How does Blazor works ? Part 1 : building class from .razor

I started this blog one year ago, I am proud to say that I published 22 blog posts, it’s one every 2.5 weeks, which is a bit under my first goal of one every 2 weeks, but it’s ok. I now have a bit less than 3000 visitors per month, I guess it’s not a lot, but it’s enough to keep me going.

I am starting a post serie about Blazor internal. I will try to find out how Blazor works, how a project is build, how is it loading on the browser, how is the UI updated … While I will focus on Blazor client-side, some of this can be applied to Blazor server-side as well (most of the build process and rendering logic). In this serie I’ll learn as I write the post, trying to figure out how the code works. I will try to add references and code sample from Microsoft repositories. If you think that my understanding is wrong, please add a comment, and I’ll fix the post.

In this first post I will try to figure out what happens after we enter “dotnet build” on a Blazor project and how C# class are generated from razor files.

dotnet cli

The dotnet cli source code is located here. When we enter “dotnet build”, the dotnet program looks for a “build” in its internal command catalog (this enables developpers to add other command to the catalog).

The BuildCommand inherit from MSBuildForwardingApp which uses MSBuildForwardingAppWithoutLogging. This class will then start a command line for executing MsBuild with the good arguments. As I found out with procexp the command looks like this

"dotnet.exe" exec "C:\Program Files\dotnet\sdk\3.0.100-preview5-011568\MSBuild.dll" -maxcpucount -verbosity:m -restore -consoleloggerparameters:Summary -target:Build "-distributedlogger:Microsoft.DotNet.Tools.MSBuild.MSBuildLogger,C:\Program Files\dotnet\sdk\3.0.100-preview5-011568\dotnet.dll*Microsoft.DotNet.Tools.MSBuild.MSBuildForwardingLogger,C:\Program Files\dotnet\sdk\3.0.100-preview5-011568\dotnet.dll"

We can see it doesn’t use MsBUild.exe directly but rather a subcommand of dotnet cli called “exec”. I can’t find where this command is defined, I guess it’s not managed code (as Matt Warren says here) and it’s the component responsible for starting a new managed executable (.exe are dead there is only .dll now in the dotnet world), I think it’ nearly the same as “dotnet MsBuild.dll” .

MSBuild

The MsBuild.dll code is located here from my understanding, the entry point is here. I won’t go into MSBuild internal details, as the source code is really large. Basically MSBuild is driven by many XML files that describe the different step of the build process. For defining this build process we see that “dotnet exec” sends one thing to MSBuild : “target:Build”. A target in MSBuild is like a function or an entry point : you execute a target, MSBuild reads the tasksand dependants target connected to this target and runs it. Targets are defined in a “.targets” file located on the SDK folder, the target definition that’ll be used are defined by the “Sdk” attribute on the csproj file. In a Blazor app it’s “Microsoft.Net.Sdk.Web”, so the target definition will be in “C:\Program Files\dotnet\sdk\3.0.100-preview5-011568\Sdks\Microsoft.NET.Sdk.Web\Sdk\Sdk.targets” this file just include targets file from other sdk. For understanding the different tasks launched, it’s easier to run the build with a more detailed verbosity :

dotnet build -v detailed > build.log

The build.log file will contain the target (function) hierarchy involved in the building of this project,we’ll have to search the target name on the Sdk folder for understanding what it does. The targets about “Component” or “Blazor” are :

  • ResolveRazorComponentInputs : this task looks for all the “.razor” files and put them into a variable
  • _CheckForIncorrectComponentsConfiguration : checks there is at least one blazor file and RazorCompileOnBuild is true. I think this means that in Blazor template are build at build time, in classic ASPNET Core you can build Razor template at runtime.
  • AssignRazorComponentTargetPaths : for each razor file, it creates a path for the generated class (the “.g.cs” file)
  • _HashRazorComponentInputs : for each .razor it computes a hash and saves it in a .cache file, this must be for debugging purpose I guess (for checking the matching between a build output and the source code).
  • _ResolveRazorComponentOutputs : I don’t get it, it’s some kind of a check.
  • RazorGenerateComponentDeclaration : this is where the generation of the .g.cs class happens
  • other target that we will explore in future blog posts

RazorGenerateComponentDeclaration

The main purpose of this target is to call “RazorGenerate”. This task is defined here. This task doesn’t do much itself but generate a “dotnet exec” command line for a dll called “rzc.dll” (it must stands for “razor compiler”). After decompiling this dll with dotPeek, I found out that the source code of this dll is here. After following the Program entry point, I found the Application class which is a bit like the dotnet cli we saw before, it defines a bunch of commands that’ll be executed. The command that RazorGenerate calls is … wait for it … the “generate” command (line 51 of RazorGenerate.cs).

This GenerateCommand creates a RazorProjectEngine browse all the files in a parallel manner 4 by 4 (line 304) and calls the RazorProjectEngine.Process method on it.

RazorProjectEngine.Process

The RazorProjectEngine is the class responsible for initializing the code generation process which is made of a list of IRazorFeature which are the necessary steps for generating the C# code out of template file. A feature can be linked to a directive (keyword), so it can get content from the template file, like the “@functions” directive defined here that “simply” add all the content between “@function{}” to the class definition. Some of those steps are for “.cshtml” parsing, some are for “.razor”, when a directive is registered it must pass the kind of file it applies to (the “function” directive for instance applies to both), the file kind choosing is done on the FileKinds class. A few of these steps :

  • ComponentInjectDirectivePass : add an attribute “Microsoft.AspNetCore.Components.InjectAttribute” for every match with the “inject” directive.
  • ComponentDocumentClassifierPass : will generate most of the .g.cs file (the class name, namespace, and BuildRenderTree method ) via subclass like ComponentCodeTarget or its base class DocumentClassifierPassBase.

You can check them all out here : (https://github.com/aspnet/AspNetCore-Tooling/tree/master/src/Razor/src/Microsoft.AspNetCore.Razor.Language/Components).

Conclusion

Writing this post I learned many thing, among those are 2 thing important :

  • Because the language processing logic is located in the SDK, I understand why I need to update my SDK before every Blazor update.
  • Blazor/Component are deeply integrated into the Razor code ans the dotnet SDK, so we can say this project is really important for them. If it wasn’t important they would have tried to put it as much as possible on its corner so it can be removed safetly later.

This kind of blog post is new to me as it’s mostly reading and trying to understand code written by people way smarter than me, if you think I missed something, some stuff needs more clarification, or said something wrong, please send a comment, I will improve my post.

In this first part we saw how “dotnet build” creates a c# class out of all your .razor file. On the next part we will see the other steps necessary (blazor boot json file, linking etc …) for building a Blazor project.