private void Process(ComponentIntermediateNode node)
{
// First collect all of the information we have about each type parameter
//
// Listing all type parameters that exist
var bindings = new Dictionary <string, Binding>();
foreach (var attribute in node.Component.GetTypeParameters())
{
bindings.Add(attribute.Name, new Binding()
{
Attribute = attribute,
});
}
// Listing all type arguments that have been specified.
var hasTypeArgumentSpecified = false;
foreach (var typeArgumentNode in node.TypeArguments)
{
hasTypeArgumentSpecified = true;
var binding = bindings[typeArgumentNode.TypeParameterName];
binding.Node = typeArgumentNode;
binding.Content = GetContent(typeArgumentNode);
}
if (hasTypeArgumentSpecified)
{
// OK this means that the developer has specified at least one type parameter.
// Either they specified everything and its OK to rewrite, or its an error.
if (ValidateTypeArguments(node, bindings))
{
var mappings = bindings.ToDictionary(kvp => kvp.Key, kvp => kvp.Value.Content);
RewriteTypeNames(_pass.TypeNameFeature.CreateGenericTypeRewriter(mappings), node);
}
return;
}
// OK if we get here that means that no type arguments were specified, so we will try to infer
// the type.
//
// The actual inference is done by the C# compiler, we just emit an a method that represents the
// use of this component.
// Since we're generating code in a different namespace, we need to 'global qualify' all of the types
// to avoid clashes with our generated code.
RewriteTypeNames(_pass.TypeNameFeature.CreateGlobalQualifiedTypeNameRewriter(bindings.Keys), node);
//
// We need to verify that an argument was provided that 'covers' each type parameter.
//
// For example, consider a repeater where the generic type is the 'item' type, but the developer has
// not set the items. We won't be able to do type inference on this and so it will just be nonsense.
var attributes = node.Attributes.Select(a => a.BoundAttribute).Concat(node.ChildContents.Select(c => c.BoundAttribute));
foreach (var attribute in attributes)
{
if (attribute == null)
{
// Will be null for attributes set on the component that don't match a declared component parameter
continue;
}
// Now we need to parse the type name and extract the generic parameters.
//
// Two cases;
// 1. name is a simple identifier like TItem
// 2. name contains type parameters like Dictionary<string, TItem>
if (!attribute.IsGenericTypedProperty())
{
continue;
}
var typeParameters = _pass.TypeNameFeature.ParseTypeParameters(attribute.TypeName);
if (typeParameters.Count == 0)
{
bindings.Remove(attribute.TypeName);
}
else
{
for (var i = 0; i < typeParameters.Count; i++)
{
var typeParameter = typeParameters[i];
bindings.Remove(typeParameter.ToString());
}
}
}
// If any bindings remain then this means we would never be able to infer the arguments of this
// component usage because the user hasn't set properties that include all of the types.
if (bindings.Count > 0)
{
// However we still want to generate 'type inference' code because we want the errors to be as
// helpful as possible. So let's substitute 'object' for all of those type parameters, and add
// an error.
var mappings = bindings.ToDictionary(kvp => kvp.Key, kvp => kvp.Value.Content);
RewriteTypeNames(_pass.TypeNameFeature.CreateGenericTypeRewriter(mappings), node);
node.Diagnostics.Add(ComponentDiagnosticFactory.Create_GenericComponentTypeInferenceUnderspecified(node.Source, node, node.Component.GetTypeParameters()));
}
// Next we need to generate a type inference 'method' node. This represents a method that we will codegen that
// contains all of the operations on the render tree building. Calling a method to operate on the builder
// will allow the C# compiler to perform type inference.
