CLI: don't enable default features in --no-default-features builds

[nicoburns]

• Depends on CLI: only push a dioxus/* fallback feature if the crate being compiled depends on the dioxus crate. #5143. Seperate PRs because I'm less sure about this fix.

The CLI currently enables "renderless" default features (default features which don't enable a renderer) even when --no-default-features is specified as a command line argument. It also unconditionally set the variable tracking the --no-default-features flag to true unless the renderer is Unknown.

This PR removes the code modifying the --no-default-features flag, and PR inverts the condition so that default features are enabled in the --no-default-features flag is not set.

[nicoburns]

@jkelleyrtp This is failing CI on the following test:

        TestHarnessBuilder::new("harness-default-to-non-default")
            .deps(r#"dioxus = { workspace = true, features = [] }"#)
            .fetr(r#"default=["web"]"#)
            .fetr(r#"web=["dioxus/web"]"#)
            .asrt(
                r#"dx build --ios"#,
                |targets| async move {
                    let t = targets.unwrap();
                    assert!(t.server.is_none());
                    assert_eq!(t.client.bundle, BundleFormat::Ios);
                    assert_eq!(t.client.triple, TestHarnessBuilder::host_ios_triple_sim());
                    assert!(t.client.no_default_features);
                },
            ),

In my mind, the test is wrong because dx build --ios shouldn't imply no_default_features. That should only be configured if you run something like dx build --ios --no-default-features. However, I'm wondering if the no_default_features setting has some other meaning within the Dioxus CLI that I'm missing?

[jkelleyrtp]

@jkelleyrtp This is failing CI on the following test:

        TestHarnessBuilder::new("harness-default-to-non-default")
            .deps(r#"dioxus = { workspace = true, features = [] }"#)
            .fetr(r#"default=["web"]"#)
            .fetr(r#"web=["dioxus/web"]"#)
            .asrt(
                r#"dx build --ios"#,
                |targets| async move {
                    let t = targets.unwrap();
                    assert!(t.server.is_none());
                    assert_eq!(t.client.bundle, BundleFormat::Ios);
                    assert_eq!(t.client.triple, TestHarnessBuilder::host_ios_triple_sim());
                    assert!(t.client.no_default_features);
                },
            ),

In my mind, the test is wrong because dx build --ios shouldn't imply no_default_features. That should only be configured if you run something like dx build --ios --no-default-features. However, I'm wondering if the no_default_features setting has some other meaning within the Dioxus CLI that I'm missing?

The reason we specify no_default_features here is because --ios is overriding the default web platform, so we need to strip it down and then add back everything except for the web feature.

Passing --ios implies --features mobile if it exists, but we need to disable web to get the feature set we want.

[nicoburns]

The reason we specify no_default_features here is because --ios is overriding the default web platform, so we need to strip it down and then add back everything except for the web feature.

I believe in the current version of the code this happens even without the no_default_features flag being set. Scroll down to where the no_default_features variable is read, and you'll see that the "feature set" it controls is the "renderless" set of features (which I believe excludes any feature which enables a renderer)..

[nicoburns]

@jkelleyrtp This is now passing CI, and the following commands in the Dioxus repo all work:

• dx serve --example todomvc (uses default desktop renderer)
• dx serve --example todomvc --web (uses web renderer, doesn't enable desktop feature)
• dx serve --example todomvc --renderer native (uses native renderer, doesn't enable desktop feature)

[ealmloff]

This seems the right behavior. The current version has inverted logic for default features. I added a test to prevent regressions like this in the future