Out of curiosity, why not adopt Wine instead of hand rolling the Win32 API layer, and also why not use MoltenVK + dxvk for rendering? Have you tried these alternatives to compare the performance?

The thing is that both Wine and DXVK would still need the original x86 code to run through Rosetta. Which is exactly what I wanted gone. Here it’s all ARM64 before the game even starts. DXVK is also a Windows DLL that reaches Vulkan through Wine so without Wine I can’t really use it. And BO2 is 32-bit, so every Windows and D3D11 call has to be adapted anyway! Going straight to Metal ended up simpler :) No benchmark against them yet but it’s on my list! There is a short explanation in the README too if you're curious: https://github.com/adrbn/bo2-mac#pipeline

Especially since Rosetta is EOL in the next macOS. I've been playing with D3DMetal recently which depends on Rosetta. I wonder if Apple are going to just drop support for this.

   > till this day you can't use linq to render a sequence of elements. Like people.Select(p => <div>{p.FullName}</div>);

In Blazor, you can do it with something like this:

   <div>
       @people.Select(p => @<div>@p.FullName</div>)
   </div>

Rendering-wise, it's better to use a @foreach [0], to avoid creating too many rendering fragments, but it works.

[0]: https://learn.microsoft.com/en-us/aspnet/core/blazor/compone...

that is not equivalent - for instance you can't save the result to a variable and work on it further:

    var thePeople = people.Select(p => <div>{p.FullName}</div>);
    var tranformed = thePeople.Select(d => d.Children.First()).DoSomethingElse();

The more I use blazor, the more I realise it's just old-school Webforms with a succinct pseudo-C#-syntax. it's very limiting compared to JSX

Interesting to notice that the best benchmarks of the model are not on Meta's own agent harness, but instead on mini-swe-agent:

   > 1. All models run at max reasoning effort. Artificial Analysis results are from Intelligence Index v5, as published November 9, 2026; cost per task uses each model's list pricing. Internal results are from Meta's evaluation suite. The Personal Superintelligence Index (PSI) is an internal composite of eight evaluations: Agents' Last Exam, AutomationBench, FocusBench, JobBench, OfficeQA Pro, GDP.pdf, GDP.xlsx and VBT. Terminal-Bench 4.0 uses each model's best published harness: mini-swe-agent (Muse Ember), Claude Code (Claude Haiku 5.5), and Codex (GPT-6 Luna). OSWorld 2.1 uses partial-credit reward; PerceptionBench is scored with tools.

This is absolutely huge. If the Expo team can demonstrate that this development strategy is viable, we will finally see a resurgence of native applications and more efficient software, getting rid of the Electron bloat.

It's a long shot for now, though! There are lot of gritty nitty details they have to get right. One of the first issues I can imagine is this framework being great to use with Opus 5.5 but not good enough with i.e. Gemini 4 Argon or DeepSeek v4.1 Flash 1017, due to the overreliance on having AI models write the code.

If anyone is going to get it right - I'd bet on the expo team. They know these sharp edges and have been building for them for years.

Rust-based Discord achieves both, though... You can absolutely get an application to run faster while using less memory, given there's enough low-hanging fruit (and with today's software, there certainly is...).

If you're optimizing sorting algorithms, video decoding or compression software, then that's another story, because using more memory can trade time complexity for space complexity in your algorithm (i.e. Merge Sort vs Insertion Sort).

Good, finally! Yet this branch started with the claim which isn't about Discord at all.

This is a thread on the post about Discord and we're discussing the original statement below:

   > All the performance and other arguments are just not true, a well build webapp is often more performant than a native app especially for long scroll lists not even to mention accessibilty, extensions and more.

To which I think Discord's Rust rewrite is actually a great counter example. Epic Games's new launcher could've been too, but it's just vibe slop, unfortunately, not real engineering work.

The announced performance and memory usage also shows that using a browser tab (reducing overhead for separate browser processes) is still not enough to save memory compared to a native application, as this new Discord rewrite, with multiple accounts logged in, uses about the same memory as an empty new Chromium tab.

For starters, logic doesn't work this way at all. Counterexample doesn't make a valid argument against non-absolute statement. And the claim is that in-browser code can be as performant as natively compiled (as you conveniently quoted). So apparently building the case on Discord misses the point twice.

Also for the sake of correctness: Discord CTO said the Rust-based app with 2 accounts uses around 290 MB which is significantly more than even the first empty tab uses (and each of the next will use smaller amount due to sharing).

   > the text renderer in Chrome is just wildly optimized and definitely can outperform an OS list view.

The question is not whether it can win against GTK/Cocoa/WinUI but whether it's faster than other retained-mode renderers such as the one Discord is using, GPUI.

Also, even if just running on a browser tab, each tab has a lot of memory overhead compare to what an actual binary running on the OS has. That's memory that gets wasted from the outset, before the developer even writes a single line of JavaScript.

To anyone that doesn't know: Vivaldi's code is partially closed source [0]. It's not much better than Google Chrome on that regard.

[0]: https://vivaldi.com/blog/technology/why-isnt-vivaldi-browser...

If I don't remember bad its just the GUI part, anyway for me is fundamental in my workflow due to its usabillity and productivity improvements.

They say it's just the GUI.

Shame there's no way to check something else isn't included too.

For those interested in a GPL-licensed browser from Norway:

https://nordstjernen.org/index.html

replying to my own post to follow up:

It is apparently furtive slop.

https://news.ycombinator.com/item?id=48151418

Is it just me or is it a bit sad that even GPL'd software has websites that are full of Claudisms these days?

Like have some pride when showing off your project. Write a few sentences. Why should I give a shit about your project when you don't even care enough to describe it in your own words?

Is that the case though?

The homepage mentions this:

"It ships no telemetry of any kind, and deliberately includes no AI assistant and no AI-style web APIs."

I am not sure the website is AI generated. Can you be more specific why you think this is the case?

OpenSSH considers this to be a vulnerability in musl and not OpenSSH, but shipped a fix anyways since that helps mitigate impact... Whether you agree with that or not, it's up to you :)

You are... That's why you can guess the driver :)

Only available to Google AI Pro subscribers. Google AI Plus hasn't got Gemini 3.8 Flash nor Gemini 4 Argon.

What's then the use of the Plus subscription?

Same as the ChatGPT Go: higher usage limits, more image and pdf uploads, more tools

ChatGPT Go gives you access to Sol which is much better than the ancient 3.1 Pro.

I agree, I don't think Plus is a good value proposition

But also no storage

Oh okay. Though OpenAI's model (GPT-6 Luna) seems a lot better than Google's (Gemini 3.5 Flash-Lite), so I don't see why anyone would pay for Gemini AI Plus.