EgorBo
2.3K posts
EgorBo
@EgorBo
I work on .NET JIT Compiler 🚀 at Microsoft
Warsaw, Poland Katılım Kasım 2010
179 Takip Edilen5.5K Takipçiler
@EgorBo I have not seen many myself in my areas, so not overwhelmed, at least yet. But people do wonder if they can use GenAI for their work, hence the policy decision, AFAIU.
How are your repos doing?
English
OpenJDK Interim Policy on Generative AI: openjdk.org/legal/ai. This aligns with my experience using GenAI tools: they are good at grasping some dusty context corners, but for surgical OpenJDK work you need to absolutely own every single line you push into the project.
English
was messing with the OpenAI base URL in Cursor and caught this
accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast
so composer 2 is just Kimi K2.5 with RL
at least rename the model ID
Cursor@cursor_ai
Composer 2 is now available in Cursor.
English
🫤 you know high quality contributions are coming when the name has "Claw" in it #issuecomment-4081653414" target="_blank" rel="nofollow noopener">github.com/ollama/ollama/…
English
Let this sink in: "The output you are getting from your coding LLM is a direct representation of how good your thinking is."
English
🙈
Let's face it: for such a small tool, entirely vibe-coded, there is zero advantage to using .NET.
Among the people who use coding agents as a CLI, I expect that cargo is more widely deployed than the dotnet sdk.
And people who use VS will prefer a VS extension to a console tool. x.com/nietras1/statu…
English
I like Copilot CLI but the billing model stresses me out a little. Because there's no hourly or daily limit, I worry that I might run out of requests before the end of the month, and use them very conservatively. Then I end the month with a lot of unused quota.
So I built a simple CLI tool to tell me how many more requests I can use for the day without worrying, along with a few additional stats.
You can install it with cargo:
cargo install copilot-usage
English
@xoofx @controlflow @James_M_South @antonfrv another issues with the tag is the thread-safety (lack thereof) which is the source of type safety issues/races for gc slots 🙂
English
@EgorBo @controlflow @James_M_South @antonfrv Yeah, I know, but you can segregate with a simple tag field + other fields per kind (object vs unmanaged field), no need to have GC support for this. I have even a 10 years old post about that topic 😅 xoofx.github.io/blog/2016/10/2…
English
"They declare a struct which uses a single object reference for storing its Value, which means: Boxing" 😮💨
Julien Couvreur@jcouv
We merged an early C# 15 preview feature into .NET 11 preview 3: unions. Adds union declarations (`union Pet(Cat, Dog, Bird) { ... }`) and union types (attributed with `[Union]`). They can be treated by pattern matching/switch expressions as a closed set for exhaustiveness.
English
@xoofx @controlflow @James_M_South @antonfrv some things are simply not possible in GC (esp, precise GC) languages, e.g. we'll never be able to make a tagged union of a gc and a non-gc field in the same slot
English
@controlflow @James_M_South @antonfrv 1. It supports non-boxed via custom [Union] struct, but not for the union declaration syntax itself.
2. Not really. (Tagged) Unions for example in Rust are squashing value into similar slot + a discriminant slot (when not optimized, like Option<&T> that occupies only a pointer)
English
Jupiter shots from Saturday; with little animation included. app.astrobin.com/i/hl230w
English
The people have spoken; they yearn for io_uring
Top all time in dotnet runtime 👍
github.com/dotnet/runtime…+
Ben {chmark} Adams ⟠@ben_a_adams
It is time. Let's go faster! Let's use io_uring for dotnet System.Net.Sockets on Linux! PR124374 Replaces epoll with managed C# io_uring engine: batching, multishot accept/recv, zero-copy send, registered buffers, SQPOLL (zero-syscall), and more 🚀
English
@EgorBo Got it to write my PR description but I think it might have overdone it? github.com/dotnet/runtime…
English
@jhigh2000 There are ~400 PRs authored by Copilot in runtime - most of them are trivial things. But there are many helpful commits = suggested changes in human PRs. github.com/dotnet/runtime…
English
@EgorBo Aren't most (if not all) of those copilot commits because it's listed as a co-author? It would be cool to see what copilot is actually committing, including autonomously.
English
@nonlinear_james I am sure there will be bugs introduced in Copilot PRs (just like with human PRs). But my gut feeling is that Copilot PR reviewer alone has already helped a lot in detecting bugs in my PRs early. So, I would say it's a by far net win for overall reliability of .NET
English
@EgorBo Note to self: Do not upgrade to .net 11 when it comes out. It will be enshitified like Windows 11 and Notepad.
I hope someone at Microsoft engages their brain soon and realizes that they're committing suicide as a company.
English
@nickersfpdx Copilot has found many potential bugs in my PRs that human reviewers might have missed. Even if a bug occasionally sneaks through, I think the net result is a significant improvement in reliability.
English
@EgorBo Is it weird that seeing this makes me trust the product less... 😯
English
Do you know that just changing the C# langversion might affect performance?
The code generation for method group conversion has changed in C# 11 to avoid allocations in certain cases.
English
Есть абсолютно человеческая "вера в справедливый мир", которая заставляет искать рациональные причины в происходящем, чтобы уберечь себя от мысли, что в любой момент может случиться любая левая хрень, абсолютно вне твоего контроля. Вот и у айтишников в эпоху сокращений так же.
Русский