This thread shows there is still a large amount of FUD around .NET and it’s community.
Folks here still think it’s Windows-only and comparable to Java. Yet latest TechEmpower benchmarks shows .net running on Linux and being faster than Go, Python, Node, and Rust.
The performance really is astonishing in many areas now.
The part that keeps me hooked is the fact that I can stand up something capable of producing those numbers in 30 lines of code using 1st party dependencies only. I can then have a full prototype to demo by late afternoon, again having antagonized over exactly zero 3rd parties.
For me, the performance numbers aren’t just about speed while in production. It’s also about speed to production. Having the confidence that it’s almost certainly going to be fast enough by default keeps me from worrying about optimizing random bullshit throughout.
I still haven’t seen anything that comes remotely close to the combination of speed and stability offered here. I do some pretty nasty things with the runtime and it just copes. GC seems like pure magic now too. I continue to avoid things like LOH allocations with streams, but I don’t worry about it like I used to in the 4.x days.
> If you like the productivity, just wait till you try Rails.
I like Rails (vintage, Rails 5). Unfortunately DotNet and EF combined now have virtually the same productivity but with massively better performance. Which is a shame as Ruby is a great language.
PHP is still a thing since time immemorial… And will remain there for a long time to come. (Seems like they made quite some improvements in the latest iterations as well.)
.NET core can run on any platform it wants, but as long as it's from Microsoft, the company as we know it, I won't even write "Hello, World!" with it.
A language can't be the fastest at everything in today's mature language ecosystem. OTOH, Debian's "Programming Language Games" benchmarks shows it's on par with Java (which is not slow in any means), and not as fast as you claim [0].
In the page I shared, some C# benchmarks are impressively fast, because they are written with explicit hand crafted SSE3/AVX vectors, which a run of the mill programmer won't want to touch (for most of the time, anyway).
Comparing .NET Core, which is a hybrid JIT language with purely interpreted ones like NodeJS and Python also makes no sense at all, considering Python Compiler does not do any optimizations whatsoever, and is still confined to a single core per process unless you pull some tricks.
Choosing horses for courses is fine, and we all shall do it, but claiming a language as winner over a single benchmark suite, including the one I referenced is wrong.
The difference between the shootout microbenchmarks and the TechEmpower benchmarks is that the former test computation-intensive tasks and the latter test an end-to-end web backend example. It tests how good the compiler is, how good the DB drivers are, how good the HTTP stack is.
What's so good about MS ranking in the TechEmpower benchmarks is that it's the fastest full-featured "enterprise" framework. Frameworks like drogon or just-js are impressive feats of engineering, but the only reason they exist is so that their authors can mention that on their resume. If you run them in production, you are on your own.
There's equally fast Vert.x, which you can buy support from Red Hat for, but most Java shops use Spring Boot (vmware or Red Hat support available), which is easier, but hopelessly slow in comparison.
> The difference between the shootout microbenchmarks and the TechEmpower benchmarks is that the former test computation-intensive tasks and the latter test an end-to-end web backend example.
That proves my point even further, because I don't develop anything remotely web-related, hence that performance scenario is completely moot for me.
> What's so good about MS ranking in the TechEmpower benchmarks is that it's the fastest full-featured "enterprise" framework.
Again, the same because the software I develop is not "Enterprise" either.
What I develop is scientific software, and needs to be extremely performant. In my case C++ is the best language, but it's not the best for every case. This is why I also learn Go and use Python for other tasks.
In my case, Debian's benchmarks are directly related to my use cases, however even that's not a definitive measurement for my case. It's just a bunch of data to keep in mind.
> That proves my point even further, because I don't develop anything remotely web-related, hence that performance scenario is completely moot for me.
I hope you realize non-web development is a niche nowadays. So it's perfectly reasonable to prioritize .NET benchmarks that target ASP.NET Core web apps/APIs.
I hope you realize that web-development in my discipline is a niche nowadays. So, it's perfectly reasonable to prioritize bare metal and micro performance benchmarks that target naked performance, OS and GPU apps/APIs.
Our vantage points are different, and these different vantage points need not to see the complete ecosystem as a whole. It's too big to generalize from a single vantage point, and just because our view reveal us a little of something doesn't mean it is, in fact, little.
There's a whole invisible world out there, from DBs themselves to OS kernels to embedded platforms, and everything in between.
> I hope you realize non-web development is a niche nowadays.
That's a terrible attitude. If you want a certain ecosystem to get a better rep, you have to be ready to talk with different communities, and accept their feedback.
Python is what it is largely because it can cater to a large number of these very different communities, which has ensured its long-term success. Compare with Ruby, which became effectively a web-only language and has since struggled to go anywhere.
