@azure/ai-form-recognizer type definitions do not match actual response data

• Package Name: @azure/ai-form-recognizer
• Package Version: 4.0.0-beta.4
• Operating system: Linux
• nodejs
  • version: 16
• browser
  • name/version:
• typescript
  • version: 4.7
• Is the bug related to documentation in
  • README.md
  • source code documentation
  • SDK API docs on https://docs.microsoft.com

Describe the bug The returned analysed document data from calling beginAnalyzeDocument with the prebuilt-document model has data that doesn’t match the declared type declarations. The pages[].lines[] array element objects have an incorrect type definition. Expected type defined in type definitions:

{
  polygon?: Point2D[]
}

Actual type of returned value:

{
  polygon?: number[]
}

Reproduce

import { DocumentAnalysisClient } from "@azure/ai-form-recognizer";

const client = new DocumentAnalysisClient(...);
const res = await client.beginAnalyzeDocument("prebuilt-document", documentFile);
// TS type: { polygon?: Point2D[] }
// Actual data: { polygon?: number[] }
console.log(res.pages[0].lines[0]);

Note that the other objects that have the polygon property (words, paragraphs, keyValuePairs, etc.) are correctly returned as Point2D objects; only line polygons are returned as an array of numbers.

[witemple-msft]

@wilsonzlin Thanks for reporting this. We transform the number[] into Point2D[] for convenience in the API, but due to some loose type constraints in one of our internal functions, we missed it specifically on DocumentLine objects.

We’re scheduled to release a new beta: 4.0.0-beta.6 today that should resolve this issue.

[jsquire]

It’s got the right labels on it to qualify for getting picked-up the next time we run training.