Tag Archives: Lamar

Lamar v3 is Released: Faster, smaller, quicker cold starts, internal type friendly

Lamar has had some churn lately with me almost giving up on it, changing my mind, and look at that, publishing a very significantly improved version up to Nuget just now.

I just published Lamar v3.0 with some bug fixes and big internal improvements. This release is mostly significant because it eliminates Lamar’s previous reliance on Roslyn for runtime code generation in favor of a more common model that builds Expression trees and compiles dynamic lambdas — inevitably with the excellent FastExpressionCompiler for maximum performance and faster cold starts.

All told, Lamar v3.0:

  • Has much faster “cold start” times compared to ❤
  • Is faster across the board, but especially when internal types are involved. The very early user feedback is “holy moses that’s much faster”
  • I think that we’re through with compatibility with internal types being registered into the container because the dynamic Lambdas don’t have to care about the type visibility
  • Will use way less memory without the Roslyn dependency tree in memory
  • Has far fewer dependencies now that some users complained about

There are no API changes to Lamar itself, but there’s now a different set of Nuget libraries:

  1. LamarCompiler — only has Lamar’s old helper for executing the Roslyn runtime compilation, but this is no longer referenced by Lamar itself
  2. LamarCodeGeneration — the Frame / Variable model that Lamar uses to generate code and now Expression trees in memory. This is used by Lamar itself, could be used as a stand alone library, and has no direct coupling to LamarCompiler
  3. Lamar itself

A touch of history…

I’ve been absolutely exhausted for years with maintaining StructureMap and dealing with user questions and oddball problems. I also knew it had some pretty ugly performance issues and it had been a massive pain in the ass force fitting StructureMap’s long standing functionality into ASP.Net Core’s new compliance requirements (that appear to have been made up by the ASP.Net team out of whole cloth with ZERO input from the major OSS IoC tool authors). I also knew from some experimentation that I wasn’t going to be able to fix StructureMap very easily, so…

Lamar was an attempt by me to throw together a new IoC tool that would mostly mimic StructureMap‘s public API (everybody has opinions, but there are actually quite a few people who like StructureMap’s usability and it’s had ~6 million downloads) and a large subset of its functionality. At the same time, I did change its lifecycle semantics and behavior to be ASP.Net Core-compliant from the very get go to head off more problems with that.

Lamar’s original model was based around using dynamic code generation and compilation with Roslyn. Woohoo, it was easy to use and I was able to get a large subset of StructureMap’s old functionality working over Christmas break a couple years ago. Unfortunately, that model doesn’t play well at all with internal types and my limited workarounds were pretty clumsy. It turns out that lots and lots of people and common frameworks like to put internal types into IoC containers (looking at you ASP.Net Core). So here we are. Lamar switched over to using Expressions compiled into Funcs like basically every other IoC container I’m aware of, but hopefully it starts fulfilling my real goal for Lamar:

Get folks complaining about StructureMap off of that and make my life easier overall by reducing user problems.


Jasper v0.9.9 is Released!

Jasper is an open source project I’ve been furiously working on for the past couple years as what I have to admit is mostly a chance to resurrect the best parts of the earlier FubuMVC framework while solving its technical and usability shortcomings. Along the way, Jasper has evolved a bit away from its FubuMVC roots to be more consistent and compatible with ASP.Net Core. At this point, Jasper is a fancy command execution pipeline that lives inside of the ASP.Net Core ecosystem. Jasper can be used as any mix of a lightweight messaging framework, an in-memory service bus, or handling HTTP requests as an alternative ASP.Net Core framework.

I was just able to push a new v0.9.9 release of Jasper that I want to effectively be the last alpha release. At this point, I think the public API surface is pretty well set and only a handful of features left before making the big ol’ 1.0 release. I think the most important thing for Jasper is to try to get folks to take it for a spin, or glance through tutorials, and generally try to get some feedback and visibility about the project.

Here’s some links to get you started:

There’ll hopefully be plenty of blog posts on Jasper in the next couple weeks, starting with how Jasper’s usability contrasts with MVC Core in the HTTP space or NServiceBus/MediatR/etc. for messaging and command execution.