var documentNode = (DocumentIntermediateNode)Ancestors[Ancestors.Count - 1];
CreateTypeInferenceMethod(documentNode, node);
}
private void Process(ComponentIntermediateNode node)
{
// First collect all of the information we have about each type parameter
//
// Listing all type parameters that exist
var bindings = new Dictionary <string, Binding>();
var componentTypeParameters = node.Component.GetTypeParameters().ToList();
var supplyCascadingTypeParameters = componentTypeParameters
.Where(p => p.IsCascadingTypeParameterProperty())
.Select(p => p.Name)
.ToList();
foreach (var attribute in componentTypeParameters)
{
bindings.Add(attribute.Name, new Binding()
{
Attribute = attribute,
});
}
// Listing all type arguments that have been specified.
var hasTypeArgumentSpecified = false;
foreach (var typeArgumentNode in node.TypeArguments)
{
hasTypeArgumentSpecified = true;
var binding = bindings[typeArgumentNode.TypeParameterName];
binding.Node = typeArgumentNode;
binding.Content = GetContent(typeArgumentNode);
// Offer this explicit type argument to descendants too
if (supplyCascadingTypeParameters.Contains(typeArgumentNode.TypeParameterName))
{
node.ProvidesCascadingGenericTypes ??= new();
node.ProvidesCascadingGenericTypes[typeArgumentNode.TypeParameterName] = new CascadingGenericTypeParameter
{
GenericTypeNames = new[] { typeArgumentNode.TypeParameterName },
ValueType = typeArgumentNode.TypeParameterName,
ValueExpression = $"default({binding.Content})",
};
}
}
if (hasTypeArgumentSpecified)
{
// OK this means that the developer has specified at least one type parameter.
// Either they specified everything and its OK to rewrite, or its an error.
if (ValidateTypeArguments(node, bindings))
{
var mappings = bindings.ToDictionary(kvp => kvp.Key, kvp => kvp.Value.Content);
RewriteTypeNames(_pass.TypeNameFeature.CreateGenericTypeRewriter(mappings), node);
}
return;
}
// OK if we get here that means that no type arguments were specified, so we will try to infer
// the type.
//
// The actual inference is done by the C# compiler, we just emit an a method that represents the
// use of this component.
// Since we're generating code in a different namespace, we need to 'global qualify' all of the types
// to avoid clashes with our generated code.
RewriteTypeNames(_pass.TypeNameFeature.CreateGlobalQualifiedTypeNameRewriter(bindings.Keys), node);
//
// We need to verify that an argument was provided that 'covers' each type parameter.
//
// For example, consider a repeater where the generic type is the 'item' type, but the developer has
// not set the items. We won't be able to do type inference on this and so it will just be nonsense.
foreach (var attribute in node.Attributes)
{
if (attribute != null && TryFindGenericTypeNames(attribute.BoundAttribute, out var typeParameters))
{
var attributeValueIsLambda = _pass.TypeNameFeature.IsLambda(GetContent(attribute));
var provideCascadingGenericTypes = new CascadingGenericTypeParameter
{
GenericTypeNames = typeParameters,
ValueType = attribute.BoundAttribute.TypeName,
ValueSourceNode = attribute,
};
foreach (var typeName in typeParameters)
{
if (supplyCascadingTypeParameters.Contains(typeName))
{
// Advertise that this particular inferred generic type is available to descendants.
// There might be multiple sources for each generic type, so pick the one that has the
// fewest other generic types on it. For example if we could infer from either List<T>
// or Dictionary<T, U>, we prefer List<T>.
node.ProvidesCascadingGenericTypes ??= new();
if (!node.ProvidesCascadingGenericTypes.TryGetValue(typeName, out var existingValue) ||
existingValue.GenericTypeNames.Count > typeParameters.Count)
{
node.ProvidesCascadingGenericTypes[typeName] = provideCascadingGenericTypes;
}
}
if (attributeValueIsLambda)
{
// For attributes whose values are lambdas, we don't know whether or not the value
// covers the generic type - it depends on the content of the lambda.
