Roslyn Source Generators: Reduction of Resource Consumption in IDEs – Part 9

In this article we will see how to reduce the resource consumption of a Source Generator when running inside an IDE by redirecting the code generation to RegisterImplementationSourceOutput.

In this article:

Pawel Gerr is architect consultant at Thinktecture. He focuses on backends with .NET Core and knows Entity Framework inside out.

More information about the Smart Enums and the source code can be found on GitHub:

After a detour to the new high-level in previos article (TODO: link), we will see how to reduce the resource consumption of a Source Generator when running inside an IDE by redirecting the code generation to RegisterImplementationSourceOutput.

Roslyn API: RegisterImplementationSourceOutput

With newer compiler versions, Roslyn team introduced the method RegisterImplementationSourceOutput to give developers a means to split the code generation in 2 categories: the code that can be used during development directly and the code which is important during runtime of the application only. If the code must be able to be referenced by the developers, like property Items generated in the first part of this series, then continue using RegisterSourceOutput, otherwise use RegisterImplementationSourceOutput. An IDE has a choice whether to run the pipeline registered with RegisterImplementationSourceOutput during development in the background or to skip the code generation until build (of the assembly) as a performance optimization.

At the time of writing, as I can see, the IDEs doesn’t treat RegisterImplementationSourceOutput any different than RegisterSourceOutput. But this should change in the future to be able to handle growing number of Roslyn Source Generators.

Pipeline for Generation of Translations

The TranslationAttribute, generated in the part 7, is nothing that can be used by a developer, so, it is a perfect candidate for RegisterImplementationSourceOutput.

Everything that is related to translations is moved to a new generation pipeline in InitializeTranslationsGenerator. The pipeline became more fine grained to leverage the efficiency of memoization described in part 4.

					   public void Initialize(IncrementalGeneratorInitializationContext context) 
      var enumTypes = context.SyntaxProvider 
                             .SelectMany((enumInfos, _) => enumInfos.Distinct());
      InitializeTranslationsGenerator(context, enumTypes); 
   private static void InitializeTranslationsGenerator( 
      IncrementalGeneratorInitializationContext context, 
      IncrementalValuesProvider<DemoEnumInfo> enumTypes) 
      // fetch namespace and name only, so the generation is not triggered anew,
      // if something else (unimportant) changes.
      var enumNames = enumTypes.Select((t, _) => (t.Namespace, t.Name)); 
      var mergedTranslations = context.AdditionalTextsProvider 
            .Where(text => text.Path.EndsWith("translations.json",
            .Select((text, token) => text.GetText(token)?.ToString())
            // Parse JSON before generation
            .Select((json, _) => ParseTranslations(json)) 
            .Where(translations => !translations.IsEmpty)
            // Merge translations from different JSON files
      var translationInfos = enumNames.Combine(mergedTranslations) 
            .SelectMany((tuple, _) => 
                var (ns, name) = tuple.Left; 
                var translationsByClassName = tuple.Right; 
                // search for the translations of current smart enum,
                // it there are none, then don't generate anything
                if (!translationsByClassName.TryGetValue(name, out var translations)) 
                    return ImmutableArray<EnumTranslationInfo>.Empty; 
                var translationInfo = new EnumTranslationInfo(ns, name, translations); 
                return ImmutableArray.Create(translationInfo); 
      context.RegisterImplementationSourceOutput(translationInfos, GenerateCode); 
The type EnumTranslationInfo implements Equals and GetHashCode for proper caching.
					public readonly struct EnumTranslationInfo : IEquatable<EnumTranslationInfo> 
   public string? Namespace { get; } 
   public string Name { get; } 
   public ImmutableDictionary<string, string> Translations { get; } 
   public EnumTranslationInfo( 
      string? ns, 
      string name, 
      ImmutableDictionary<string, string> translations) 
      Namespace = ns; 
      Name = name; 
      Translations = translations; 
   // Equals and GetHashCode