The Road to 1.0

I’ve got to put Jasper down for awhile to focus on some Lamar and help out a lot more with Marten for probably a couple months before I do much more on Jasper, but I’d still love to get 1.0 out by at least the end of the summer.

I’m thinking out loud in this section, so everything is subject to change. Before I flip the switch to the big, giant 1.0, I think these things might need to happen:

  • There are some optimizations I want to make to the database backed message persistence I couldn’t quite get to for v0.9.9.
  • I’m very tempted just to wait until the netcoreapp3.0 wave of updates comes out. It’s a near guarantee that ASP.Net Core v3.0 will break some of Jasper’s internals, and it’d be very helpful to slim Jasper’s package dependency tree down if Jasper could depend on the new generic host model instead of IWebHostBuilder when 3.0 unifies that model somewhat.
  • Jasper heavily depends on the Task Parallel Library from Microsoft, but that seems to be somewhat deprecated now. I might look to rewire Jasper’s internals to use the newer System.Threading.Channels instead. I haven’t done any research into this one yet.
  • If the HTTP support is going to live on, it’ll need some kind of Swagger/Swashbuckle integration for the Jasper HTTP API support
  • I’d like to spend some time on supporting the Techempower Benchmarks for Jasper and some performance optimization. My goal here is to make Jasper the fastest HTTP application framework for .Net Core and be just barely slower than the raw ASP.Net Core benchmarks (not sure how feasible that is, but let me dream on).
  • Jasper is going to have to adjust to whatever becomes of Lamar. I don’t think this is going to change Jasper at development time, but might introduce a new production build step to optimize Jasper application’s “cold start” times for better hosting in Docker kind of worlds.

Lamar stays and how that enfolds

I wrote a blog post at the end of last week about Lamar when I just happened to be feeling discouraged about a couple things (it happens sometimes, and I swear that having to support IoC tools exposes me to more difficult people than every other project I work on combined). I got some rest this weekend, a bit of positive reinforcement from other folks, and actually thought through how to fix the issues. Long story short, I’m not giving up on anything and here’s what I think the very doable game plan is for Lamar (and the closely related Jasper project):


  • Short term: Get a small bug fix release out soon that has some options to “force” all the compilation upfront in one dynamic assembly. That’s gonna hurt the cold start time, but should help the memory usage. We’ll also look to see if there’s any places where Lamar is holding on unnecessarily to the Roslyn compilation objects to ensure that they can be garbage collected
  • Medium term: Introduce an alternative compilation model based on Expressions compiled to Lambdas with FastExpressionCompiler. This model will kick in automatically whenever there’s one or more internal types in the “build plan”, and could be opted into globally through a container level switch. I didn’t want to do this originally because the model just isn’t very fun to work with.  After thinking it through quite a bit over the weekend, I think it won’t be bad at all to retrofit this alternative to Lamar’s existing Frame and Variable model. This will knock out the performance issues with internal types and address all the memory issues.
  • Long term: Probably split up LamarCompiler a little bit to remove the actual code compilation to significantly slim down Lamar’s dependency tree and move Lamar to a purely Expression based model. Introducing the Expression model will inevitably make the exception stacktrace messages coming out of Lamar explode, so there might have to be an effort to rewrite them to make them more user friendly (I had to do this in StructureMap 3 several years ago).



I’m very close to pulling the trigger on a Jasper v1.0, with the understanding that it’s inevitable that there will be a Jasper v2.0 later in the year to incorporate .Net Core 3.0. I don’t think that Jasper will have any issues with memory usage related to Lamar because it uses Lamar very differently than MVC in any flavor. The changes to Lamar will impact Jasper though, so:

Short term: Jasper v1.0 with Lamar as it is.

Medium term: Jasper gets a model where you can happily use the runtime codegen and compilation during development time while things are churning, but for production usage you have the ability to just drop the code that would be generated to disk, have that compiled into your system in the first place, and let Jasper use those types. Ultra fast production time cold start times, no worries at all about Roslyn doing bad things from memory. I’ve already done successful proof of concept development on this one.

Long term: profit.