> Otherwise we would be coding web applications in assembly
Some people write web-apps in C. Just because you don't like something, it doesn't mean the world agrees.
>Ruby, wouldn't even be a blip in history radar if it wasn't for Ruby on Rails
Hard disagree. Ruby was getting traction on its own, as "the purest OOP language you can use in the real world", around the same time Python was starting to get traction (early 2000s). Then RoR blew up, effectively coopting the entire ecosystem. Since then, Python has slowly gone from strength to strength in so many different fields, whereas Ruby died on its ass as soon as people moved on to other tools for web.
Curious, what is it with Microsoft that makes it so poisonous for you that you won't touch anything they make with a ten foot pole? As an example, I'd say Facebook (Meta) is a way more toxic company but I've never heard of a developer have moral arguments against using React
I'm using Linux for 20 years, and witnessing what Microsoft is doing for more than 25. Being openly hostile towards Linux and plotting ways to lock it out of hardware platforms and tech ecosystems is enough of a motivation for me.
It's not only their stance against Linux only, but everything competing with them.
I believe hardware and software should be open and platforms shall compete openly. I don't use any vendor which openly kills this interoperability and try to corner market with underhanded tactics.
It's not limited to Microsoft, though. I use Java because there's OpenJDK, Go because there's gcc-go toolchain, etc. Similarly I don't use Rust because it's LLVM only for now.
However, this doesn't mean that I want these to disappear. I want them to compete fairly, so we can improve. I'm not a web-dev so I have no need for this frameworks, but I'd make similar choices if I enter to that arena, too.
FWIW I also use Linux exclusively, develop (and host) dotnet applications on it, and have my own gripes with it (mostly with Linux still being treated as a second-tier platform which is only good for servers as far as MS is concerned — I'm not talking about the abomination that VS is — try to compare the official profiling & debugging tooling).
There's an important distinction: License. OpenJDK is GPL2.0, .NET runtime is MIT. They may be same in terms of openness, but not in freedom sense.
As a result, OpenJDK is much more sustainable in the long term, since it can't be closed down and crippled as easily.
While I am no enemy of MIT/BSD style licenses, the "freedoms" they provide to corporations are damaging to open source ecosystem in my eyes.
OpenJ9 looks like EPL licensed, and indeed interesting. It's worth a look.
I'm not writing Java with any considerable volume for some time (because, I didn't need that), hence I was just keeping that in my peripheral vision without paying much attention. However, I'm using Eclipse because it's a great IDE for my needs, and comes with OpenJ9 embedded.
Thanks for bringing this to my attention, will look deeper.
> I believe hardware and software should be open and platforms shall compete openly. I don't use any vendor which openly kills this interoperability and try to corner market with underhanded tactics. [..] Similarly I don't use Rust because it's LLVM only for now.
I'm curious, what do you have against LLVM? I'm sure it supports fewer backends than GCC, no surprise given their relative ages, but it's still got quite a few and it's open enough that it has documented support for adding new backends.
I'm not strictly against LLVM. My objections are the license, and how companies use it against GCC/GNU.
I open everything I can, and strictly with GNU/GPLv3+. I want this code can be built and improved upon, even after I abandon it voluntarily or involuntarily. Hence I choose GPL licensed (or close as much as possible) toolchains and tools as much as possible.
For example, I use Eclipse, and include .project files in the repositories, so one can import the project as is, and continue playing, or maintaining the code. I'm improving my Eclipse knowledge to use agnostic paths, so anyone cloning the repo can directly import to Eclipse, and just click build.
I also plan to make my Eclipse and toolchain settings available to make my development environment completely reproducible.
On the case of the Rust, I'm waiting gccrs, which is arriving as an unofficial language in gcc-13. This will both provide an alternative and GPL licensed implementation of the language, and I can be sure that the code I leave behind can be built with the public version of the GCC.
Moreover, I use no compiler specific extensions, because I'm not trying to lock anyone to any toolchain. I'm just trying to make sure that the repo is buildable. I may even add a LLVM (or go-gc) test path to make sure that I don't break anything between implementations.
IOW, I have strong opinions about software freedom, and I apply them to my code, but not force people to obey them (except the license, because I need to underpin it). It's a strong stance, yet there's no insistence. I just set an example, and try to make it a very good one, to show it can be done.
I think the point was not against LLVM itself, but rather against languages for which only a single implementation is available, which is a pretty strong argument IMHO.
us old farts remember Microsoft (and Gates) as true villains. Things have changed but those wounds run deep and the scars remain. Every time i see the Bill and Melinda Gates foundation in the news all i can see is Bill trying to buy his way into heaven.
