As you probably know based on news from BUILD Windows conference last week, Roslyn, the new .NET Compiler Platform with code analysis APIs for C# and VB.NET, has been pushed to Preview phase and open sourced. In case the event slipped your attention, here’s a nice tour of Roslyn on the C# FAQ blog at MSDN.
Immediately we at JetBrains were faced with multiple questions on the perspectives of ReSharper using Roslyn for its code analysis and how the two tools might compete. The flow of questions wouldn’t end, to the point of introducing a template to answer them:
Seriously though, it was clear that we needed to elaborate on the ReSharper vs Roslyn issue. Hence this post.
We sat down with Sergey Shkredov (@serjic), ReSharper Project Lead and .NET Tools Department Lead at JetBrains, and Alex Shvedov (@controlflow), a Senior Developer on the ReSharper team who’s responsible for ReSharper’s Generate functionality, code annotations and support for XML-based languages. The following Q&A is a summary of the conversation that we had with them.
What’s JetBrains stance towards Roslyn? Do we consider the technology and its Open Source status important and valuable?
Roslyn is definitely important and a good step forward for Microsoft in that it should help Visual Studio users take advantage of more C# and VB.NET code editing and analysis features in Visual Studio out of the box.
It should also help Visual Studio extension developers write code-centric extensions against a consistent API while having the opportunity to know how it works inside, thanks to the Open Source status of the project. This is not to mention hackers who are willing to spend their time forking the compiler and tuning it to make, say, C# the ideal language they’ve always envisioned.
We also believe that Roslyn is no less important for Microsoft itself. Faced with the burden of maintaining a plethora of Visual Studio integrated tools including code editing tools, IntelliTrace and code designers, the folks at Microsoft are interested in making these tools as flexible and easy to update as possible. Roslyn should enable updating .NET languages and experimenting with them faster than before. Apart from that, the old compiler wouldn’t let launch compilation steps in parallel, and Roslyn is expected to enable that, bringing more scalability to the table.
What’s the point of making Roslyn Open Source?
As to the act of letting Roslyn go Open Source, we don’t believe that Microsoft is expecting anyone from outside of the company to develop the compiler for them. Programming languages are entities too monolithic and complex to justify accepting external changes of any significance. Therefore, we expect Microsoft to keep the function of designing .NET languages totally to itself without depending on the community.
The true value of Roslyn going Open Source lies in enabling extension developers to look into Roslyn code that is relevant to their purposes: how it’s written and whether it’s efficient. They might debug or profile it to see if it’s the culprit of unexpected behavior in their extensions or if it introduces performance issues. This is possibly the workflow whereby meaningful pull requests might start coming in to the Roslyn repository.
As to possible endeavors to fork and modify the compiler to address application- or domain-specific tasks, this scenario appears to be a shot in the foot. Even if the default compiler in Visual Studio can be replaced with a fork, instrumental support for the fork ends as soon as you go beyond Visual Studio. In theory we can imagine a custom
INotifyPropertyChanged implementation based on a Roslyn fork that can even gain certain popularity. However, we can barely imagine supporting it in ReSharper as our intention is to focus on supporting the official version of Roslyn.
Will ReSharper take advantage of Roslyn?
The short answer to this tremendously popular question is, no, ReSharper will not use Roslyn. There are at least two major reasons behind this.
The first reason is the effort it would take, in terms of rewriting, testing and stabilizing. We’ve been developing and evolving ReSharper for 10 years, and we have a very successful platform for implementing our inspections and refactorings. In many ways, Roslyn is very similar to the model we already have for ReSharper: we build abstract syntax trees of the code and create a semantic model for type resolution which we use to implement the many inspections and refactorings. Replacing that much code would take an enormous amount of time, and risk destabilizing currently working code. We’d rather concentrate on the functionality we want to add or optimize, rather than spend the next release cycle reimplementing what we’ve already got working.
The second reason is architectural. Many things that ReSharper does cannot be supported with Roslyn, as they’re too dependent on concepts in our own code model. Examples of these features include Solution-Wide Error Analysis, code inspections requiring fast lookup of inheritors, and code inspections that require having the “big picture” such as finding unused public classes. In cases where Roslyn does provide suitable core APIs, they don’t provide the benefit of having years of optimization behind them: say, finding all derived types of a given type in Roslyn implies enumerating through all classes and checking whether each of them is derived. On the ReSharper side, this functionality belongs to the core and is highly optimized.
The code model underlying ReSharper features is conceptually different from Roslyn’s code model. This is highlighted by drastically different approaches to processing and updating syntax trees. In contrast to ReSharper, Roslyn syntax trees are immutable, meaning that a new tree is built for every change.
Another core difference is that Roslyn covers exactly two languages, C# and VB.NET, whereas ReSharper architecture is multilingual, supporting cross-language references and non-trivial language mixtures such as Razor. Moreover, ReSharper provides an internal feature framework that streamlines consistent feature coverage for each new supported language. This is something that Roslyn doesn’t have by definition.
Will it be practical to use both ReSharper and Roslyn-based functionality in Visual Studio?
This is a tricky problem as it’s still uncertain whether we would be able to disable Roslyn-based features (such as refactorings or error highlighting) when integrating into new releases of Visual Studio. If we’re unable to do that, performance would take a hit. Apart from ReSharper’s own immanent memory and performance impact, Roslyn’s immutable code model would increase memory traffic, which would in turn lead to more frequent garbage collection, negatively impacting performance.
We’re hopeful that this problem would be solved in favor of letting us disable Roslyn features that ReSharper overrides, because otherwise ReSharper would have to work in a highly resource-restricted environment. Irrelevant of whether this happens though, we’ll keep doing what we can do, minimizing ReSharper’s own performance impact.
As Roslyn is now Open Source, which parts of its code are going to be of particular interest to ReSharper developers?
We’ll be sure to peek into Roslyn code and tests from time to time, to see how C# and VB.NET language features are implemented. We don’t rule out that actual code supporting them is going to emerge before formal specifications are finalized. In fact, we’ve already started.
That’s more or less the picture of living in the Roslyn world as we see it today. As time goes by, we’ll see if things turn out the way we expected them to.
Meanwhile, if you have questions that were not addressed in this post, please ask in comments and we’ll try to come up with meaningful answers.