As an aside, I got quizzed quite a bit about why Jasper has to be specific to Lamar as its IoC container and can’t just support whatever tool folks want to use. The reason is that Jasper uses Lamar’s very specific code generation in its pipeline to avoid using an IoC container at runtime whatsoever and also to avoid forcing users to have to conform to all kinds of Jasper specific adapter interfaces. I could maybe force Jasper to still pull this off with the built in DI container or another IoC container with Jasper-centric adapters to expose all of its metadata in a way such that the codegen understands it, but just ick.

If you took Lamar’s runtime codegen away, I think Jasper inevitably looks like a near clone of NServiceBus or Brighter both in its usability and runtime pipeline and why does the world need that?



Lamar asks: Do I stay or do I go?

I happened to look into the Lamar issue list on GitHub late this afternoon at the end of a very long work day. What I saw was nothing to feel good about. A huge memory usage problem (probably related to Roslyn but I don’t know that for sure), a user who’s been very difficult in the past telling me he switched to another tool because it was easier to use and faster, bugs that unfortunately point to a huge flaw in Lamar’s basic premise, and folks really wanting a feature I’ve put off because it’s going to be a lot of work.

Lamar started its life as a subsystem of Jasper, the project that’s been my main side-project for the past couple years and my biggest ambition for several years before that. Jasper’s special sauce is the way that it uses Lamar to generate and compile code at runtime to glue together the runtime pipeline in the most efficient way possible. At the same time, it was killing me to support StructureMap and I wanted out of it, but didn’t want to leave a huge mess of folks with an abandoned project, so I turned Lamar into an IoC tool that was purposely meant to be a smooth transition for StructureMap users to an improved and more efficient IoC tool.

Right now, it’s becoming clear that the basic model of Lamar to use the runtime code generation isn’t working out. First because of what looks like some severe memory issues related to the runtime compilation, and second because it turns out that it’s very, very common for folks to use internal .Net types in IoC containers which pretty well defeats the entire code generation model. Lamar has workarounds for this to fall back to dynamic expressions for internal types, but it’s not optimized for that and Lamar’s performance degrades badly when folks use internal types.

As I see it, my choices with Lamar are to:

  • Retool it to use dynamic Expression compilation all the way down in the IoC engine usage like basically every other IoC tool including the older StructureMap library. That takes care of the performance issues most likely and knocks out the bugs due to internal types, but just sounds miserable.
  • Deal with the memory problems by trying to engage with the actual Roslyn team to figure out what’s going on and work around it. Which doesn’t solve the internal type issue.
  • Use pre-generated code build by Lamar during testing time baked into the real application for production deployments. There’s a bit more than a proof of concept already baked in, but this still doesn’t fix the internal problem

On the Jasper side I could:

  • Try to make it work with the built in IoC container but still try to do its “no IoC at runtime because it’s all codegen’d” model with a cutdown version of Lamar

Or, just call the whole thing off because while I would argue that Jasper shows a lot of promise technically and I still believe in its potential, it has zero adoption and I most likely won’t get to use it myself in any projects in the next year. If I give up on Jasper, I’m likely to do the same to Lamar. In that case, I’ll write off Lamar as a failed project concept, apologize profusely to all the folks that started to use it to replace StructureMap, and walk away.

Nobody needs to feel sorry for me here, so no need to try to cheer me up about any of this, but, I wouldn’t mind some feedback on whether it’s worth keeping Lamar around and what you think about the proposed improvements.

My OSS Plans for 2019

I wrote a similar post last year on My OSS Plans for 2018, and it wasn’t too far off what actually happened. I did switch jobs last May, and until recently, that dramatically slowed down my rate of OSS contributions and my ambition level for OSS work. I still enjoy building things, I’m in what’s supposed to be a mostly non-coding role now (but I absolutely still do), and working on OSS projects is just a good way to try to keep my technical skills up. So here we go:

Marten — I’m admittedly not super active with Marten these days as it’s a full fledged community driven project now. There has been a pile up of Linq related issues lately and the issue backlog is getting a bit big, so it really wouldn’t hurt for me to be more active. I happen to love Stephen King’s Dark Tower novels (don’t bother with the movie), but he used to complain that they were hard for him to get into the right head space to write more about Roland and his Ka-tet. That’s basically how I feel about the Linq provider in Marten, but it’s over due for some serious love.