I never considered Microsoft as villains in the 90s. They were the only company making computers usable, and thanks to them I didn't have to use the archaic Unix systems that were prevalent at the time. Microsoft was being unfairly punished for things that are commonplace today, like integrating browsers with operating systems.
They have strategy called Embrace, Extend, Extinguish.
They are still running this strategy, but now also buy good PR buying paying high profile developers thru their developer advocate positions.
"""
Microsoft finally is fulfilling its dream of EEE to Linux. Windows Subsystem for Linux (WSL) embraced the API – letting you run Linux binaries (and GUIs!) on Windows natively. Next, WSL extended the kernel with GPU driver support.
"" " - https://matt-rickard.com/embrace-extend-extinguish/
I only looked at the binary example and it’s some of the worst C# code I’ve ever seen. It’s synchronizing every task in a loop instead of waiting for the tasks to complete like the Java example. It’s returning a result instead of using an array like the Java example. The 2 examples are not equivalent.
I know no C#, but you can contribute a better one if you prefer. Some of the languages have multiple implementations in the games and their performance vary a lot.
I know, I've written NodeJS for a very short time, and remember liking it. However I can't position it into my current ecosystem where code is probably be seen by multiple people and need to understand/improve it.
I recently worked with some .Net developers who wanted a primarily Java based company to let them work with .Net. In their opinion .Net was so much better (faster, larger ecosystem, more developers) that it would pay off quickly. When I asked them for documentation, they pointed to some microbenchmarks that had no significance to the type of work they would do. They also complained that Java was more "corporate" than ".Net", which is ridiculous in my opinion. Even if that was the case, the differences would be a "rounding error"
Until Microsoft can match the dev experience of writing front and backend in typescript with pnpm, I’m not touching the dotnet/nuget cli(s) with a 10-ft poll. If you want to write C# just write go
I don't get this. If you want to write C#, write C#. If you want to write Go, just write Go.
Go is a great language because of its simplicity and ecosystem, has lots of useful things in the standard library, has pretty good runtime performance and also compiles to static executables easily (which is especially useful for projects like Nomad). It's both a good fit for web development, as well as writing smaller or larger utilities. Honestly, its packaging situation is leaps and bounds ahead of something like Python, so I predict great future for it in DevOps too.
C# is a more advanced language with a rich history and a lot of the functionality for web development in particular coming as first party packages. Some of that historical baggage weighs it down and the complexity can be annoying, but ASP.NET Core, EF Core, Kestrel and many other components are great. Plus, running on *nix is good, even though the single executable/runtime/deployment situation isn't quite as easy as with Go.
Write code in whatever technologies work for you. My caveat to add would be that I'd (almost) always develop front end and back end separately, since the React/Angular/Vue webapp really shouldn't care much about what technologies are behind the APIs. Of course, being able to use a single language for both FE and BE development is also a worthwhile approach!
That’s because my argument has nothing to do with the language itself but you still felt compelled to write an essay. I’m saying the modern Linux-based tooling, ecosystem, and devex of C# is inferior to so many other choices.
I don't get it because the claim that "If you want to write code in language X, just use language Y" doesn't vibe well with me. It feels needlessly dismissive somehow.
Hence the exploration of the good things about each of the languages and genuine reasons why people might enjoy using them for slightly different use cases.
At the end of the day, everyone can make that choice for themselves, unless their org makes it for them.
VS Code has a C# debugger on Linux. There are certainly limitations to the open source C# support in VS Code, but for students it should be easily sufficient.
If you want a language where not only the core and main libraries are open source, but also all the tools then C# is probably not the right choice. If free (but not open) tools for most use cases and paid tools for larger companies or higher-level demands are okay, then C# fits.
OmniSharp (VS Code extension for C#) has been discontinued and there is no replacement yet. Besides, Microsoft purpusefully sabotaged OmniSharp to make it so unusable that it forces people to use Visual Studio, so even if a beginner on Linux would decide to try C# with a discontinued product they would only face difficulties and a basically horrendous mostly broken experience. .NET on Linux is non existent without Rider and there is no way to sugarcoat this fact. It's sad how ignorant some people are about this and keep peddling untruths about C# that it is well supported on Linux when it isn't.
It is not discontinued, it is regularly updated, and all the interesting parts are coming straight out of the compiler project (Roslyn) which will be the source for all future alternative extensions. Surely, they have different interest and do not put too much money into OmniSharp.
Do I like that some part of the stack is closed source (like the debugger and in future the extension host): No! Does it lead to a broken or discontinued future: No!
Compared to my experiences with PHP, Java and JavaScript, C# on Linux is just fine.
I use vscode on Linux without any problems. Can debug and do all I need to do without any problems. Omnisharp may not be perfect but it works. It will be interesting to see what is going to replace it but short of the closed source concerns I’m fairly sure it will still work.
