I have a package (it's name is Tree),
Tree contains 10 internal services (interfaces) and they are resolved by Autofac.
Tree contains 2 external services:
TreeCommand
TreeQuery
I want to use multiple Trees in my application.
StudentTree
TeacherTree
CompanyTree etc.
Each implementation can use different component registrations (for sub services)
How can I implement this kind of requirement by Autofac?
My solution looks a little weird without multiple containers:
Sample registration: (It is simplified version of original)
foreach (var domainInfo in domainInfoList)
{
var domainInfoParameter = new NamedParameter("domainInfo", domainInfo);
var domain = domainInfo.Domain;
builder.RegisterType<TreeFactory>()
.Named<TreeFactory>(domain)
.WithParameter(domainInfoParameter)
.SingleInstance();
builder.RegisterType<SqlTreeRepository>()
.Named<ITreeRepository>(domain)
.WithParameter(domainInfoParameter);
builder.RegisterType<TreeCommandService>()
.Named<TreeCommandService>(domain)
.WithParameter(domainInfoParameter);
builder.RegisterType<TreeQueryService>()
.Named<TreeQueryService>(domain)
.WithParameter(domainInfoParameter);
}
Sample resolving:
container.ResolveNamed<TreeFactory>("Teacher");
container.ResolveNamed<TreeCommand>("Student");
container.ResolveNamed<TreeQuery>("Company");
What is the best practise in Autofac for this kind of requirement?
Related
im using
integration toolkit for each external system ... externalService
definition, Servers registation datas, ENV refers the J2C creds to use, datamapping, bussines-errors handling etc.
(Bussines-Layer TK ex. TK_SAP)
for the common functionality such as logging, tokenizing,
pseudomizing, common http-error handling i want to use another one
toolkit (Generic implementation for Transport-layer ex. TK_COM).
So its looks like this dependicies chain:
ProcessApp -> TK_SAP -> TK_COM
There is the serviceFlow with inputs externalServiceName, operationName and a serviceFlow ask for oAuth-token and call to target system using externalServiceName, operationName.
The problem is - when i try to invoke the BPMRESTRequest from TK_COM, i get NullPointerException because "externalServiceName" cant be resolved.
var request = new BPMRESTRequest();
request.externalServiceName = "language-translator-v2";
request.operationName="checkout";
...
var response = tw.system.invokeREST(request);
is it possible to store service definition in another TK (upper) and refer it from Toolkit-invoker?
Or is there callbacks for BPMRESTRequest-Construct to say which ServiceDefinition must be used and avoid NPE.
Or another way to call Rest programmaticaly supporting Environments.
Im understand that switching the layers can help (serviceDefinition in lower TK-dependency), but it unlogisch is:
ProcessApp -> TK_COM -> TK_SAP
the answer is: JS-lib implementation.
Implementing common functionality as JS-Server-file in TK_COM makes that a call to it from TK_SAP will instatiate JS-execution context in the TK_SAP namespace so all defined in TK_SAP externalServices and variables will be accesible by executing of JS-code (actualy provided by lower-dependency Toolkit)
Please forgive my non-native English:
In short, What is the best way for a tenant to override default IEnumerable<T> registration?
TL;DR So I have a service ServiceToBeResove(IEnumerable<IShitty> svcs) need an IEnumerable<IShitty> dependency, but we found not all our tenants have services registered as IShitty, so in our application container we create an not implemented NoImplementShitty and register it as a TypeService of IShitty to server as a default one to make resolve process happy, we do get tenant-specific if tenant have registration and this default non-implemented if tenant forgot to register. But we soon find the ServiceToBeResove will have both tenants implemented registered IShitty and the default NoImplementShitty for its dependence of IEnumerable. What I really want for the IEnumerable<IShitty> dependency is just used tenant registered (registered 1 or more), if tenant not registered, just use the default NoImplementShitty as the IEnumerable<IShitty>. I have played with .OnlyIf(), OnlyIfRegistered(), .PreventDefault() on the app container and it really not helps since autofac will build default first and then tenant. I can certainly use the NoImplementShitty for all the tenant that missing registration of IShitty but it doesn't seem to take the advantage of multiple tenant's override-default features.