Lamar (was BlueMilk) — Users are reporting some occasional memory usage problems that I think point to issues in Roslyn itself, but in the meantime I’ve got some workarounds in mind to at least alleviate the issue (maybe, hopefully, knock on wood). The one and only big feature idea for this year I have in mind is to finally do the “auto-factory” feature that I hope will knock out a series of user requests for more dynamic behavior.

Alba — I took some time at CodeMash this year and did a pretty substantial 3.0 release that repositioned Alba as a productivity helper on top of TestServer to hopefully eliminate a whole lot of compatibility issues with ASP.Net Core. I don’t know that I have any other plans for Alba this year, but it’s the one tool I work on that I actually get to use at work, so there might be some usability improvements over time.

Storyteller — I’d really love to sink into a pretty big rewrite of the user interface and some incremental — but perfectly backward compatible — improvements to the engine. After a couple years of mostly doing integration work, I suddenly have some UI centric projects ahead of me and I could definitely use some refresher on building web UIs. More on this in a later post.

Jasper — I think I’m finally on track for a 1.0 release in the next couple months, but I’m a little unsure if I want to wait for ASP.Net Core 3.0 or not. After that, it’s mostly going to be trying to build some community around Jasper. Most of my side project time and effort the past three years has been toward Jasper, I have conceptual notes on its architecture that go back at least 5 years, and counting its predecessor project FubuMVC, this has been a 10 year effort for me. Honestly, I think I’m going to finish the push to 1.0 just to feel some sense of completion.

Oakton — I feel like it’s basically done

StructureMap — I answer questions here and there, but otherwise it’s finished/abandoned

FubuMVC — Dead, but little bits and pieces of it live on in Oakton, Alba, and Jasper


Metaprogramming in C# with LamarCompiler

I gave a talk last week at NDC London called “Dynamic Runtime Code with Roslyn” that featured an example of using LamarCompiler and Lamar to create a minimal equivalent of the Refit library. If you’re not familiar with Refit, it allows you to express the intended service proxy to a web service with a .Net interface or abstract class, apply a few attributes to explain Url patterns or other HTTP request items, and then allow Refit to dynamically generate a class that implements your interface that deals with all the repetitive gunk of building Urls, serializing or deserializing JSON, and calling HttpClient to actually make the requests.

Cool, right? I’d argue that Refit is an example of metaprogramming inside of C# where a tool is able to generate and eventually execute code gleaned from data (an interface and a handful of attributes in this case) within the application. I tend to think of metaprogramming as something specific to the Ruby and other dynamic languages arena, but now that we have the ability to compile C# code on the fly with Roslyn, I think it’s going to become something much more commonly done in .Net systems.

Today in the Lamar codebase, there’s a little project called LamarRest that contains all the code I’m going to show here if you’re curious to see more about how it all fits together. I should say that I intended this code strictly as a demonstration of what LamarCompiler can do and do not have any plans right now to make this a production capable library.

The Problem and Conceptual Solution

Okay, let’s get concrete and write a very small web service that manages users. The whole API surface in this case is implemented by the UserController class shown below (the full source code for this web service is here):

    public class UserController : Controller
        private readonly UserRepository _repository;
        private readonly ILogger _logger;

        public UserController(UserRepository repository, ILogger logger)
            _repository = repository;
            _logger = logger;

        public IActionResult Create([FromBody] User user)
            _logger.LogInformation($"I created a new user with the name '{user.Name}'");
            return Ok();
        public User GetUser(string name)
            return _repository.Users.FirstOrDefault(x => x.Name == name);

        public User[] GetUsers()
            return _repository.Users.ToArray();

Now, back in the codebase for an application that needs to communicate with the user service, we ideally want to just write an interface to act as a proxy to the User web service that looks like this:

    public interface IUserService
        Task GetUser(string name);

        Task Create(User user);

In our application, we just want to interact with the IUserService and not really have to know where it comes from. To that end, I’m going to go a step farther than Refit does and say that our solution should automatically resolve the generated type for IUserService from our application’s IoC container on demand.

