.NET 9 Is Out(github.com) |
.NET 9 Is Out(github.com) |
Large performance improvements.
[0] - https://devblogs.microsoft.com/dotnet/performance-improvemen...
And then spice up the concurrency and measure memory consumption as well, a 24h test.
I know that TechEmpower gets a lot of flak for being "unrealistic", but it's easy enough to filter out the pure middleware ones. But even those I feel are "fair game" because it shows that such optimizations are possible on the runtime if needed. The story behind one of the top entrants, Just(js) talks a bit about this[3]
[0] https://benchmarksgame-team.pages.debian.net/benchmarksgame/...
[1] https://github.com/TechEmpower/FrameworkBenchmarks/blob/mast...
[2] https://tfb-status.techempower.com/
[2] https://just.billywhizz.io/blog/on-javascript-performance-01...
A pretty good result for just changing a dropdown in the project settings tab.
https://learn.microsoft.com/en-us/dotnet/csharp/language-ref...
The most obvious way to store backing fields for arbitrary objects is to have a global concurrent dictionary that stores weak keys, and a compiler-generated class to store the backing fields as the value. But then you have a performance problem of needing to do a lookup from a weak-concurrent-dictionary every time you access any extension backing field.
The higher-performance alternative would be a second shadow variable in code that uses the extension backing fields. Do your dictionary lookup, then cache the backing-fields-object so you don't need to redo lookups for it. But that would be function-local.
If the extension property doesn't need a new backing field, it's trivial, you're just making a new getter and setter function pair.
Edit: include ;-) at the end. Client filtered it out...
twinBasic https://twinbasic.com
The 2 apps in production are MAUI Blazor Hybrid apps. The learning curve was very small since I already was familiar with C# and Razor syntax after having built many C# server-side rendered web apps. Development is quite rapid since it's essentially using web technologies. For my use cases, the users do not care about the app looking native as these are B2B apps. Both apps are deployed in the 2 major mobile app stores and 1 app is also deployed in the Microsoft Store.
The 3rd app was an internal-only Android app built with the UI written entirely in C# instead of XAML. I chose to write the UI in C# instead of XAML primarily just for fun and to see how that would work out. I ran into a lot more issues using native UI when it came to me wanting to customize things, such as removing an annoying bottom text underline/border that is added by default to Entry (textbox) controls.
I'll probably never build another native UI app if I have the choice and will stick to MAUI Blazor Hybrid apps because it is so much faster to create the app using web technologies (HTML/CSS).
Behind php and nginx.
On the more "focused" application and micro benchmarks the difference between 7, 8 and 9 is very significant. The annual compiler improvements match those that take about 3 years for OpenJDK, and more for Golang. This has reached the point where you could reasonably expect e.g. LINQ to breathe down the Rust's iterators neck in un-vectorizable code (which is 98% of it), I didn't expect this would actually happen.
I have been told that each Linux kernel update tends to hit Techempower scores particularly hard, and practically everything in 75th percentile tends to be very sensitive to specific hardware configuration and DB driver implementation quality.
GC improvements while somewhat hurting Techempower too should make everyone deploying on small containers very happy as it pushes GC design in a Go-like direction with very minor hit to throughput (which is still massively better) while yielding significant sustained RSS reduction.
Personally, I feel like .NET has an identity crisis of sorts - it is being marketed as a regular high-level language for regular applications and web development, where it has plenty of competition.
At the same time it never seems to be marketed for low-level, systems and systems-adjacent scenarios, where .NET's competition is far behind. Any new DB or a message broker or a web server that is written in Java or Go would have benefited massively from being built with .NET instead, as proven by projects like Ryujinx and Garnet/FASTER. Nor it is being marketed at FP public - F# has access to the whole .NET ecosystem and easily slots into existing .NET projects better than Kotlin, Clojure or Scala do into Java ones.
> Personally, I feel like .NET has an identity crisis of sorts - it is being marketed as a regular high-level language for regular applications and web development, where it has plenty of competition.
>
> At the same time it never seems to be marketed for low-level, systems and systems-adjacent scenarios
C# and TypeScript are already fairly similar. In my mind, T# would be like a trimmed down C# with a bit more influence from F#.
