Kiota produces different model with same command and OpenAPI description

[bmarkovic17]

I also ran into the issue with Kiota not being idempotent and after going through the already existing issues (mainly #2442) I decided to create a new one as it might provide more insight. The problem basically boils down to these few classes shown on image below where most of them don't get generated, while the one that depends on them gets generated differently.

Setting the environment variable KIOTA_GENERATION_MAXDEGREEOFPARALLELISM to 1 didn't help, but I've noticed something that might be helpful:

• generate command will almost everytime produce a wrong model
  • I have an impression that the model gets generated okay when the command takes a bit longer to execute (around 10s), while usually it takes around 5s and then it gets it wrong. This is my feeling that could be wrong
• update command will almost everytime produce a correct model
  • on one occassion it produced a TeamAndCompetitor class in the same way that the generate command produces it, even though the othere classes were not missing

Here is the the OpenApi description that I'm using with Kiota 1.11.1 and here are the commands that I've used to generate the client:

• kiota generate --openapi openapi.json --language csharp --clean-output --clear-cache --exclude-backward-compatible
• kiota update

Let me know if you need more information or I can help in any way.

Thanks.

[bmarkovic17]

EDIT: After further analysis I've noticed that the issue is actually related to the --clean-output parameter as kiota update --clean-output also started generating wrong models. When I remove the parameter, model generation mostly works okay, but from time to time fails regardless.

Here is an example with a faulty model even though the --clean-output parameter was not present.

[baywet]

Hi @bmarkovic17
Thanks for using kiota and for reaching out.
I'm pretty sure this has nothing to do with which command or argument you pass in but rather because of this in the description:

"TeamAndCompetitor": {
				"allOf": [
					{
						"$ref": "#/components/schemas/Competitor"
					},
					{
						"$ref": "#/components/schemas/Team"
					}
				]
			},

Long story short, kiota implements the following rules for the schema elements:

kind	scenario	result
allOf	single entry	not meaningful, will flatten the and use the single entry
allOf	one inline entry, one reference entry	inheritance
allOf	multiple inline, or multiple reference entries	not really implemented, it should probably deterministically elect one reference as the parent and fall back to an inheritance case
oneOf	single entry	not meaningful, will flatten the and use the single entry
oneOf	no/single/multiple inline entries and/or no/single/multiple reference entries	will generate inline types for the inline schemas, will generate an exclusive union type
anyOf	single entry	not meaningful, will flatten the and use the single entry
anyOf	no/single/multiple inline entries and/or no/single/multiple reference entries	will generate inline types for the inline schemas, will generate an inclusive union type

The main logic behind all of this is here

kiota/src/Kiota.Builder/KiotaBuilder.cs

Line 1806 in d15ad50

private CodeTypeBase? CreateModelDeclarations(OpenApiUrlTreeNode currentNode, OpenApiSchema schema, OpenApiOperation? operation, CodeElement parentElement, string suffixForInlineSchema, OpenApiResponse? response = default, string typeNameForInlineSchema = "", bool isRequestBody = false)

And you can duplicate and tweak this unit test

kiota/tests/Kiota.Builder.Tests/KiotaBuilderTests.cs

Line 3596 in d15ad50

public void UnionOfInlineSchemasWorks()

With that context in mind, is this something you'd be willing to work on a pull request for?

[bmarkovic17]

Thanks for the clarification @baywet.
I can't commit any serious time at the moment but I'll look into it during the weekends.

[baywet]

@bmarkovic17 can you test this again with the latest preview from yesterday and confirm whether you observe the problem? We've made significant improvements to the handling of allof edge scenarios with #4668 and #4381

[bmarkovic17]

@baywet Sorry for my late reply, we were busy with some other stuff that had a narrow deadline.
Today I found some time to test this again and I'm happy to say that with kiota v1.15.0 I couldn't reproduce it. The generated model was always the same, regardless how I run kiota.

Thanks.