[formrecognizer] model_id param validation

[kristapratico]

Item on arch board issues #11283

Original suggestion was to to enforce that user passes guid for methods that accept model_id parameter. Based on architect guidance we will not validate for this, but will validate that model_id is not None or an empty string.

Reasoning is that service could eventually allow users to name their models with "real" names and not just server generated guids. It's reasonable to check for None/empty string since this will likely never be a valid model_id and prevents a 404 error in the path URL.

[kristapratico]

@maririos FYI since we talked about this

[maririos]

ohh interesting.
In .NET we do validate that we can convert the string into a GUID, because currently the service expects a UUID. It is worth to checking this in Python?

FYI @samvaity as she was looking into consistency across languages for this item

[iscai-msft]

Approving, but also waiting to hear what @samvaity thinks about guid validation

[samvaity]

In .NET we do validate that we can convert the string into a GUID, because currently the service expects a UUID. It is worth to checking this in Python?

I think we should validate if it is a valid UUID. Especially since the service does not validate that eagerly and will result in a request failure.
We can save users time and cost by failing early on the client-side.
What do you think @kristapratico ?

[rakshith91]

Curious - don't we have match conditions for delete operations in form?

[kristapratico]

Can you show me where you see that?

[kristapratico]

I think we should validate if it is a valid UUID. Especially since the service does not validate that eagerly and will result in a request failure.
We can save users time and cost by failing early on the client-side.
What do you think @kristapratico ?

I talked to Johan about this and posted the reasoning above:

Reasoning is that service could eventually allow users to name their models with "real" names and not just server generated guids. It's reasonable to check for None/empty string since this will likely never be a valid model_id and prevents a 404 error in the path URL.

[samvaity]

I talked to Johan about this and posted the reasoning above:
Reasoning is that service could eventually allow users to name their models with "real" names and not just server generated guids. It's reasonable to check for None/empty string since this will likely never be a valid model_id and prevents a 404 error in the path URL.

Not sure how other languages are handling this currently, but Java will always first convert the user-provided string to a UUID, using fromString() from UUID's lib class to pass it down to the service. If the string provided to fromString() does not conform to UUID specifications it will throw an IllegalArgumentException.

[kristapratico]

Not sure how other languages are handling this currently, but Java will always first convert the user-provided string to a UUID, using fromString() from UUID's lib class to pass it down to the service. If the string provided to fromString() does not conform to UUID specifications it will throw an IllegalArgumentException.

But if the service eventually allows the user to name the model, e.g. "myInvoiceModel", won't this fail with IllegalArgumentException? This is what Johan wants to avoid

[samvaity]

But if the service eventually allows the user to name the model, e.g. "myInvoiceModel", won't this fail with IllegalArgumentException? This is what Johan wants to avoid

To allow this, the service will probably have to start accepting Strings in their API's and that's when we (Java) can plan to remove the UUID.fromString(). Currently, with the service expecting the type UUID, there is no way of escaping converting the user-provided string to UUID.