.NET 8
.NET 9 +
4.605 binarytrees,csharpcore,7
5.336 binarytrees,csharpcore,7 +
5.229 binarytrees,csharpcore,1
6.112 binarytrees,csharpcore,1 +
4.725 binarytrees,csharpcore,2
5.365 binarytrees,csharpcore,2 +
12.565 binarytrees,csharpcore,8
13.504 binarytrees,csharpcore,8 +
5.076 binarytrees,csharpcore,3
6.452 binarytrees,csharpcore,3 +
2.219 fannkuchredux,csharpcore,5
2.267 fannkuchredux,csharpcore,5 +
11.337 fannkuchredux,csharpcore,3
11.014 fannkuchredux,csharpcore,3 +
9.076 fannkuchredux,csharpcore,9
8.399 fannkuchredux,csharpcore,9 +
8.665 fannkuchredux,csharpcore,6
9.930 fannkuchredux,csharpcore,6 +
40.509 fannkuchredux,csharpcore,8
43.795 fannkuchredux,csharpcore,8 +
29.589 fannkuchredux,csharpcore,2
30.376 fannkuchredux,csharpcore,2 +
11.103 fasta,csharpcore,8
11.152,0 fasta,csharpcore,8 +
1.206 fasta,csharpcore,9
1.207 fasta,csharpcore,9 +
1.502 fasta,csharpcore,5
1.482 fasta,csharpcore,5 +
1.838 fasta,csharpcore,1
1.696 fasta,csharpcore,1 +
3.043 knucleotide,csharpcore,1
3.184 knucleotide,csharpcore,1 +
40.687 knucleotide,csharpcore,7
42.742 knucleotide,csharpcore,7 +
70.598 knucleotide,csharpcore,8
75.208 knucleotide,csharpcore,8 +
5.057 knucleotide,csharpcore,6
5.632 knucleotide,csharpcore,6 +
11.672 knucleotide,csharpcore,4
12.063 knucleotide,csharpcore,4 +
6.822 mandelbrot,csharpcore,3
6.793 mandelbrot,csharpcore,3 +
26.665 mandelbrot,csharpcore,8
26.408 mandelbrot,csharpcore,8 +
3.195 mandelbrot,csharpcore,7
3.206 mandelbrot,csharpcore,7 +
45.641 mandelbrot,csharpcore,2
46.193 mandelbrot,csharpcore,2 +
3.233 mandelbrot,csharpcore,1
3.929 mandelbrot,csharpcore,1 +
4.039 mandelbrot,csharpcore,5
4.055 mandelbrot,csharpcore,5 +
7.796 nbody,csharpcore,1
7.656 nbody,csharpcore,1 +
4.793 nbody,csharpcore,7
4.803 nbody,csharpcore,7 +
6.645 nbody,csharpcore,3
6.647 nbody,csharpcore,3 +
6.440 nbody,csharpcore,0
7.656 nbody,csharpcore,0 +
7.254 nbody,csharpcore,8
7.214 nbody,csharpcore,8 +
3.164 nbody,csharpcore,9
3.174 nbody,csharpcore,9 +
4.822 nbody,csharpcore,6
4.828 nbody,csharpcore,6 +
3.747 nbody,csharpcore,4
3.749 nbody,csharpcore,4 +
6.831 nbody,csharpcore,2
6.893 nbody,csharpcore,2 +
1.359 pidigits,csharpcore,4
1.291 pidigits,csharpcore,4 +
0.928 pidigits,csharpcore,5
0.839 pidigits,csharpcore,5 +
6.590 pidigits,csharpcore,7
7.148 pidigits,csharpcore,7 +
0.960 pidigits,csharpcore,6
0.818 pidigits,csharpcore,6 +
1.371 regexredux,csharpcore,2
1.361 regexredux,csharpcore,2 +
2.108 regexredux,csharpcore,5
1.741 regexredux,csharpcore,5 +
4.598 regexredux,csharpcore,1
4.598 regexredux,csharpcore,1 +
7.309 revcomp,csharpcore,3
8.472 revcomp,csharpcore,3 +
0.585 revcomp,csharpcore,9
0.586 revcomp,csharpcore,9 +
2.639 revcomp,csharpcore,5
2.534 revcomp,csharpcore,5 +
1.642 revcomp,csharpcore,6
1.551 revcomp,csharpcore,6 +
1.559 revcomp,csharpcore,7
1.543 revcomp,csharpcore,7 +
2.972 revcomp,csharpcore,1
3.140 revcomp,csharpcore,1 +
0.933 spectralnorm,csharpcore,5
0.938 spectralnorm,csharpcore,5 +
5.387 spectralnorm,csharpcore,8
5.405 spectralnorm,csharpcore,8 +
1.519 spectralnorm,csharpcore,2
1.521 spectralnorm,csharpcore,2 +
1.544 spectralnorm,csharpcore,3
1.539 spectralnorm,csharpcore,3 +IronPython and IronRuby had a lot of potential.