Disable default OpenAPIRuntime traits

[madsodgaard]

Disables the default traits for OpenAPIRuntime, so that anyone that depends on urlsession transport, can choose disable the default. runtime traits

Related:

• Move to FoundationEssentials swift-server/swift-openapi-async-http-client#59
• Disable traits for imported runtime swift-openapi-generator#878

[simonjbeaumont]

We're not anticipating that many folks are using this transport on non-Darwin platforms. Are we expecting that dropping the foundation dependency is bringing any value here?

[madsodgaard]

The issue is that if you have a cross-platform lib, where you use URLSession transport on Darwin and AsyncHTTPClient on non-Darwin, like:

.package(url: "https://github.com/apple/swift-openapi-urlsession", from: "1.2.0"),
.package(url: "https://github.com/swift-server/swift-openapi-async-http-client", from: "1.4.0"),
.package(url: "https://github.com/apple/swift-openapi-runtime", from: "1.11.0", traits: []),

The traits: [] won't work, as long as swift-openapi-urlsession also depends on the runtime lib, but with the default traits being enabled. So, we'd have to conditionally include the urlsession transport using something like canImport(Darwin) in the manifest file, which seems suboptimal.

This is even if you conditionally include the target:

.product(
  name: "OpenAPIURLSession",
  package: "swift-openapi-urlsession",
  condition: .when(platforms: [.macOS, .iOS])
),

[simonjbeaumont]

This is even if you conditionally include the target:

This seems like a serious limitation of the trait-based resolution in SwiftPM. Have we filed an issue for that?

[madsodgaard]

@simonjbeaumont Hm, I am not sure that is a "bug". swift-openapi-urlsession depends on openapi-runtime like any other package would. Let's say that swift-openapi-urlsession in fact used some of the features that was enabled by certain traits from the runtime. Then the "outer" package that depended on both, cannot just override that, otherwise urlsession would not compile. Maybe @FranzBusch can chime in on this as well.

[madsodgaard]

@simonjbeaumont Would you be against getting this merged?

[madsodgaard]

@simonjbeaumont Hey, just wanted to check-in on this PR.

We are currently forced to do an ugly #canImport(Darwin) to conditionally include URLSession transport, to prevent linking Foundation on our cross-platform lib.

We've done the same empty traits change for other repos, anything stopping us from doing the same here, and merging this in?

[simonjbeaumont]

Sorry. Still on vacation. I don't mind getting this merged. I kicked the CI. Maybe @guoye-zhang can shepherd it through while I'm out.

[simonjbeaumont]

Looks like something is wrong with the CI related to the import statements:

  |  `- error: consecutive statements on a line must be separated by ';'

[guoye-zhang]

We should probably get #89 merged first, but I don’t have the admin rights to do so

[simonjbeaumont]

OK, I merged #89 and updated this PR branch to unblock progress on this PR.

[simonjbeaumont]

Looks like the C++ interop is unrelated. This PR is good to review now. I'll take a look.

[simonjbeaumont]

@madsodgaard thanks for your patience here while I was OOO. We've now dropped 6.0, so can we remove the additional manifest? Otherwise this LGTM.

[madsodgaard]

@simonjbeaumont no problem! I removed the 6.0 manifest

[simonjbeaumont]

Thanks @madsodgaard!