17th Feb, 2018
- Upgraded to framework 4.6
- Updated references to version 2.6.1 of Microsoft compilers platform.
This provides programmer friendly view of your application via the .NET Compiler Platform (code named Roslyn) from Microsoft.
- The best thing about .NET Compiler Platform, Roslyn is that it exposes the syntactic and semantic trees of the compiler
- The worst thing about .NET Compiler Platform, Roslyn is that it exposes the syntactic and semantic trees of the compiler
It's a freakin' awesome compiler, but you don't think like that. No one thinks like that.
You might want to use this tool, RoslynDom to
- Perform complex refactoring that isn't provided by the services of Roslyn or easily created via an Analyzer. For example I added the property assigning constructors to RoslynDom using RoslynDom.
- Do many of the things you would do with a syntax walker/rewriter when performance isn't an issue
- Perhaps, get better performance in iterative updates that require multiple rewriter passes in Roslyn
- Do things with the compiler information without having to think like a compiler
It's probably easiest to understand RoslynDom by understanding it's components
A set of mutable (yes, mutable) classes that represent your code as close to how you think about code as I've been able to accomplish so far. That's sort of an "average you" so you may not love everything, but I'm kind of a geek on the human/programming language interface and I worked hard and watched closely how we talk about code, which is a good clue about how we think about code.
These classes are built on a set of characteristic based interfaces (see below).
I planned RoslynDom to be VB/C# language agnostic and to be friendly to the idea of interacting with other languages in parallel trees.
There is pretty much a one-to-one relationship between the RoslynDom classes and a C# factory.
The C# factories take The .NET Compiler Platform, Roslyn, syntax and semantic trees and sort through them to load RoslynDom. It's messy work sometimes.
The C# factories also take RoslynDom classes and recreate The .NET Compiler Platform, Roslyn syntax trees.
Not yet created.
For certain scenarios, this will allow large scale VB/C# conversions. I say for certain scenarios because VB and C# have some significant differences. The thing about this approach is that we can reason about the code, look for problem areas, refactor, etc. in the common tree.
One of the frustrating things about The .NET Compiler Platform, Roslyn is that it has a very shallow syntax tree, which means very little polymorphism. There are very good reasons to do that in the compiler, and I'm also not a fan of deep hierarchies for anything, particularly where the commonality is as fluid as in the syntax tree.
So, I created an interface to correspond with each entity in RoslynDom, and a series of interfaces for characteristics. Sometimes these characteristic interfaces are quite small with only one property.
There are two important benefits of these targeted characteristic interfaces. First, you can go look up an entities interface and find out what it can do. I want some visualization of this, but I'm not there yet.
Second, as you're working with the tree, you can simply attempt a cast (an as cast) and see if you have that property to work with. No need to understand any other aspect of the entity.
There's a fair amount of sometimes messy code that supports all this.
There's over 600 passing unit tests, and about 28 tests that are inconclusive because I know I need them and havne't gotten to it.
There is a lot of testing still to do. I'm only at 87% coverage and I think a core library should be at near 100%. You might notice that I have a tendency to increase what's at near 100%, while postponing testing in other areas. I think it makes it easier to know where to still be careful.
It's been an enormous effort.
I didn't do it just to do it.
I did it because I have two generation projects - CodeFirstMetadata and ExpansionFirstTemplates that entirely rely on it. The prototypes of those projects used The .NET Compiler Platform, Roslyn directly. The complexity that arose from combining the concerns of compiler detail and quirks with trying to get the generation job done made for exceedingly complex code.
And in a workshop last spring I said "I'm sure the community will build a wrapper for Roslyn" and a few decades of experience echoed in my head "you are community."
My only real feedback to date has been those projects. I turn to them,make some headway and turn back to make RoslynDom better. The vast majority of my work has been to make RoslynDom good (it's not done, and could use refactoring, etc, but I'm very happy with the core). As a result, those two projects, which tackle much more difficult problems are relatively small and as easy to understand as complex mapping problems can be.
So, here you go. I am community! You are community, and what I most need right now from community is feedback. This is still a young project and I'm still available to answer questions like "is RoslynDom a good idea to do xxx."
Load a project, explore or make changes, recreate the syntax tree.
Use LINQ, Ancestors, Descendants, and OfType often in your code.
That's it.
You can see samples in the RoslynDom examples on NuGet.
Yes, there still be dragons. This is a large work in progress.
RoslynDom entities retain a reference to the syntax and symbol tree. This is still a very active area and don't trust them. The eventual intention is to allow you to access symbolic services directly, but this requires knowing that the symbol and model are up to date, and as soon as you make a change, they aren't.
Another place the "youngness" of the symbolic work shows is that a few symbolic lookups (like an invocation method's reference to it's method) is created at load time and not updated.
Eventually, your changes will selectively invalidate the .NET Compiler Platform, Roslyn semantics, allowing some very neat tricks.
Another place you'll find dragons is that RoslynDom is incomplete. You can find Issues relating to most of the C# 5 missing pieces. Expressions are generally handled as strings right now (except a few). I want to understand the value as I dive into this space - it's an area where you sometimes do think in code so the text is not so bad.
Groupings are challenging because they affect how you think. There is an issue on this, namespaces, the lack of async support and handling combined declarations like int i,j,k are all part of this larger problem.
I have not yet incorporated C# 6 syntax - it's not yet final.
If you find ways your code breaks RoslynDom, please tell me. If you send a sample, it will make the breaking tests and will therefore be a permanent part of testing and never break again. Obviosuly, the smaller the better here. Obviously not secret code.
Special thanks to Benson Joeris for his frequent brilliant insight, and constant patience when GitHub confused me.
Thanks to Llewellyn Falco for his support, help, and pushing me off the high dive.
Thanks to the audiences at my workshops. So far, you have always sent me home with work to do, and I thank you.
Install-Package RoslynDOM -Pre
You can find more info on [my blog] (http://msmvps.com/blogs/kathleen/default.aspx)
IRoot root = RDom.CSharp.LoadFromFile("something.cs");
INamespace nameSpace = root.Namespaces.First();
string name = nameSpace.Name;
// Consider using the Immediate or Watch windows to explore code
// You can also make changes directly to the tree
SyntaxTree = RDom.CSharp.GetSyntaxNode(root); var ws = MSBuildWorkspace.Create();
var solution = ws.OpenSolutionAsync(slnFile).Result;
Project project = solution.Projects.Where(x => x.Name == "RoslynDomExampleTests").FirstOrDefault();
IRootGroup rootGroup = RDom.CSharp.LoadGroup(project);
IRoot = rootGroup.Roots.First();
// Explore as aboveAsk on Twitter: @kathleendollard