Fully support complex types with the InMemory provider

[roji]

Some query/materialization support is missing, see tests DbContextTest.SaveChanges_calls_DetectChanges_by_default, DbContextTest.Auto_DetectChanges_for_SaveChanges_can_be_switched_off

[ajcvickers]

Note from triage: we are very unlikely to implement this. See https://github.com/dotnet/efcore/issues/18457#issuecomment-615411282

[aliastopan]

Is the KeyNotFoundException part of the reason why ComplexProperty has yet compatible with in-memory database?

"title": "KeyNotFoundException", "code": 500, "source": "System.Private.CoreLib", "detail": "The given key 'Property: UserAccount.User#User.EmailAddress (string) Required MaxLength(255)' was not present in the dictionary.", "stackTrace": [ " at TValue System.Collections.Generic.Dictionary<TKey, TValue>.get_Item(TKey key)", " at Expression Microsoft.EntityFrameworkCore.InMemory.Query.Internal.InMemoryShapedQueryCompilingExpressionVisitor+ShaperExpressionProcessingExpressionVisitor.VisitMethodCall(MethodCallExpression methodCallExpression)",

[roji]

@aliastopan support hasn't been done for InMemory, this specific KeyNotFoundException may be part of that (or not). We're unlikely to implement support, since using the InMemory database for testing is discouraged - consider testing against your containerized database (e.g. SQL Server) using testcontainers.net.

[IanWitham]

Is this documented somewhere? I've just sunk hours into trying to "solve" this before I came across this thread.

[roji]

@IanWitham yes, the InMemory provider documentation contains the following note:

New features are not being added to the in-memory database.

Aside from the numerous other warnings and notices to stay away from the provider in general.

[marchy]

@roji the right guidance to give on that page would be to suggest people to use the SQLite with in-memory option set.

There are definitely gaps there as well, but they are more or less workable and has been a godsend to enabling us to work in a rapid development model for new features (ie: not having to worry about migrating a live DB until you have the full domain model for that feature built out)

[roji]

the right guidance to give on that page would be to suggest people to use the SQLite with in-memory option set.

We generally recommend against using database fakes, either InMemory or Sqlite - though Sqlite is indeed a much better option. In today's world, setting up SQL Server or PG via testcontainers and testing against that is both easy and very fast, and so testing against a fake is much less needed than it used to be.

not having to worry about migrating a live DB until you have the full domain model for that feature built out

You can just use EnsureCreated during rapid development, including with a real database - no need to deal with migrations.

[marchy]

@roji What about tests? We run all tests on Sqlite DB's that we create rapidly for each test instance (ie: in test creation/tear-down)

Can this run disconnected from the Internet on local and test servers?

It looks like a LOT more setup to pull off – but if that is indeed the guidance there should be a full guide for how to set it up in both local environment and Azure DevOps at the least.

[roji]

@marchy tests is what we're discussing here above (at least what I was discussing).

Can this run disconnected from the Internet on local and test servers?

Of course, the major databases can run in a docker container, which is very easy to manage. As I wrote above, I suggest taking a look at testcontainers, and also reading the EF testing docs which discuss a lot of this stuff.

[jukkahyv]

Of course, the major databases can run in a docker container, which is very easy to manage. As I wrote above, I suggest taking a look at testcontainers, and also reading the EF testing docs which discuss a lot of this stuff.

Can you give a practical example of how it would work in the ideal situation? I mean some medium-large open source website that makes use of testcontainers to efficiently run 100s of tests. Because the problem is, I've only seen bad examples. Projects that try to do it that way, then 2-3 years afterwards it takes 20 minutes to run the tests, tests randomly fail because of database timeouts, and test data management is a nightmare because it's being reused between tests. I'm not against using real database for tests, I just haven't seen it working.

[roji]

@jukkahyv please take a thorough look at the testing docs I linked to above; these are also related to these code samples which should be helpful. I do think we should improve these docs to specifically show testing with testcontainers (they were written before that was a thing), and be more practically-oriented, but the general idea is there.

EF itself runs almost all of its test against real databases - that's around 20k tests that run against SQL Server alone, in a very reasonable amount of time. It's a matter of structuring test databases appropriately, not e.g. creating one for every test, and running everything in parallel while ensuring that modification tests e.g. use transactions which get rolled back, so as to not cause interference with other tests.

[dta5003]

It's totally understandable that no new features are being added to the InMemory provider.

If possible, a better error message/exception would be super helpful when attempting to interact with Complex Types via the InMemory provider. As far as I can tell after many hours yesterday, finding this GitHub issue via searching is the only way to discover the root cause of this "KeyNotFoundException".