Rider is free for students and that is how I started using it.
Also, VS Code with OmniSharp isn't that bad for smaller projects for development and is pretty good for debugging for apps of all sizes (my 2 year experience with it).
So to claim that you cannot debug C# on Linux for free is false.
Only for some students. If you don't have an .edu address and your college or university is not in their pre-approved list, it's difficult to get a free license. It's a pretty typical attitude towards us coming from third-world countries though, and we developed our own ways of dealing with this.
And please stop this fake reality bullshit. NOBODY, not a single freaking person who wants to learn a new language will start off by searching the internet for an IDE and debugger where they can apply for a free license and then wait a few days so they can try it out. That is just ridiculous to even suggest.
If one wants to learn a language they download the SDK, open VS Code and look for an extension. If there is none or the one that exists doesn't work they basically say "WHAT THE FUCK IS THIS SHIT", close it and walk away with the conclusion that it's not well supported. Let's stay real please okay.
Then you should say: "a person who just starts with .Net".
> And please stop this fake reality bullshit. NOBODY, not a single freaking person who wants to learn a new language will start off by searching the internet for an IDE and debugger
I started with .net only 4 years ago. I went ahead and got myself Rider because I dislike VS Code and I already had 6 years experience with other JetBrains IDEs.
Oh look. I'm that "no single person" who went ahead and looked for an IDE. And bought it
> Let's stay real please okay.
Screaming nonsense at the top of your lungs doesn't make you any closer to reality.
> And please stop this fake reality [...]. NOBODY, not a single freaking person who wants to learn a new language will start off by searching the internet for an IDE and debugger where they can apply for a free license and then wait a few days so they can try it out. That is just ridiculous to even suggest.
I actually did. JetBrains products were really nice to try out whilst studying in university and eventually lead to me purchasing the ultimate tools package for any personal or professional projects that I might want to do.
It's really amazing to have the same set of tools for almost all of the languages that you might want to work with, that work consistently across all of the OSes that you might use. Normally I avoid paid software because of the vendor lock in, but a lot of the time it feels like JetBrains are just the company that you should reach for to avoid headaches, for better or worse.
For comparison, in the Java ecosystem, you also have Eclipse which has performance problems and an odd/unstable plugin ecosystem (though some swear by it) and NetBeans which has been handed over to Apache but doesn't really get much love or attention towards its development at all. Of course, some might use Visual Studio Code or another text editor with plugins but not being able to do refactoring across a project with 4000+ source files makes that approach dead on arrival for me, for all but the smallest projects.
The situation is pretty similar in regards to C#: you have Visual Studio which limits the platforms, JetBrains Rider which is a paid project, Visual Studio Code plugins which are insufficient and also something like MonoDevelop, which has pretty much been abandoned.
Every benchmark is a cheat and not a real-world problem.
I also thought their benchmark is a cheat. But it is not. It is testing exactly what the benchmark was testing in that scenario. That is why TechEmpower Benchmark suite is so powerful. It uses different web server scenarios (connection, JSON parsing, database connections, ...). The different test case implementation .NET and others have are making sure no other factor is influencing this. For example: when you benchmark connections why dealing with MVC, validation and database connections.
It is not .NET who cheats, it is the others who do not properly isolate the tested performance aspect.
And to make the comparison a bit more real and day-to-day applicable, you need to pick the right framework configuration (like aspcore-mw @ 80% / aspcore-mvc @ 37%) and compare with other reasonable scenarios (like express @ 1% / spring @ 2% / quarkus @ 8%). Both are then far of the top 10 ten list but still comparable. The top 10 combinations are for cases where developer productivity does not matter but throughput is the key (e.g. a DoH resolver or a system like a Cache / Database).
Coming from Java, Go, and Javascript, C#/.net just plain sucks. Things that should be simple, e.g. Collections APIs.. aren't. I stopped trying after this realization.
I preferred the C# from 15 years ago compared to whatever is going on over there at present. If you're already indoctrinated into the ecosystem, I'm sure it's fine. Otherwise, buckle up - things are going to get choppy.
Maybe I was using it wrong? As I recall, "the usual" APIs present in most other OO languages were not available when I tried to help a friend with a project 3 montjs ago.
Which collection APIs were you missing? Between the standard interfaces and their default implementations[1] and the extension methods to IEnumerable[2] you can do quite a lot in my experience.
You might've forgotten to reference LINQ and Collections.Generic in your project? Because C# definitely has good collection APIs that influenced many languages including Java Streams.
Folks here still think it’s Windows-only and comparable to Java. Yet latest TechEmpower benchmarks shows .net running on Linux and being faster than Go, Python, Node, and Rust.