As you’ll notice, I did have to add a couple attributes ([Get] and [Post]) that just specify the Url pattern for the two proxy methods. In this case as well though, the presence of the route marker attributes tells the code we’re about to write that these methods should be generated to call HttpClient.

Implementing Our Refit Clone

To generate the concrete HTTP proxy classes, I’m going to leverage LamarCompiler’s Frame model. If you write out implementation of IUserService above, you’ll notice discrete set of logical steps, each of which becomes a LamarCompile Frame in this model:

  1. Resolving an HttpClient from the HttpClientFactory for our application.  That became BuildClientFrame.
  2. Fill the real Url by substituting any route arguments with method parameters from our interface method signatures. That’s implemented (crudely and incompletely) in LamarRest by the FillUrlFrame.
  3. If necessary, serialize the input boy argument to Json using Newtonsoft.Json’s default settings (Refit being a real library allows you to configure the Json serialization, LamarRest being a demo worked up the day before a conference talk, does not). For that purpose, I created SerializeJsonFrame.
  4. To actually send an Http request, we need to create an HttpRequestMessage object with the right Url, Http method, and the serialized Json if there is any. This one is a little easier, as I just subclass Lamar’s built in support for calling a class constructor with any mix of arguments and setters in the BuildRequestFrame class.
  5. To actually invoke HttpClient.SendAsync() , there’s just a built in MethodCall Frame inside of LamarCompiler.
  6. If there is a response body expected, as in the IUserService.GetUser() method, we need a Frame that can deserialize the contents of the HTTP response body into the desired model. I implemented that in DeserializeObjectFrame.

Okay, so that’s a lot of steps, but once these Frame classes were built, it was easy to assemble them to match an interface method. In LamarRest, there’s a fair sized class called GeneratedServiceType that handles all the code generation and compilation for these proxy classes. Most applicable to this discussion is the method that builds out the concrete proxy methods by assembling the necessary list of Frame objects as shown below:

public static void BuildOut(
    // The contract type
    Type interfaceType, 
    // The MethodInfo from Reflection that describes
    // the Url structure through attributes, the input type
    // if any, and the .Net type of the response (if any)
    MethodInfo definition, 
    // This models the method being generated by LamarCompiler
    GeneratedMethod generated)
    var path = definition.GetCustomAttribute();

    // Get the right HttpClient from IHttpClientFactory
    generated.Frames.Add(new BuildClientFrame(interfaceType));

    // Build out the Url from the method arguments and route
    // pattern
    var urlFrame = new FillUrlFrame(definition);

    // See if there is an input type that should be serialized
    // to the HTTP request body
    var inputType = DetermineRequestType(definition);
    if (inputType == null)
        // Just build the HttpRequestMessage
        generated.Frames.Add(new BuildRequestFrame(path.Method, urlFrame, null));
        // Add a step to serialize the input model to Json
        generated.Frames.Add(new SerializeJsonFrame(inputType));
        // Just build the HttpRequestMessage
        generated.Frames.Add(new BuildRequestFrame(path.Method, urlFrame, new SerializeJsonFrame(inputType)));

    // Actually call the HttpClient to send the request
    generated.Frames.Call(x => x.SendAsync(null));

    // Is there a response type that should be serialized out of the HTTP
    // response body?
    var returnType = DetermineResponseType(definition);
    if (returnType != null)
        // Deserialize the response JSON into a new variable
        var deserialize = new DeserializeObjectFrame(returnType);
        // Return that deserialized object from the method

I should note that LamarCompiler does some work of connecting variables, method arguments, and class fields by type and sometimes by name — which is why you see no explicit mention of how IHttpClientFactory comes into play. LamarCompiler is smart enough to see that a Frame object needs an IHttpClientFactory object, know that that service is a singleton-coped object in the application’s IoC container, and therefore generate a field for that service.