Parsing of the JSON file is virtually the same but is using ImmutableDictionary to be able to return empty dictionaries without creation of new instances.
					   private static ImmutableDictionary<string, ImmutableDictionary<string, string>> ParseTranslations(string? json) 
      if (String.IsNullOrWhiteSpace(json)) 
         return ImmutableDictionary<string, ImmutableDictionary<string, string>>.Empty; 
         return JsonConvert.DeserializeObject<ImmutableDictionary<
                                            string, ImmutableDictionary<string, string>>>(json!) 
                ?? ImmutableDictionary<string, ImmutableDictionary<string, string>>.Empty; 
      catch (Exception) 
         return ImmutableDictionary<string, ImmutableDictionary<string, string>>.Empty; 

The source generator must merge the translations if there are multiple JSON files with translations.

					   private static ImmutableDictionary<string, ImmutableDictionary<string, string>> MergeTranslations( 
      ImmutableArray<ImmutableDictionary<string, ImmutableDictionary<string, string>>>
      CancellationToken cancellationToken) 
      if (collectedTranslations.IsDefaultOrEmpty) 
         return ImmutableDictionary<string, ImmutableDictionary<string, string>>.Empty; 
      if (collectedTranslations.Length == 1) 
         return collectedTranslations[0]; 
      var mergedTranslations = ImmutableDictionary<string,
                                                   ImmutableDictionary<string, string>>.Empty; 
      foreach (var translationsByClassName in collectedTranslations) 
         foreach (var kvp in translationsByClassName) 
               var className = kvp.Key; 
               var translationByLanguage = kvp.Value; 
               if (mergedTranslations.TryGetValue(className, out var otherTranslations)) 
                  translationByLanguage = translationByLanguage.AddRange(otherTranslations); 
               mergedTranslations = mergedTranslations.SetItem(className, translationByLanguage); 
            catch (Exception) 
               // Report the error 
      return mergedTranslations; 
The actual code generation has been moved out of DemoCodeGenerator to EnumTranslationsGenerator.
					   private static void GenerateCode(
      SourceProductionContext context,
      EnumTranslationInfo translationInfo)
      var ns = translationInfo.Namespace is null ? null : $"{translationInfo.Namespace}.";
      var code = EnumTranslationsGenerator.Instance.Generate(translationInfo);

      if (!String.IsNullOrWhiteSpace(code))
         context.AddSource($"{ns}{translationInfo.Name}.Translations.g.cs", code);
// -------------------------------
public class EnumTranslationsGenerator 
   public static readonly EnumTranslationsGenerator Instance = new(); 
   public string Generate(EnumTranslationInfo translationsInfo) 
      var ns = translationsInfo.Namespace; 
      var name = translationsInfo.Name; 
      var sb = new StringBuilder(@$"// <auto-generated /> 
#nullable enable 
{(ns is null ? null : $@"namespace {ns}; 
      GenerateTranslationAttributes(sb, translationsInfo.Translations); 
partial class {name} 
      return sb.ToString(); 
   private static void GenerateTranslationAttributes( 
      StringBuilder sb, 
      IReadOnlyDictionary<string, string> translations) 
      foreach (var kvp in translations.OrderBy(kvp => kvp.Key)) 
[global::DemoLibrary.TranslationAttribute(""").Append(kvp.Key).Append("\", \"").Append(kvp.Value).Append("\")]"); 



Even if this feature is not supported (by all) IDEs yet. I recommend start using it now, so our IDEs stay responsive with growing number of Roslyn Source Generators as soon as RegisterImplementationSourceOutput is treated properly.

Current articles, screencasts and interviews by our experts

Don’t miss any content on Angular, .NET Core, Blazor, Azure, and Kubernetes and sign up for our free monthly dev newsletter.

EN Newsletter Anmeldung (#7)
Related Articles
.NET 8 brings Native AOT to ASP.NET Core, but many frameworks and libraries rely on unbound reflection internally and thus cannot support this scenario yet. This is true for ORMs, too: EF Core and Dapper will only bring full support for Native AOT in later releases. In this post, we will implement a database access layer with Sessions using the Humble Object pattern to get a similar developer experience. We will use Npgsql as a plain ADO.NET provider targeting PostgreSQL.
Originally introduced in .NET 7, Native AOT can be used with ASP.NET Core in the upcoming .NET 8 release. In this post, we look at the benefits and drawbacks from a general perspective and perform measurements to quantify the improvements on different platforms.
.NET 8 introduces a new Garbage Collector feature called DATAS for Server GC mode - let's make some benchmarks and check how it fits into the big picture.