To be more specific, In our base AgreementModule, we have
builder.RegisterType<NoOpAgreementHandler>() //NoOpAgreementHandler is the IShitty
.As<IAgreementHandler>()
.InstancePerLifetimeScope();
In our tenantA, we have
public class TenantAContainerBuilder : ITenantContainerBuilder
{
public virtual object TenantId => "1";
public virtual void Build(ContainerBuilder builder)
{
builder.RegisterType<TenantAAgreementHandler>()
.As<IAgreementHandler>()
.InstancePerLifetimeScope();
}
}
We build container as below:
var appContainer = builder.Build();
var tenantIdentifier = new ManualTenantIdentificationStrategy(); //We have our own strategy here I just use the ManualTenantIdentificationStrategy for example
var multiTenantContainer = new MultitenantContainer(tenantIdentifier, appContainer);
//GetTenantContainerBuilders will basically give you all TenantBuilder like TenantAContainerBuilder above
foreach (IGrouping<object, ITenantContainerBuilder> source in GetTenantContainerBuilders().GroupBy(x => x.TenantId))
{
var configurationActionBuilder = new ConfigurationActionBuilder();
configurationActionBuilder.AddRange(source.Select(x => new Action<ContainerBuilder>(x.Build)));
multiTenantContainer.ConfigureTenant(source.Key, configurationActionBuilder.Build());
}
When try to resolving the service, if we do:
public DisbursementAgreementManager(IEnumerable<IAgreementHandler> agreementHandlers)
{
_agreementHandlers = agreementHandlers;
}
The agreementHandlers will be an IEnumerable of NoOpAgreementHandler and TenantAAgreementHandler, seems wierd to have NoOpAgreementHandler and I thought we will only get TenantAAgreementHandler. But if we change the DisbursementAgreementManager to
public DisbursementAgreementManager(IAgreementHandler agreementHandler)
{
_agreementHandler = agreementHandler;
}
We will get only the TenantAAgreementHandler which is expected.
The default behavior of Autofac is there for a reason. Asking it to do it differently would be adding application logic at the dependency-injection level, which violates the separation of concerns (DI should only inject dependencies) and leads directly to surprising behavior ("Why did DI not inject every available component?") and undercuts the maintainability of the system.
This may be a non-issue.
The logic is self-contained inside each IAgreementHandler.
If so, at the point where they are invoked by DisbursementAgreementManager, they are all called and then perform their own logic (which may include a decision whether to do anything all). E.g.:
foreach (var ah in _agreementHandlers) ah.Agree(disbursementInfo);
or maybe something like
foreach (var ah in _agreementHandlers.Where(a => a.ShouldRun(data) || overridingCondition))
{
var agreement = ah.Agree(info);
this.Process(agreement);
}
or whatever. The point is that if NoOpAgreementHandler is doing what it is supposed to (that is, nothing) then it should have no effect when it is called. No problem.
If the situation is other than described, then NoOpAgreementHandler and possibly IAgreementHandler need to be refactored.
There is another point of concern:
The reason we add the no-op is we have unit tests for registration/resolve in order to make sure all registration is properly configured.
Your testing requirements are bleeding into your primary logic. These DI configuration tests should be independent of the production DI configuration. NoOpAgreementHandler shouldn't even be in your primary project, just a member of the unit test project.
What I'm trying to achieve is that if i have 2 components nodes :
component1
clientlib
component1.js
component2
clientlib
component2.js
and i drag them into page1, then when page1 is generated, only component1.js and component2.js will be loaded when navigating to page1 .
One approach i saw is to use custom Tag Library as described here : http://www.icidigital.com/blog/best-approaches-clientlibs-aem-part-3/
I have two questions :
1) is there an existing feature in AEM to do this ?
2) if not, what is the easiest way to create such custom Tag Library ?
EDIT:
Assume that there is no ability to just include all component clientLibs, rather load only those that are added to the page.
There is no built in feature to do this. Although I've heard that the clientlib infrastructure is being looked at for a re-write so I'm optimistic that something like this will be added in the future.
We have, and I know other company have, created a "deferred script tag." Ours is a very simple tag that take a chunk of html like a clientlib include, add it to a unique list and then on an out call at the footer, spits it all out one after another.