// For example, "() => 123" can cover Func<T>, but "() => null" cannot. So we'll
// accept cascaded generic types from ancestors if they are compatible with the lambda,
// hence we don't remove it from the list of uncovered generic types until after
// we try matching against ancestor cascades.
if (bindings.TryGetValue(typeName, out var binding))
{
binding.CoveredByLambda = true;
}
}
else
{
bindings.Remove(typeName);
}
}
}
}
// For any remaining bindings, scan up the hierarchy of ancestor components and try to match them
// with a cascaded generic parameter that can cover this one
List <CascadingGenericTypeParameter> receivesCascadingGenericTypes = null;
foreach (var uncoveredBindingKey in bindings.Keys.ToList())
{
var uncoveredBinding = bindings[uncoveredBindingKey];
foreach (var candidateAncestor in Ancestors.OfType <ComponentIntermediateNode>())
{
if (candidateAncestor.ProvidesCascadingGenericTypes != null &&
candidateAncestor.ProvidesCascadingGenericTypes.TryGetValue(uncoveredBindingKey, out var genericTypeProvider))
{
// If the parameter value is an expression that includes multiple generic types, we only want
// to use it if we want *all* those generic types. That is, a parameter of type MyType<T0, T1>
// can supply types to a Child<T0, T1>, but not to a Child<T0>.
// This is purely to avoid blowing up the complexity of the implementation here and could be
// overcome in the future if we want. We'd need to figure out which extra types are unwanted,
// and rewrite them to some unique name, and add that to the generic parameters list of the
// inference methods.
if (genericTypeProvider.GenericTypeNames.All(GenericTypeIsUsed))
{
bindings.Remove(uncoveredBindingKey);
receivesCascadingGenericTypes ??= new();
receivesCascadingGenericTypes.Add(genericTypeProvider);
// It's sufficient to identify the closest provider for each type parameter
break;
}
bool GenericTypeIsUsed(string typeName) => componentTypeParameters
.Select(t => t.Name)
.Contains(typeName, StringComparer.Ordinal);
}
}
}
// There are two remaining sources of possible generic type info which we consider
// lower-priority than cascades from ancestors. Since these two sources *may* actually
// resolve generic type ambiguities in some cases, we treat them as covering.
//
// [1] Attributes given as lambda expressions. These are lower priority than ancestor
// cascades because in most cases, lambdas don't provide type info
foreach (var entryToRemove in bindings.Where(e => e.Value.CoveredByLambda).ToList())
{
// Treat this binding as covered, because it's possible that the lambda does provide
// enough info for type inference to succeed.
bindings.Remove(entryToRemove.Key);
}
// [2] Child content parameters, which are nearly always defined as untyped lambdas
// (at least, that's what the Razor compiler produces), but can technically be
// hardcoded as a RenderFragment<Something> and hence actually give type info.
foreach (var attribute in node.ChildContents)
{
if (TryFindGenericTypeNames(attribute.BoundAttribute, out var typeParameters))
{
foreach (var typeName in typeParameters)
{
bindings.Remove(typeName);
}
}
}
// If any bindings remain then this means we would never be able to infer the arguments of this
// component usage because the user hasn't set properties that include all of the types.
if (bindings.Count > 0)
{
// However we still want to generate 'type inference' code because we want the errors to be as
// helpful as possible. So let's substitute 'object' for all of those type parameters, and add
// an error.
var mappings = bindings.ToDictionary(kvp => kvp.Key, kvp => kvp.Value.Content);
RewriteTypeNames(_pass.TypeNameFeature.CreateGenericTypeRewriter(mappings), node);
node.Diagnostics.Add(ComponentDiagnosticFactory.Create_GenericComponentTypeInferenceUnderspecified(node.Source, node, node.Component.GetTypeParameters()));
}
// Next we need to generate a type inference 'method' node. This represents a method that we will codegen that
// contains all of the operations on the render tree building. Calling a method to operate on the builder
// will allow the C# compiler to perform type inference.
var documentNode = (DocumentIntermediateNode)Ancestors[Ancestors.Count - 1];
CreateTypeInferenceMethod(documentNode, node, receivesCascadingGenericTypes);
}
