrs` field in rustdoc json[/+]obi1kenobi commentedon May 30, 2026
SwishSwushPow commentedon May 30, 2026
Thanks everyone for handling this so quickly. :)
“Beta backport” means that this change would go into the 1.97.0 release already and not the 1.98.0 release, do I understand this correctly? In that case I would also be happy about a backport so that BuFFI can pick up the support sooner rather than later.
Although I’m not 100% sure if that is still on the table since that comment seems to have vanished. 😁
But in any case thanks again to everyone that chimed in!
Rollup merge of rust-lang#157150 - JonathanBrouwer:fix-attrs, r=jdons…
Unrolled build for #157150
JonathanBrouwer commentedon May 30, 2026
Beta backport decision thread opened here: #t-compiler/backports > #157150: beta-nominated @ 💬
Yes