Here's the core of a simple tag implementation that extends BodyTagSupport. Then in your footer grab the attribute and write it out.
public int doEndTag() throws JspException {
SlingHttpServletRequest request = (SlingHttpServletRequest)pageContext.getAttribute("slingRequest");
Set<String> delayed = (Set<String>)request.getAttribute(DELAYED_INCLUDE);
if(delayed == null){
delayed = new HashSet<String>();
}
if(StringUtils.isNotBlank(this.bodyContent.getString())){
delayed.add(this.bodyContent.getString().trim());
}
request.setAttribute(DELAYED_INCLUDE, delayed);
return EVAL_PAGE;
}
Theoretically the possible way of doing is to write script in your page component/abstract page component that does something like this -
Step1 : String path = currentPage.getPath()
Step2 : Query this path for components (one way is to have a master list do a contains clause on sling:resourceType)
Step 3: User resource resolver to resolve the resourceType in Step 3, this will give you resource under your apps.
Step 4: From the above resource get the sub-resource with primary type as cq:ClientLibraryFolder
Step 5: from the client libs resource in Step 4 get the categories and include the JS from them
you could actually write a model to adapt a component resource to a clientLibrary to actually clean the code.
Let me know if you need actual code, I can write that in my free time.
In my project, there are additional (non-wicket) applications, which need to know the URL representation of some domain objects (e.g. in order to write a link like http://mydomain.com/user/someUserName/ into a notification email).
Now I'd like to create a spring bean in my wicket module, exposing the URLs I need without having a running wicket context, in order to make the other application depend on the wicket module, e.g. offering a method public String getUrlForUser(User u) returning "/user/someUserName/".
I've been stalking around the web and through the wicket source for a complete workday now, and did not find a way to retrieve the URL for a given PageClass and PageParameters without a current RequestCycle.
Any ideas how I could achieve this? Actually, all the information I need is somehow stored by my WebApplication, in which I define mount points and page classes.
Update: Because the code below caused problems under certain circumstances (in our case, being executed subsequently by a quarz scheduled job), I dived a bit deeper and finally found a more light-weight solution.
Pros:
No need to construct and run an instance of the WebApplication
No need to mock a ServletContext
Works completely independent of web application container
Contra (or not, depends on how you look at it):
Need to extract the actual mounting from your WebApplication class and encapsulate it in another class, which can then be used by standalone processes. You can no longer use WebApplication's convenient mountPage() method then, but you can easily build your own convenience implementation, just have a look at the wicket sources.
(Personally, I have never been happy with all the mount configuration making up 95% of my WebApplication class, so it felt good to finally extract it somewhere else.)
I cannot post the actual code, but having a look at this piece of code will give you an idea how you should mount your pages and how to get hold of the URL afterwards:
CompoundRequestMapper rm = new CompoundRequestMapper();
// mounting the pages
rm.add(new MountedMapper("mypage",MyPage.class));
// ... mount other pages ...
// create URL from page class and parameters
Class<? extends IRequestablePage> pageClass = MyPage.class;
PageParameters pp = new PageParameters();
pp.add("param1","value1");
IRequestHandler handler = new BookmarkablePageRequestHandler(new PageProvider(MyPage.class, pp));
Url url = rm.mapHandler(handler);
Original solution below:
After deep-diving into the intestines of the wicket sources, I was able to glue together this piece of code
IRequestMapper rm = MyWebApplication.get().getRootRequestMapper();
IRequestHandler handler = new BookmarkablePageRequestHandler(new PageProvider(pageClass, parameters));
Url url = rm.mapHandler(handler);
It works without a current RequestCycle, but still needs to have MyWebApplication running.
However, from Wicket's internal test classes, I have put the following together to construct a dummy instance of MyWebApplication:
MyWebApplication dummy = new MyWebApplication();
dummy.setName("test-app");
dummy.setServletContext(new MockServletContext(dummy, ""));
ThreadContext.setApplication(dummy);
dummy.initApplication();
this post ideally continues my other post on MEF plugins, but my first post was too full of comments and this sample is more complete. Here I summarize my updated scenario, with all my findings up to this point. Hope this can be useful to other CM newbies like me.
You can download a full repro sample scenario: it's an almost do-nothing dumb skeleton for a CM + MEF plugins-based application:
VS2010 repro solution (updated)
This is a minimal stripped down solution representing my issues with CM+MEF.
There are 3 projects:
the main UI (CmRepro).
a core DLL shared among all the addins (AddinCore), with a couple of interfaces and custom attributes used for MEF metadata.
a sample addin DLL (AlphaAddin), with a view and a viewmodel implementing the interfaces.
