[AVRO-4011] Schema generated via AvroSchema is not compatible with itself

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 1.11.3
• Fix Version/s: 1.12.0, 1.11.4
• Component/s: rust

Environment

Rust 1.79.0

• Assignee: Martin Tzvetanov Grigorov
• Reporter: Marco Lugo
• Created: 02/Jul/24 12:54
• Updated: 04/Jul/24 08:23
• Resolved: 04/Jul/24 08:23

Description

We encountered an issue where our Avro schema generated with the AvroSchema macro is not compatible with itself. That is, the can_read function fails to read the generated schema. Here's a minimal reproducible example:

use apache_avro::AvroSchema;

#[derive(AvroSchema)]
struct Foo {
    a: Vec<Bar>,
    b: Vec<Bar>,
}

#[derive(AvroSchema)]
struct Bar {
    value: i32,
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn schema_reflexivity() {
        let schema = Foo::get_schema();
        println!("{}", &schema.canonical_form());
        assert_eq!(&schema, &schema);
        assert!(apache_avro::schema_compatibility::SchemaCompatibility::can_read(&schema, &schema)); // this fails
    }
}

This is with the only dependency in Cargo.toml being:

apache-avro = { version = "0.16.0", features = ["derive"] }

The automatically generated Avro schema is:

{
    "name": "Foo",
    "type": "record",
    "fields": [
        {
            "name": "a",
            "type": {
                "type": "array",
                "items": {
                    "name": "Bar",
                    "type": "record",
                    "fields": [
                        {
                            "name": "value",
                            "type": "int"
                        }
                    ]
                }
            }
        },
        {
            "name": "b",
            "type": {
                "type": "array",
                "items": "Bar"
            }
        }
    ]
}