Compilation error: the trait bound Clone is not satisfied (for generated types) + same for `serde::Serialize` trait

[antkmsft]

This affects

• specification/quantum/Quantum.Workspace
• specification/eventgrid/data-plane/EventGridNamespace
• specification/communication/Communication.Messages
• specification/devcenter/DevCenter
• specification/computeschedule/ComputeSchedule.Management
• specification/cognitiveservices/OpenAI.Inference
• specification/healthdataaiservices/HealthDataAIServices.DeidServices

I think this happens when a type is also used as an OperationStatus.

I.e.

...
/// A job to be run in the workspace.
#[derive(Default, Deserialize, SafeDebug)]
#[serde(rename = "Job", tag = "itemType")]
pub struct JobDetails {
    /// The time when the item began execution.
    ///
    /// Operational visibility: Read
    #[serde(
        default,
        rename = "beginExecutionTime",
        skip_serializing_if = "Option::is_none",
        with = "azure_core::time::rfc3339::option"
    )]
    pub begin_execution_time: Option<OffsetDateTime>,

    /// The time when a job was successfully cancelled.
    ///
    /// Operational visibility: Read
    #[serde(
        default,
        rename = "cancellationTime",
        skip_serializing_if = "Option::is_none",
        with = "azure_core::time::rfc3339::option"
    )]
    pub cancellation_time: Option<OffsetDateTime>,

    /// The blob container SAS uri, the container is used to host job data.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "containerUri", skip_serializing_if = "Option::is_none")]
    pub container_uri: Option<String>,

    /// Cost estimate.
    ///
    /// Operational visibility: Read
    #[serde(rename = "costEstimate", skip_serializing_if = "Option::is_none")]
    pub cost_estimate: Option<CostEstimate>,

    /// The identity that created the item.
    ///
    /// Operational visibility: Read
    #[serde(rename = "createdBy", skip_serializing_if = "Option::is_none")]
    pub created_by: Option<String>,

    /// The type of identity that created the item.
    ///
    /// Operational visibility: Read
    #[serde(rename = "createdByType", skip_serializing_if = "Option::is_none")]
    pub created_by_type: Option<CreatedByType>,

    /// The creation time of the item.
    ///
    /// Operational visibility: Read
    #[serde(
        default,
        rename = "creationTime",
        skip_serializing_if = "Option::is_none",
        with = "azure_core::time::rfc3339::option"
    )]
    pub creation_time: Option<OffsetDateTime>,

    /// The time when the item finished execution.
    ///
    /// Operational visibility: Read
    #[serde(
        default,
        rename = "endExecutionTime",
        skip_serializing_if = "Option::is_none",
        with = "azure_core::time::rfc3339::option"
    )]
    pub end_execution_time: Option<OffsetDateTime>,

    /// Error information.
    ///
    /// Operational visibility: Read
    #[serde(rename = "errorData", skip_serializing_if = "Option::is_none")]
    pub error_data: Option<WorkspaceItemError>,

    /// Id of the job.
    ///
    /// Operational visibility: Read
    #[serde(skip_serializing_if = "Option::is_none")]
    pub id: Option<String>,

    /// The format of the input data.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "inputDataFormat", skip_serializing_if = "Option::is_none")]
    pub input_data_format: Option<String>,

    /// The input blob URI, if specified, it will override the default input blob in the container.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "inputDataUri", skip_serializing_if = "Option::is_none")]
    pub input_data_uri: Option<String>,

    /// The input parameters for the job. JSON object used by the target solver. It is expected that the size of this object is
    /// small and only used to specify parameters for the execution target, not the input data.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "inputParams", skip_serializing_if = "Option::is_none")]
    pub input_params: Option<JobDetailsInputParams>,

    /// The type of job.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "jobType", skip_serializing_if = "Option::is_none")]
    pub job_type: Option<JobType>,

    /// The identity that last modified the item.
    ///
    /// Operational visibility: Read
    #[serde(rename = "lastModifiedBy", skip_serializing_if = "Option::is_none")]
    pub last_modified_by: Option<String>,

    /// The type of identity that last modified the item.
    ///
    /// Operational visibility: Read
    #[serde(rename = "lastModifiedByType", skip_serializing_if = "Option::is_none")]
    pub last_modified_by_type: Option<CreatedByType>,

    /// The timestamp of the item last modification initiated by the customer.
    ///
    /// Operational visibility: Read
    #[serde(
        default,
        rename = "lastModifiedTime",
        skip_serializing_if = "Option::is_none",
        with = "azure_core::time::rfc3339::option"
    )]
    pub last_modified_time: Option<OffsetDateTime>,

    /// The last time the item was updated by the system.
    ///
    /// Operational visibility: Read
    #[serde(
        default,
        rename = "lastUpdatedTime",
        skip_serializing_if = "Option::is_none",
        with = "azure_core::time::rfc3339::option"
    )]
    pub last_updated_time: Option<OffsetDateTime>,

    /// The job metadata. Metadata provides client the ability to store client-specific information
    ///
    /// Operational visibility: Create, Read, Update
    #[serde(skip_serializing_if = "Option::is_none")]
    pub metadata: Option<JobDetailsMetadata>,