The core contains 2 interfaces representing a viewmodel and its view, and 2 attributes to be used for decorating the viewmodels and the views. The viewmodel interface describes a class which should compose a greeting message from some person name, so it exposes a couple of properties and a method for this. The view interface just exposes a property returning its DataContext cast to the viewmodel interface.
The sample addin has an implementation for a viewmodel and a view; both are MEF-exports, decorated by the corresponding attribute. In the real-world solution several properties of these attributes are used for filtering; here I just have a dummy Language property which should allow for other plugins for different languages.
The main UI has a MEF bootstrapper which adds code for retrieving MEF exports from an Addins folder. I modified this code to include exports from MEF directories and get a better understanding of some MEF exceptions, but still I cannot figure out how to properly "register" them with CM.
The main viewmodel has 2 methods: one (A) uses a MEF catalog to retrieve a viewmodel and its view, bind them and show them into a window. Another (B) uses the same catalog to get a viewmodel, and then a CM window manager to locate, create, bind and show the corresponding view according to CM naming conventions. These methods represent two alternative ways I should deal with in my real-world code, i.e. instantiating some crucial objects "by myself" just using MEF but then let them work for CM, or letting CM (with a MEF-bootstrapper) do most of the work starting from a viewmodel.
Anyway, it seems that in both cases I am missing something as for registration with CM. Issues:
(A) how do I wire up VM+V for CM so that the conventions for databinding etc are applied? At this time I can build my MEF parts together, but CM ignores them as it was not used to instantiate none of them.
I answer to myself here:
ViewModelBinder.Bind(viewmodel, (UserControl)view, null);
(B) how do I register the exports from MEF in CM so that the CM window manager can find the view? Currently it does not manage to locate the view from the viewmodel.
Addition (21 jun)
I try to explain better for whom cannot access the repro solution. I use a "standard" MEF bootstrapper, changing the Configure override like:
_container = new CompositionContainer(
new AggregateCatalog(AssemblySource.Instance.Select(
x => new AssemblyCatalog(x)).OfType()
.Union(GetAddinDirectoryCatalogs())));
this creates a MEF composition container which aggregates the catalog from AssemblySource, with CM types like event aggregator or window manager, with a catalog from several addin directories, which contain exports for both V and VM's.
In my sample main viewmodel I create a new VM from a plugin, found in the host application directory among others, and I'd like a CM window manager to locate, instantiate and show its view in a dialog, e.g.:
viewmodel = GetMyViewModelFromAddin();
windowmanager.ShowDialog(viewmodel);
CM anyway cannot locate the view. AFAIK, naming conventions are honored: both V and VM are in the same addin assembly, marked as MEF exports, named like SomethingViewModel / SomethingView. Anyway, as Leaf pointed out in his clarification, AssemblySource.Instance is a static IObservableCollection collection of assemblies and I have not added my addins to it. But this is right the point: I would not like to add all of them in advance, because this means loading ALL the addins without yet knowing which (if any) will be ever used. A robust plugin system is the reason for using MEF, after all. I'm new to CM and not sure if it is possible (and where) to find an extension point for CM in this scenario. The window manager does not call my bootstrapper implementation at all, clearly because there is nothing to be instantiated by IoC, as no match was found in assembly source Instance. So, it seems I'm stuck here, the only solution being loading in advance all the assemblies in the Instance, but this seems defeating the whole purpose of using MEF.
The plugin-based application I'm developing loads tons of V+VM CM "pairs" representing user interface widgets, which in turn often use a window manager to popup other V+VM's pairs as dialogs. I can bypass instantiation by CM and use MEF to retrieve V+VM for each widget, but still I'm facing the same view location issue for each widget requiring a window manager. The other alternative (workaround) I can see is avoding the use of window manager and implement my own mechanism for the purpose of showing dialogs from widgets, but this makes feel me a little wrong about CM...:). Usually when I find myself writing much more code than expected I am inclined to think I'm not using the tool in the right way. Any idea?
Caliburn.Micro goes through 3 stages to find a view from a viewmodel instance.
Does text transforms to transform view type name to viewmodel type name. There are a set of standard conventions for these transformations, e.g. SomeNamespace.ViewModels.CustomerViewModel might map to SomeNamespace.Views.CustomerView.
With the view type name(s), Caliburn.Micro then uses AssemblySource.Instance (a static IObservableCollection<Assembly> collection of assemblies) to find the first matching Type.
Caliburn.Micro attempts to create an instance of that Type via one of the IoC.GetInstance() methods (which delegate to your bootstrapper and hence MEF).
