Recursive paths are not correctly detected

[SOF3]

use std::collections::HashMap;

use serde::*;

#[derive(Debug, Deserialize, Clone)]
pub struct Schema {
    #[serde(flatten)]
    typed: TypedSchema,
    description: Option<String>,
}

#[derive(Deserialize, Debug, Clone)]
#[serde(tag = "type")]
pub enum TypedSchema {
    Object(ObjectSchema),
    Array(ArraySchema),
}

#[derive(Deserialize, Clone, Debug)]
pub struct ObjectSchema {
    properties: HashMap<String, Schema>,
}

#[derive(Deserialize, Clone, Debug)]
pub struct ArraySchema {
    items: Box<Schema>,
}

fn main() {
    let r: Result<Schema, _> = dbg!(serde_path_to_error::deserialize(&mut serde_json::Deserializer::from_str(json())));
    if let Err(err) = r {
        println!("path: {}", err.path());
    }
}

fn json() -> &'static str{
    r#"
{
    "type": "Object",
    "properties": {
      "subkeys": {
          "properties": {
            "emails": { "type": "Array", "items": {} },
            "subkeys": { "type": "Array", "items": {} }
        }
      }
    }
}
    "#
}

[src/main.rs:30] serde_path_to_error::deserialize(&mut serde_json::Deserializer::from_str(json())) = Err(
    Error {
        path: Path {
            segments: [],
        },
        original: Error("missing field `type`", line: 12, column: 1),
    },
)
path: .

The errors are at emails and subkeys having empty objects, but the path was reported to be at ..

[dtolnay]

Would you be able to minimize this to the smallest data structure that reproduces the problem? There is a lot going on in that code (attributes and types and so forth) and probably not all of it is related to this issue.

[SOF3]

It appears that this issue might be caused by an unnoticed upstream bug in serde or serde_json.

https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=c68d1d013cbc59fbee638653af809ab2

The line and column reported by serde_json is also incorrect, pointing to EOF rather than the actual problematic line. However I am not sure if it is caused by serde or serde_json.

[dtolnay]

I'll go ahead and close this in the absence of a repro involving serde_path_to_error.