(annotations): annotations that include tokens are output as [object Object]

[swachter]

Describe the bug

If a stack with name 'some-stack' includes an info annotation

Annotations.of(this).addInfo(`stackId: ${this.stackId}`);

then the following output results:

[Info at /some-stack] [object Object]

Regression Issue

• Select this option if this issue appears to be a regression.

Last Known Working CDK Version

No response

Expected Behavior

The string representation of the token should be output.

Current Behavior

The message is output as "[object Object]"

Reproduction Steps

class MyStack extends Stack {
  constructor(scope: Construct, id: string) {
    super(scope, id)
    Annotations.of(this).addInfo(`stackId: ${this.stackId}`)
  }
}

Possible Solution

No response

Additional Information/Context

No response

CDK CLI Version

2.1000.2 (build bc82193)

Framework Version

No response

Node.js Version

v22.12.0

OS

Ubuntu

Language

TypeScript

Language Version

No response

Other information

No response

[pahud]

aws-cdk/packages/aws-cdk-lib/core/lib/annotations.ts

Line 146 in 241bda3

let normalizedMessage = typeof message === 'string' ? message : JSON.stringify(message);

I think the fix would be to use Token.isUnresolved() to detect if the string contains any Tokens, and if it does, use Token.asString() to ensure proper resolution at synthesis time.

e.g.

private addMessage(level: string, message: string | IResolvable) {
  const isNew = !this.scope.node.metadata.find((x) => x.data === message);
  if (isNew) {
    // If message already implements IResolvable (like a Token), or is a string containing tokens,
    // we want to preserve its resolvable nature rather than prematurely converting to string
    const normalizedMessage = Token.isUnresolved(message) ? Token.asString(message) : message;
    this.scope.node.addMetadata(level, normalizedMessage, { stackTrace: this.stackTraces });
  }
}

Making it a p1 and requesting inputs from the team.

[go-to-k]

@pahud

I think the fix would be to use Token.isUnresolved() to detect if the string contains any Tokens, and if it does, use Token.asString() to ensure proper resolution at synthesis time.

Even with Token.asString(), all tokens are resolved before the timing of message output by annotation, so I think it is necessary to avoid resolving only annotation tokens in the first place.

However, in that case, the tokens are also output to manifest.json of the cloud assembly. The tokens are for CDK apps, but the cloud assembly is for CFn or CDK CLI. Therefore, the cloud assembly should be output with the token resolved. (It already contains strings in other CFn formats such as ${AWS::AccountId}.)

So I submitted a PR to the CDK CLI repository without modifying the aws-cdk-lib. Because I thought that was the best way to fix just the message output part without changing the manifest format.

Please see the details: aws/aws-cdk-cli#101

[go-to-k]

@godwingrs22 @mrgrain @pahud

Could you please put this issue back into aws-cdk(lib)?

After talking to Rico, it was agreed that this should be done on the library side. (aws/aws-cdk-cli#101 (comment))

I submitted the PR but it is now P2; moving the issue would change it to P1.

[mrgrain]

Could you please put this issue back into aws-cdk(lib)?

🫡