I'm guessing (your file share site is blocked here) that the problem with resolving views from viewmodels is due to the second step and the AssemblySource.Instance collection not containing your dynamic assembly.
One solution might be to add each dynamically loaded addin's assembly to AssemblySource.Instance when they are loaded, or if you know all assemblies at startup then you can override the Bootstrapper.SelectAssemblies method to return list of assemblies you expect to contain views and viewmodels.
Update to show how you could pull assemblies from MEF loaded parts
If you are using DirectoryCatalog to load your parts from other assemblies then you could find the assemblies used like this:
var directoryCatalog = new DirectoryCatalog(#"./");
AssemblySource.Instance.AddRange(
directoryCatalog.Parts
.Select(part => ReflectionModelServices.GetPartType(part).Value.Assembly)
.Where(assembly => !AssemblySource.Instance.Contains(assembly)));
If your addins folder changes during the runtime of the application then you would have to DirectoryCatalog.Refresh() the catalog and run the code to add any new assemblies to AssemblySource.Instance
I found a workaround. It's not that pretty, but it allows CM and its window manager to work. I summarize here my findings, hope this will help others or let someone point me to a better solution.
Given that (a) I do not want to load ALL the assemblies including all their dependencies in my plugins folder, to avoid polluting my app domain with unused stuff; and that (b) the only available CM extension point for this seems SelectAssemblies, my goal is add my assemblies there, but only the plugin assemblies requiring to be registered with CM.
So I started looking for a way of loading all the DLL's into a temporary app domain, scan them for some aspect marking them as plugins, and then load into the current app domain only the plugin ones, passing them to SelectAssemblies. This is far from being the optimal solution, as I cannot use a general purpose scanning way like MEF, and I feel I'm duplicating the effort, but at least it's a working solution.
First, to provide at least a way for loading only the plugins, I decorate my plugin assemblies requiring CM registration with a custom attribute marking them as plugins, and further enumerate their types looking for the ones which should be later used by MEF.
The attribute is as simple as:
[AttributeUsage(AttributeTargets.Assembly)]
public class AssemblyRegisteredWithCMAttribute : Attribute {}
I then found this very good piece of code for scanning assemblies into another temporary AppDomain:
http://sachabarber.net/?p=560
I had to modify it a bit, because it was failing when loading assemblies with dependencies, and it was scanning the assemblies types while in my case I just need to check for the attribute.
Here is my code:
using System;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using System.Security.Policy;
using System.Reflection;
using System.Diagnostics.CodeAnalysis;
namespace CmRepro
{
///
/// Separate AppDomain assembly loader.
///
/// Modified from http://sachabarber.net/?p=560 .
public class SeparateAppDomainAssemblyLoader
{
///
/// Loads an assembly into a new AppDomain returning the names of the files
/// containing assemblies marked with the assembly attribute name matching
/// the specified name. The new AppDomain is then Unloaded.
///
/// list of files to load
/// assemblies directory
/// matching attribute name
/// list of found namespaces
/// null files, assembly directory or matching attribute
public List LoadAssemblies(string[] aFiles, string sAssemblyDirectory, string sMatchingAttribute)
{
if (aFiles == null) throw new ArgumentNullException("aFiles");
if (sAssemblyDirectory == null) throw new ArgumentNullException("sAssemblyDirectory");
if (sMatchingAttribute == null) throw new ArgumentNullException("sMatchingAttribute");
List<String> namespaces = new List<String>();
AppDomain childDomain = BuildChildDomain(AppDomain.CurrentDomain);
try
{
Type loaderType = typeof(AssemblyLoader);
if (loaderType.Assembly != null)
{
AssemblyLoader loader = (AssemblyLoader)childDomain.
CreateInstanceFrom(
loaderType.Assembly.Location,
loaderType.FullName).Unwrap();
namespaces = loader.LoadAssemblies(aFiles, sAssemblyDirectory, sMatchingAttribute);
} //eif
return namespaces;
}
finally
{
AppDomain.Unload(childDomain);
}
}
/// <summary>
/// Creates a new AppDomain based on the parent AppDomains
/// Evidence and AppDomainSetup.