    /// The name of the item. It is not required for the name to be unique and it's only used for display purposes.
    ///
    /// Operational visibility: Create, Read, Update
    #[serde(skip_serializing_if = "Option::is_none")]
    pub name: Option<String>,

    /// The format of the output data.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "outputDataFormat", skip_serializing_if = "Option::is_none")]
    pub output_data_format: Option<String>,

    /// The output blob uri. When a job finishes successfully, results will be uploaded to this blob.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "outputDataUri", skip_serializing_if = "Option::is_none")]
    pub output_data_uri: Option<String>,

    /// Priority of job or session.
    ///
    /// Operational visibility: Create, Read, Update
    #[serde(skip_serializing_if = "Option::is_none")]
    pub priority: Option<Priority>,

    /// The unique identifier for the provider.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "providerId", skip_serializing_if = "Option::is_none")]
    pub provider_id: Option<String>,

    /// Quantum computing data.
    ///
    /// Operational visibility: Read
    #[serde(
        rename = "quantumComputingData",
        skip_serializing_if = "Option::is_none"
    )]
    pub quantum_computing_data: Option<QuantumComputingData>,

    /// The ID of the session that the job is part of.
    ///
    /// Operational visibility: Create, Read
    #[serde(rename = "sessionId", skip_serializing_if = "Option::is_none")]
    pub session_id: Option<String>,

    /// The status of the job.
    ///
    /// Operational visibility: Read
    #[serde(skip_serializing_if = "Option::is_none")]
    pub status: Option<JobStatus>,

    /// List of user-supplied tags associated with the job.
    ///
    /// Operational visibility: Create, Read, Update
    #[serde(skip_serializing_if = "Option::is_none")]
    pub tags: Option<Vec<String>>,

    /// The target identifier to run the job.
    ///
    /// Operational visibility: Create, Read
    #[serde(skip_serializing_if = "Option::is_none")]
    pub target: Option<String>,

    /// Resource consumption metrics containing provider-specific usage data such as execution time, quantum shots consumed etc.
    ///
    /// Operational visibility: Read
    #[serde(skip_serializing_if = "Option::is_none")]
    pub usage: Option<Usage>,
}
...

and then it is used elsewhere, i.e. here it is used in Pageable

/// Paged collection of JobDetails items
#[derive(Clone, Default, Deserialize, SafeDebug, Serialize)]
#[non_exhaustive]
pub struct PagedJobDetails {
    /// The link to the next page of items
    #[serde(rename = "nextLink", skip_serializing_if = "Option::is_none")]
    pub next_link: Option<String>,

    /// The JobDetails items on this page
    #[serde(default)]
    pub value: Vec<JobDetails>,
}

But it's not only when Pageable, i.e.

/// The error object.
#[derive(Default, Deserialize, SafeDebug)]
#[non_exhaustive]
pub struct WorkspaceItemError {
    /// One of a server-defined set of error codes.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub code: Option<String>,

    /// An array of details about specific errors that led to this reported error.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub details: Option<Vec<ErrorDetail>>,

    /// An object containing more specific information than the current object about the error.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub innererror: Option<ErrorDetail>,

    /// A human-readable representation of the error.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub message: Option<String>,

    /// The target of the error.
    #[serde(skip_serializing_if = "Option::is_none")]
    pub target: Option<String>,
}

...

/// A workspace item.
#[derive(Serialize)]
pub(crate) struct UnknownItemDetailsItemType<'a> {
    ...

    /// Error information.
    ///
    /// Operational visibility: Read
    #[serde(rename = "errorData")]
    pub error_data: &'a Option<WorkspaceItemError>,

    ...
}

I think the thing that is common is that the pub struct that doesn't have Clone is being used as LRO operation status. Maybe the fix is simple - to update a place in the code to include Clone there - there is a specific line responsible for that, maybe at a time we didn't add it simply because we thought it is not needed and kept things to a minimum.

[antkmsft]

See also #918

[jhendrixMSFT]

While adding Clone to the emitted types would help (it's unclear at present why it's missing in the first place), without Clone on azure_core::error::ErrorDetail we're still broken.

[antkmsft]

I updated the bug, I see the very similar complaints for Serialize trait for those very same types

[jhendrixMSFT]

I see. I think this is how we can summarize the issue. For types that are Clone/Serialize, all transitive types must also be Clone/Serialize (same for Deserialize but apparently we haven't hit that case yet). For the latter, we add Serialize if the type is marked as output. I'm guessing there's a bug somewhere (tcgc and/or us) calculating this. Not yet sure about Clone.

[jhendrixMSFT]

const isOperationStatus = hasAzureErrorDetailFields(model);

Is part of the problem here. e.g. for JobDetails it is marked as an I/O type but isOperationStatus === true which causes us to omit Serialize.

[jhendrixMSFT]

I think there's also a high-level question: when should a model type not derive Clone?

[antkmsft]

for the information, there are currently no errors complaining the lack of Deserialize, only for Serialize it is.

[antkmsft]

For Serialize, maybe this - #918 (comment)
We could be able to get away with not implementing it, however that would require all the specs marking it as Lifecycle.Read which maybe they do need, maybe not.