[SwiftBindings] Added post processing step to compute struct frozenness #2912

jkurdek · 2025-01-08T17:39:33Z

Implements #2886

Initialises a post ABI parsing step which calculates information about types and populates type database.

Follow-ups:

Add tooling sorting modules in topological order for processing
Remove type database usage in ABI parsing
Make type database immutable after post-processing step

stephen-hawley

Love to see the tests in place. This is a good start, but there are some things that need to be addressed. If it's not practical to resolve them, open issues and keep going.

stephen-hawley · 2025-01-08T20:20:14Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+            ProcessStructFields(structDecl);
+
+            // TODO: Remove loading dylib
+            IntPtr metadataPtr = DynamicLibraryLoader.invoke(_dylibPath, $"{structDecl.MangledName}Ma");


As we discussed, we shouldn't do this.

I agree. This will be the next step. Here is a tracking issue #2914. Should also cover the comments about blittability and size, stride, alignment.

stephen-hawley · 2025-01-08T20:22:10Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+        /// <returns>True if blittable, false otherwise.</returns>
+        private unsafe bool EvaluateBlittability(SwiftTypeInfo swiftTypeInfo)
+        {
+            // TODO: Replace with a manual approach.


👍 Is there an issue for this?

stephen-hawley · 2025-01-08T20:24:53Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+        {
+            _typeDatabase.Registrar.RegisterType(namedTypeSpec.Module,
+                                                 namedTypeSpec.NameWithoutModule,
+                                                 new TypeRecord


This is a separate issue, but we will need the size, stride, and alignment as well.

stephen-hawley · 2025-01-08T20:26:51Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+        /// <param name="structDecl">The struct declaration.</param>
+        private void ProcessStruct(NamedTypeSpec namedTypeSpec, StructDecl structDecl)
+        {
+            // Ensure all fields are processed or known in the database.


All nominal types can have inner types. Because fields in the structure may depend on them, we should process any inner types before we process fields.

They are processed during ABI parsing so we have full information. Added test for that and patched some logic.

stephen-hawley · 2025-01-08T20:35:00Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+                                                 namedTypeSpec.NameWithoutModule,
+                                                 new TypeRecord
+                                                 {
+                                                     Namespace = namedTypeSpec.Module,


I'd like to see a separate type for holding a type name. Something like:

public class /* or record or struct */ TypeName { public string Namespace { get; init; } public string Name { get; init; } public string FullName { get; } => Namespace + "." + Name; // if not a record, make ToString, Equals, etc }

This will make it easier to set up import statements in Swift and using statements in C#.

stephen-hawley · 2025-01-08T20:37:52Z

src/Swift.Bindings/src/Parser/SwiftABIParser.cs

@@ -146,7 +146,7 @@ public ModuleDecl GetModuleDecl()

            moduleDecl.Fields = decls.OfType<FieldDecl>().ToList();
            moduleDecl.Methods = decls.OfType<MethodDecl>().ToList();
-            moduleDecl.Declarations = decls.Where(d => !(d is MethodDecl) && !(d is FieldDecl)).ToList();
+            moduleDecl.Types = decls.OfType<TypeDecl>().ToDictionary(d => new NamedTypeSpec(d.FullyQualifiedName), d => d);


One thing that we will run into right here is that the *Decl types are already C# and that type may not be the same as the Swift from which it came. For example, if we get a Swift type that is unpronounceable in C# (eg, contains emoji) then we've just created an incorrect NamedTypesSpec.

stephen-hawley · 2025-01-08T20:39:46Z

src/Swift.Bindings/src/Parser/SwiftABIParser.cs

-            typeRecord.IsProcessed = true;
-
-            if (node.Children != null && decl != null)
+            if (decl != null)


favor is not null over != null to avoid possible issues that arise from operator implementations.

jkurdek · 2025-01-09T09:17:30Z

CI failures unrelated - #2896 broke compilation. should be fixed once #2909 gets in

kotlarmilos · 2025-01-09T09:36:58Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

@@ -0,0 +1,223 @@
+using Swift.Runtime;


Suggested change

using Swift.Runtime;

// Copyright (c) Microsoft Corporation.

// Licensed under the MIT License.

using Swift.Runtime;

kotlarmilos · 2025-01-09T10:02:58Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+    /// Calculates properties on types which are not directly available from the ABI.
+    /// Generates type database entries for structs, enums, and classes.
+    /// </summary>
+    internal class ModuleProcessor


I like the idea of having a processor/marshaller that based on ABI and dylib inputs, registers types in the database. For that, we need to wire up a Mach-O reader and a demangler.

kotlarmilos · 2025-01-09T10:06:59Z

src/Swift.Bindings/src/Program.cs

                // Register the module and set the filter
                var moduleRecord = typeDatabase.Registrar.RegisterModule(moduleName);
                moduleRecord.Path = dylibPath;

                // Parse the Swift ABI file and generate declarations
                var decl = swiftParser.GetModuleDecl();

+                var moduleProcessor = new ModuleProcessor(moduleName, dylibPath, decl.Types, typeDatabase);
+                moduleProcessor.FinalizeTypeProcessing();


If ModuleProcessor is exclusively creating a type registrar, should it be returned here?

kotlarmilos · 2025-01-09T10:08:11Z

src/Swift.Bindings/src/Parser/ModuleProcessor.cs

+                    break;
+
+                default:
+                    throw new NotSupportedException($"Unknown or unsupported type declaration '{typeDecl.GetType().Name}'.");


It should gracefully handle unknown types to generate partial bindings.

jkurdek added 4 commits January 2, 2025 15:41

Removed unnecessary nulls

8f64055

Moved type database to BindingsGeneration

ac37677

Initialized tests for frozenness detection

6fc4c02

Added post abi parsing step

ce540e1

jkurdek added the area-SwiftBindings Swift bindings for .NET label Jan 8, 2025

jkurdek requested review from kotlarmilos and stephen-hawley January 8, 2025 17:39

jkurdek self-assigned this Jan 8, 2025

jkurdek changed the title ~~[SwiftBindings]~~ [SwiftBindings] Added post processing step to compute struct frozenness Jan 8, 2025

stephen-hawley requested changes Jan 8, 2025

View reviewed changes

kotlarmilos reviewed Jan 9, 2025

View reviewed changes

Added support for nested structs

2198c9c

stephen-hawley approved these changes Jan 9, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SwiftBindings] Added post processing step to compute struct frozenness #2912

[SwiftBindings] Added post processing step to compute struct frozenness #2912

jkurdek commented Jan 8, 2025

stephen-hawley left a comment

stephen-hawley Jan 8, 2025

jkurdek Jan 9, 2025

stephen-hawley Jan 8, 2025

stephen-hawley Jan 8, 2025

stephen-hawley Jan 8, 2025

jkurdek Jan 9, 2025

stephen-hawley Jan 8, 2025

stephen-hawley Jan 8, 2025

stephen-hawley Jan 8, 2025

jkurdek commented Jan 9, 2025

kotlarmilos Jan 9, 2025

kotlarmilos Jan 9, 2025

kotlarmilos Jan 9, 2025

kotlarmilos Jan 9, 2025

[SwiftBindings] Added post processing step to compute struct frozenness #2912

Are you sure you want to change the base?

[SwiftBindings] Added post processing step to compute struct frozenness #2912

Conversation

jkurdek commented Jan 8, 2025

stephen-hawley left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jkurdek commented Jan 9, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment