Add ILLink/ILCompiler support for feature checks

[sbomer]

This will substitute static boolean properties with false when decorated with [FeatureGuard(typeof(SomeFeature))], where SomeFeature is known to be disabled.

The "feature" of unreferenced code, represented by RequiresUnreferencedCodeAttribute, is always considered disabled. For ILC, RequiresDynamicCodeAttribute and RequiresAssemblyFilesAttribute are also disabled. ILLink also respects the feature switch for IsDynamicCodeSupported to disable the RequiresDynamicCodeAttribute feature.

We don't want this kicking in when trimming in library mode, so this adds an ILLink option to disable the optimization.

Additionally, a property is substituted if it has [FeatureSwitchDefinition("SomeFeatureName")] and "SomeFeatureName" is set in the command-line arguments.

XML substitutions take precedence over this behavior.

Includes a few tweaks and cleanup to the analyzer logic added in #94944.

See #94625 for notes on the design.
See #96859 for the API proposal.

[ghost]

Tagging subscribers to this area: @agocke, @sbomer, @vitek-karas
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This will substitute static boolean properties with false when decorated with [FeatureCheck(typeof(SomeFeature))], where SomeFeature is known to be disabled.

The "feature" of unreferenced code, represented by RequiresUnreferencedCodeAttribute, is always considered disabled. For ILC, RequiresDynamicCodeAttribute and RequiresAssemblyFilesAttribute are also disabled.

Additionally, a feature is considered disabled if the type has [FeatureSwitchDefinition("SomeFeatureName")] and "SomeFeatureName" is set to false in the command-line arguments.

We don't want this kicking in when trimming in library mode, so this adds an ILLink option to disable the optimization.

XML substitutions take precedence over this behavior.

Includes a few tweaks and cleanup to the analyzer logic added in #94944.

See #94625 for notes on the design.
See #96859 for the API proposal.

Author:	sbomer
Assignees:	-
Labels:	area-Tools-ILLink
Milestone:	-

[sbomer]

I will update this to match the approved API shape once #99340 is merged.

[sbomer]

This is ready for review.

[jtschuster]

Thank you!

[jtschuster]

Thank you!

[MichalStrehovsky]

Native AOT side LGTM modulo the questions. Thanks!

[MichalStrehovsky]

Just double checking that we don't allow feature types to be generic/arrays/pointers (non-definition type will never be EcmaType).

[sbomer]

The feature types aren't restricted in any way, but everything will be a no-op except for RequiresUnreferencedCodeAttribute, RequiresDynamicCodeAttribute, and RequiresAssemblyFilesAttribute. A third-party analyzer could theoretically add support for other types (including generics) so we don't block other types.

[MichalStrehovsky]

Do we need this? IsDynamicCodeSupported/Compiled is hardcoded to false in native AOT corelib.

[sbomer]

I did this to match the behavior of some substitutions I noticed for System.Linq, that only kick in when the feature is explicitly passed on the command-line. It's kind of a moot point unless somebody tries to set DynamicCodeSupport to true in an AOT app.

[MichalStrehovsky]

Hm, that's an unintended consequence of #89638. I don't think we should allow IsDynamicCodeSupported to be set to true. It would produce unactionable warnings (we don't have user-facing warning messages for these) and a broken app.

Submitted #99590 to ban this.

[sbomer]

Thanks!