/// </summary>
/// <param name="parentDomain">The parent AppDomain</param>
/// <returns>A newly created AppDomain</returns>
private AppDomain BuildChildDomain(AppDomain parentDomain)
{
Evidence evidence = new Evidence(parentDomain.Evidence);
AppDomainSetup setup = parentDomain.SetupInformation;
return AppDomain.CreateDomain("DiscoveryRegion", evidence, setup);
}
/// <summary>
/// Remotable AssemblyLoader, this class inherits from <c>MarshalByRefObject</c>
/// to allow the CLR to marshall this object by reference across AppDomain boundaries.
/// </summary>
private class AssemblyLoader : MarshalByRefObject
{
private string _sRootAsmDir;
/// <summary>
/// ReflectionOnlyLoad of single Assembly based on the assemblyPath parameter.
/// </summary>
/// <param name="aFiles">files names</param>
/// <param name="sAssemblyDirectory">assemblies directory</param>
/// <param name="sMatchingAttribute">matching attribute name</param>
[SuppressMessage("Microsoft.Performance", "CA1822:MarkMembersAsStatic")]
internal List<string> LoadAssemblies(string[] aFiles, string sAssemblyDirectory, string sMatchingAttribute)
{
AppDomain.CurrentDomain.ReflectionOnlyAssemblyResolve += OnReflectionOnlyAssemblyResolve;
_sRootAsmDir = sAssemblyDirectory;
List<string> aAssemblies = new List<String>();
try
{
sMatchingAttribute = "." + sMatchingAttribute;
foreach (string sFile in aFiles)
Assembly.ReflectionOnlyLoadFrom(sFile);
aAssemblies.AddRange(from asm in AppDomain.CurrentDomain.ReflectionOnlyGetAssemblies()
let attrs = CustomAttributeData.GetCustomAttributes(asm)
where attrs.Any(a => a.ToString().Contains(sMatchingAttribute))
select asm.FullName);
return aAssemblies;
}
catch (FileNotFoundException)
{
/* Continue loading assemblies even if an assembly
* can not be loaded in the new AppDomain. */
return aAssemblies;
}
}
private Assembly OnReflectionOnlyAssemblyResolve(object sender, ResolveEventArgs e)
{
// http://blogs.msdn.com/b/junfeng/archive/2004/08/24/219691.aspx
System.Diagnostics.Debug.WriteLine(e.Name);
AssemblyName name = new AssemblyName(e.Name);
string sAsmToCheck = Path.GetDirectoryName(_sRootAsmDir) + "\\" + name.Name + ".dll";
return File.Exists(sAsmToCheck)
? Assembly.ReflectionOnlyLoadFrom(sAsmToCheck)
: Assembly.ReflectionOnlyLoad(e.Name);
}
}
}
}
Now in my bootstrapper I override the SelectAssemblies method as follows:
...
protected override IEnumerable SelectAssemblies()
{
string sAddinPath = GetAbsolutePath(ADDIN_PATH);
FileCheckList list = new FileCheckList(sAddinPath);
// check only DLL files which were added or changed since last check
SeparateAppDomainAssemblyLoader loader = new SeparateAppDomainAssemblyLoader();
List<string> aAssembliesToRegister = loader.LoadAssemblies(list.GetFiles(null),
sAddinPath, "AssemblyRegisteredWithCM");
string[] aFilesToRegister = (from s in aAssembliesToRegister
select Path.Combine(sAddinPath, s.Substring(0, s.IndexOf(',')) + ".dll")).ToArray();
// update checklist
foreach (string sFile in aFilesToRegister) list.SetCheck(sFile, true);
list.UncheckAllNull();
list.Save();
// register required files
return (new[]
{
Assembly.GetExecutingAssembly(),
}).Union((from s in list.GetFiles(true)
select Assembly.LoadFrom(s))).ToArray();
}
...
As you can see, I am calling the loader not for all the DLL's in my addins path, but only for the ones which a cached files checklist tells me have been added or modified in that folder since the last full scan. This should speed up things a bit and does not require the checklist file to exist: if not found it will be recreated at startup by scanning all the files, if found only added or changed files will be scanned again (I use a CRC to detect changes). So I get the addins folder, create a files checklist for that folder, get from it a list of new or changed files, and pass it to the assemblies loader. This returns only the names of the DLL files which should be registered (i.e. the ones containing assemblies marked with my attribute); I then update the checklist for the next startup, and register only the required files. This way I can let my addin VM's use window managers and correctly locate the view for each required viewmodel. Somewhat ugly, but working. Thanks again to Leaf, who explained me the working of CM.