For the IUserService interface, LamarRest will generate this code for the concrete class, but I warn you now, it’s generated code and hence will be as ugly as sin:

    public class UserService : LamarRest.Testing.IUserService
        private readonly System.Net.Http.IHttpClientFactory _httpClientFactory;

        public UserService(System.Net.Http.IHttpClientFactory httpClientFactory)
            _httpClientFactory = httpClientFactory;

        public async System.Threading.Tasks.Task GetUser(string name)

            // From BuildClientFrame
            var httpClient = _httpClientFactory.CreateClient("IUserService");

            // From FillUrlFrame
            var url = $"/user/{name}";
            var httpRequestMessage = new System.Net.Http.HttpRequestMessage(new System.Net.Http.HttpMethod("GET"), url);
            using (var httpResponseMessage = await httpClient.SendAsync(httpRequestMessage))

                // From DeserializeObjectFrame
                var user = Newtonsoft.Json.JsonConvert.DeserializeObject(await httpResponseMessage.Content.ReadAsStringAsync());
                return user;


        public Task Create(UserApp.Controllers.User user)

            // From BuildClientFrame
            var httpClient = _httpClientFactory.CreateClient("IUserService");

            // From FillUrlFrame
            var url = $"/user/create";

            // From SerializeJsonFrame
            var json = Newtonsoft.Json.JsonConvert.SerializeObject(user);
            var httpRequestMessage = new System.Net.Http.HttpRequestMessage(new System.Net.Http.HttpMethod("POST"), url){Content = new System.Net.Http.StringContent(json, System.Text.Encoding.UTF8, "application/json")};
            return httpClient.SendAsync(httpRequestMessage);

Hooking it up to the Application

Lastly, we need to hook up our new LamarRest library to our application’s IoC container and application configuration. Assuming that you’re using Lamar as your IoC container, the configuration might be something like this:

            var container = new Container(x =>
                    // You'd pull this from actual configuration in real life
                    c => c.BaseAddress = new Uri("http://localhost:5000"));
                // Adding a Lamar policy here that will auto-resolve LamarRest
                // proxies on demand

In the code above, I’m heavily leveraging the standard ASP.Net Core AddHttpClient() registration to deal with base Urls and default HTTP request attributes like authentication credentials. I’d probably switch it to being strong typed later, but for now the convention is to name the HttpClient based on the interface name.

The second registration is Lamar specific to add a policy to allow Lamar to auto-resolve a concrete class for any interface that is otherwise unknown to the container and has methods marked with any kind of LamarRest route attribute.

Just for completeness sake, that’s shown below:

    public class LamarRestPolicy : Lamar.IFamilyPolicy
        public ServiceFamily Build(Type type, ServiceGraph serviceGraph)
            if (type.GetMethods().Any(x => x.HasAttribute()))
                var rules = new GenerationRules("LamarRest");
                var generatedAssembly = new GeneratedAssembly(rules);
                var generatedType = new GeneratedServiceType(generatedAssembly, type);

                var container = (IContainer)serviceGraph.RootScope;

                return new ServiceFamily(
                    new IDecoratorPolicy[0], 
                    new ConstructorInstance(type, 

            return null;

The final result of all this setup is to be able to do this:

            // Build an implementation of this service as needed
            // I'm using service location here, but you'd be able
            // to inject IUserService as a constructor argument as well
            var userService = container.GetInstance<IUserService>();


Lamar 2.0 is Out! StructureMap’s Successor Grows up a bit

I was just able to publish Lamar 2.0 to Nuget and it looks like it’s indexed, so here we go!

Along with a whole lot of user reported bug fixes, the two big changes are:

  • We split the dynamic code compilation and code generation (the Frames model) into its own separate library called LamarCompiler so that interested folks could easily reuse that functionality without having the IoC support bundled along with it (even though you never needed to use the Lamar IoC code to take advantage of the code compilation but .Net devs are finicky about dependencies some times)
  • After a couple user requests for it, Lamar 2.0 supports optional and mandatory setter injection based on StructureMap’s setter injection support.\

Many thanks to all the folks that helped move this along by reporting GitHub issues and providing pretty damn useful reproduction steps. Special thanks to Mark Warpool for all his help in making Lamar go. Lamar, like Marten before it, has been a very positive experience for me in my dealings with the greater OSS community.

For more information, here’s:


What next for Lamar?

I don’t know when, but I’ll eventually get around to the “AutoFactory” feature, because I think that’s the single biggest blocker for many folks moving from StructureMap to Lamar. If LamarCompiler takes off at all, there’s worlds of opportunities to add new C# code constructs to the code generation model.