Linking https://github.com/microsoft/testfx/blob/main/docs/RFCs/007-DataSource-Attribute-VS-ITestDataSource.md.

Ideally we want the new attribute to implement the same interface as the other source attributes. Outside of technical constraints, we would also like to unfold arguments and declare them on the method signature instead of using the TestContext.DataRow[".."] syntax.

Finally, I think we should drop the DataAccessMethod and consider these sources the same way we consider other parameterized tests.

@Evangelink Thanks. I updated the proposal a little bit.

One consideration for not having TestContext.DataRow[...] is how the users are using data sources. Are there any valid scenarios where the XML could contain much more data than needed?

If we go with parameters instead of TestContext.DataRow, users will have to declare everything in the correct order, and it will be noisy if part of the schema is unused for the purpose of the test. I'm not sure if that's how users are using DataSource though.

In that case, we may be able to have the test parameter as DataRow? e.g:

[TestMethod]
[XmlDataSource("MyTests.Configuration.xml", "MyTable1")]
public void TestMethod(DataRow dataRow /*alternative of TestContext.DataRow*/)
{
}

That's a good remark, we probably need wrapper type yeah.

@Evangelink Is it "our own new wrapper type"? Or could we just use "DataRow"? (available in all .NET versions without extra packages)

My 5 cents, from our side and our migration to .net 8 (and continue building .net framework in parallel of...)

We took the following approach:

The XML file looks like :

The intent was, sometimes the data in the XML need to be parsed differently :

The base class look like this

and the GetData on the datasource folder:

Which then the test look likes:

Thanks for your input @fforjan. I read it as our solution wouldn't be helping you much, am I right?

Giving some more thoughts on it, I am actually wondering if this feature is actually bringing much to our users.

I mean that either we will end up in a too generic solution where user is basically left with a too generic argument type or something too narrow where the user ends up having to write its own extension.

For XML, I don't think we are really providing much value as (nearly) everything is already available out-of-the-box. For CSV, we would need to either re-implement a parser which is a no-go or rely on some library from community (with licensing/security/support concerns) but even there, I have the feeling we will either have too generic type or too narrow which means we don't have the right abstraction. Finally, for databases, this seems to be even more true.

I am more tempted to invest some time in offering a different fixture experience that would give the flexibility to use the library and srong typing the user wants.

I'll sleep on it.

I read it as our solution wouldn't be helping you much, am I right?
yes, the painful part in my solution was to implement the GetData method and as we were successful, we didn't even think you were looking to port (and raise such request!) the original DataSourceAttribute from the .net Framework - we were using it but had to move away to allow the tests to run in both fwk and core.

In our case, we decided that the xml file would be read from the filesystem as such, relative to the cs:

We could have decided to read the file where the DLLs is executed or we could also have read the xml file as a stream from a resource from an assembly. (in the end, would the DataSourceAttribute consider the reference as a DeploymentItem implicit reference too, would the file be an explicit deployment item too?)

_ I have the feeling we will either have too generic type or too narrow which means we don't have the right abstraction. Finally, for databases, this seems to be even more true._
I would agree with you, same if you unit test is a sqllite file, where do you read it (see the previous remark on the deploy item) ? Where do you store the credential for a real DB ? etc.. I think everyone would have a different answer.

What would be helpful potentially is more end to end sample/simple for the different use case - XML/CSV/DB- which people can adapt from ?