Ideas for better performance and safety

[mlaily]

Hello :)

I imagine you did some research before starting this project, and I guess you probably wouldn't have started the project if there were already a satisfying implementation available. I know it was a long time ago, but do you remember if you had the chance to review this: https://www.codeproject.com/Articles/29922/Weak-Events-in-C ?

The author seems trustworthy ("lead developer on SharpDevelop"), and the performance of his implementation is pretty much excellent. To achieve such performance he replaces reflection with System.Reflection.Emit.DynamicMethod and IL code.

• Is there any particular reason you didn't use this technique in your library, when the running platform supports it? (beside it being black magic)

• His implementation ensures you can't use anonymous delegates by checking for the presence of CompilerGeneratedAttribute on declaring type of the delegate's method. It could be a welcomed improvement in your library, because it's way too easy to erroneously subscribe to a lambda...

---

I put up the code-project's code in a new repo, and replaced the original ad-hoc benchmark with BenchmarkDotNet. I also added a benchmark for your library. Check it out! https://github.com/mlaily/FastSmartWeakEvent

Summary:

|                                 Method |      Mean |     Error |   StdDev | Ratio | RatioSD |  Gen 0 | Gen 1 | Gen 2 | Allocated |
|--------------------------------------- |----------:|----------:|---------:|------:|--------:|-------:|------:|------:|----------:|
|                'Normal (strong) event' |  12.12 ns |  0.218 ns | 0.414 ns |  1.00 |    0.00 |      - |     - |     - |         - |
|                     'Smart weak event' | 594.21 ns |  9.563 ns | 7.467 ns | 47.41 |    0.98 | 0.0420 |     - |     - |     177 B |
| 'Fast smart weak event (2009 version)' |  41.49 ns |  0.839 ns | 1.030 ns |  3.36 |    0.12 | 0.0172 |     - |     - |      72 B |
| 'Fast smart weak event (2013 version)' |  18.16 ns |  0.157 ns | 0.131 ns |  1.45 |    0.04 |      - |     - |     - |         - |
|                    'Thomas weak event' | 567.82 ns | 10.375 ns | 9.705 ns | 45.64 |    1.33 | 0.1545 |     - |     - |     650 B |

EDIT: I almost forgot: he apparently rewrote the thing 4 years after his code-project article, but it seems to only live in a gist here: https://gist.github.com/dgrunwald/6445360#file-fastsmartweakevent-cs (I added it to my repo and benchmark)

Have a great day!

[thomaslevesque]

Salut Melvyn,

Yes, I know about this article. To be honest, I don't remember exactly why I didn't use the same approach (or even just use his code directly...). Maybe there was something I couldn't do with it, but I don't remember what. Anyway, this project started as a POC to test an idea (using open-instance delegates), and ended up being used by many people, so I didn't rethink my approach. I agree that it's probably worth considering using reflection emit.

• His implementation ensures you can't use anonymous delegates by checking for the presence of CompilerGeneratedAttribute on declaring type of the delegate's method. It could be a welcomed improvement in your library, because it's way too easy to erroneously subscribe to a lambda...

I'm not sure about that. Using a lambda is fine, as long as you keep a reference to it. But yes, it's easy to get wrong...

[mlaily]

Thanks for your answers!

Feel free to close this issue if you don't intend to implement these suggestions, or if you want to create more targeted issues for them :)

[thomaslevesque]

Closing this issue, since it's pretty old and not seeing any activity. I have some plans to redesign this library using compile-time code generation, which would probably improve performance as well. But it depends on this being implemented in the compiler, and it's probably too late for the next release...