Circular reference error when adding `MutablePersistableRecord` conformance.

[scottrhoyt]

I'm new to GRDB so it's possible I did something wrong here, but I cannot add MutablePersistableRecord conformance to my Codable structs without getting a circular reference. I copy/pasted the Authors and Books models from the recommend practices documentation to a minimal project to repro the issue. GRDB v. 7.7.1 and 7.8.0 had the same issue. Interestingly, when I use the same record objects in a minimal SPM package I do not see this, so it's possibly dependent on some of the compiler settings or linked frameworks unique to a new Xcode 26 project.

[groue]

Hello @scottrhoyt,

I suppose your project uses the MainActor isolation by default: check the SWIFT_DEFAULT_ACTOR_ISOLATION in project settings. This is a new compiler feature (SE-0466), disabled by default in new packages, but enabled by default in new Xcode projects.

I did my duty, starting GRDB 7.7.0, for supporting this feature, by marking the GRDB protocols that can not support isolated conformances as SendableMetaType.

Unfortunately this is not enough to fix the pile of compiler errors brought by this feature.

Your code might look like:

import GRDB

struct Player { // MainActor-isolated
    var id: Int64?
    var name: String
}

// ❌ Circular reference
extension Player: Codable, FetchableRecord, MutablePersistableRecord { }

Even if your type opts out of MainActor isolation with nonisolated, new errors appears:

nonisolated struct Player { // nonisolated
    var id: Int64?
    var name: String
}

// ❌ Type 'Player' does not conform to protocol 'EncodableRecord'
// ❌ Main actor-isolated instance method 'encode(to:)' cannot satisfy nonisolated requirement
extension MyRecord2: Codable, FetchableRecord, MutablePersistableRecord { }

In my current knowledge, just opt out of MainActor isolation, or define record types in a package that does not have this feature, and import them in your app. I know those are not satisfying solutions, but that's the best I can tell.

[EDIT] Please see the comment below for the recommended fix.

And if you have time, please open issues at http://github.com/apple/swift, so that they can be addressed by compiler engineers.

BTW, I also did my duty before Swift 6.2 shipped, by warning with insistance during the review of [Amendment] SE-0466 Control default actor isolation inference. My concerns were discarded without any semblance of justification. But I did not expect how bad it would be, with errors that just can not be fixed.

Maybe @DougGregor or @hborla now better?

[scottrhoyt]

Understandable, thank you!

[groue]

You're welcome! I'm probably more frustrated than you are (assuming you can use of the suggested workarounds) because this fresh compiler feature ruins the experience of GRDB users. We'll keep the issue open for a while, in case somebody (maybe the authors of the feature) suggest a better solution 🤞

[groue]

Oh, @scottrhoyt, I'm sorry, I have a better solution, which is the one that is intended by the feature authors. Declare the conformance right on the type declaration:

-struct Player { ... }
-extension Player: Codable, FetchableRecord, MutablePersistableRecord { ... }

+struct Player: Codable, FetchableRecord, MutablePersistableRecord { ... }

(I still think that this is contrary to reasonable expectations, well-organized code, and useful programming patterns enabled by extensions, but well, bad ideas sometimes do ship.)

[scottrhoyt]

That does the trick! Thanks for the followup.

I hope there is (or emerges) good documentation on how/why where a conformance is declared affects the compiler in this new world as it does break longstanding expectations and practices around extensions--and that's on the swift team. Frankly, it's all a bit too black box and restrictive for my taste. Certainly at least the error messages could be more helpful.

While this issue is bigger than GRDB, it may be helpful to add a note to the documentation here as I expect many to hit a similar issue if nothing else changes.

[groue]

Thank you for the suggestion. Now I'm not sure any note in the GRDB doc would actually help many users solve the compiler error without opening an issue like this one. Nobody expects those errors, nobody expects that they are related to MainActor isolation, nobody expects that extensions were broken by the language itself, and people do not carefully read a note when they don't even know it has any relevance to a present or future situation.

And I will not update hundreds of perfectly fine sample code, progressive documentation pieces, and existing issues and discussions, by moving the conformances up to the type declaration. That would be an unreasonable amount of work, and the end result would be documentation and code of worse quality. I'm not responsible for the incompatible language dialects we have to deal with today.

That's how bad bad design is 🤷‍♂️

[groue]

I am closing the case because I find it extremely difficult to degrade GRDB due to the poorly designed SE-0466.

Other pull requests of interest:

• Update examples to use default main actor pointfreeco/sqlite-data#248 (demo app is isolated to MainActor by default)
• Enforce database functions to be nonisolated pointfreeco/swift-structured-queries#205 adds notes about default MainActor isolation)