Extension modules are implemented in C because the interpreter is written in C. If it were written in another language, folks would write extension modules for that language instead. Also, it would be less relevant if people used portable C bindings like cffi, which are portable to PyPy and other interpreters… but they don’t.

You tried to apply far too much pressure over too large a surface area. Either make a more focused approach by not chasing Free Software and XMPP supremacy at the same time, or find ambient ways to give people options without forcing them to make choices in the direction you want. In particular, complaining about bridges usually doesn’t get the discussion to a useful place; instead, try showing people on the other side of the bridge how wonderful your experience is.

Also, I get that you might not personally like IRC, but you need to understand its place in high-reliability distributed systems before trying to replace it; the majority of them use IRC instead of XMPP for their disaster recovery precisely because its protocol jankiness makes it easier to wield in certain disaster situations.

Watch the video. Wedson is being yelled at by Ted Ts’o. If the general doesn’t yell, but his lieutenants yell, is that really progress? I will say that last time I saw Linus, he was very quiet and courteous, but that likely was because it was early morning and the summit-goers were starting to eat breakfast and drink their coffee.

I want to run PipeWire as a system user and have multiple login users access it. My current hack is to run it as one login user and then do something like:

export XDG_RUNTIME_DIR=/run/user/1001

Where 1001 is the user ID. Is there a cleaner approach?

Well, I don’t want to pull the kernel-hacker card, but it sounds like you might not have experienced being yelled at by Linus during a kernel summit. It’s not fun and not worth the money. Also it’s well-known that LF can’t compete with e.g. Collabora or Red Hat on salary, so the only folks who stick around and focus on Linux infrastructure for the sake of Linux are bureaucrats, in the sense of Pournelle’s Iron Law of Bureaucracy.

I already helped build a language Monte based on Python and E. Guido isn’t invited, because he doesn’t understand capabilities; I’ve had dinner with him before, and he’s a nice guy but not really deep into theory.

Sounds like it’s time to start training code-writing models on leaked Microsoft source code. Don’t worry, it’s not like it’ll “emit memorized code”.

I’ve been using NixOS for nearly a decade. It took me several days to understand the filesystem layout, and I had the advantage of knowing some capability theory beforehand. However, once I understood the Nix store, my paradigm shifted and I haven’t had any further “unexpected troubles.”

As far as I can tell, AppImages and Flatpaks are extraneous, heavy, improperly isolated, and propagate a sprawling filesystem which is hard to secure. Compare and contrast with Impermanent NixOS, which only persists data that the user has explicitly marked to be saved and has systemwide caching of installed applications.

Maybe you should actually try one before sharing your thoughts.

I’d love to link you to their Wikipedia pages, but both of them are redlinked. As far as I can tell, Dr. V. Ronald was an educator who moved from Canada to the USA as part of the whole Xerox PARC thing and probably was valued for mainframe experience; does anybody have a full bio? The current maintainer is Ron Sunk, who did a full run at MIT up through postdoc before going to Red Hat. The names are a coincidence; runk implements what we now call Sunk summation, after Sunk’s thesis. (As you might guess, that’s an instance of Stigler’s law, since clearly Dr. Ronald discovered Sunk summation first!)

Also, as long as we’re here, I want to empathize a little with Sunk. The GUIs that folks have placed on runk, like GNOME’s Gunk or Enlightenment’s enk, look very cool, and there’s rumors of an upcoming unified number-counting protocol that will put them all on equal ground. But @MossyFeathers@pawb.social wasn’t joking; Dr. Arnold’s code literally only reads punch cards, and there’s a façade to make it work on modern Linux and BSD transparently. It predates X11, if that’s any help. The tech debt is real.

Because frankly, Ronald (the current maintainer, not the original author) is very competent. I say this as somebody who has personally been yelled at by Ronald at a kernel summit; I didn’t deserve it, but none of his technical points were wrong. I like to think of myself as the kind of person that, given enough time and documentation, can maintain anything; I think it’d still take three of me to do Ronald’s job. (Well, “job.” I think he technically works for Red Hat or something?) Not to excuse his conduct, just to explain why he’s not been replaced yet.

By induction, that will only make your user base stupider.

They have two. If the complaint is that neither wiki is as rich as the Gentoo or Arch wiki, consider that perhaps NixOS users don’t need as much supplementary advice for configuring their systems.

It’s been two decades. What kinks do you think NixOS has yet to iron out?

You’re not crazy or harsh. This is a very real problem. I have been stuck doing business development at every company I’ve worked for. There’s always some shitty load-bearing Django app whose schema determines what the business is capable of doing, and somebody’s gotta maintain it. It’s gotten to the point where I assume that any interesting things I do will be outside of work and not for pay.

python3Packages.scikit-image appears to be available and non-broken in nixpkgs; on my machine, I get /nix/store/w8681ncsw92cn4gq6gyraw4z19r0r6c3-python3.11-scikit-image-0.21.0. Do you have an actual example?

I understand your point, but given nixpkgs’ position in the community, it might be a moot point.

There are subfields of computer science dedicated to this question. A good starting point for the theory would be Pessimal algorithms and simplexity analysis, which lays out two concepts:

• The time & space simplexity of an algorithm indicates best-case lower bounds on resource usage, and
• An algorithm is pessimal if no equivalent algorithm wastes more time/space/etc.

For example, common folklore is that sorting has O(n lg n) time complexity, depending on assumptions. In the paper, they give that sorting has Ω(n ** (lg n / 2)) time simplexity; any algorithm which takes more time, like bogosort, must do so through some sort of trickery like non-determinism or wasting time via do-nothing operations.

It looks alright. You’ll have to use it for a few months before knowing whether it’s comfortable.

To be honest, I’m not a fan of variables; I’m in the tacit/concatenative camp. But I think it’s good to try new things and learn for yourself why they are good or bad.

The original workflow, which is still in use today, is git-blame followed by git-show: look up candidate commits, then examine their history individually. This can be accelerated with a GUI; e.g. GitHub and GitLab support blame-style views.

I agree that the UX of existing git commands is not great. They evolved in a particular insular environment – Linux kernel development with heavy mailing-list usage and large multi-headed merges, with occasional pull requests and manual integration testing.

Check out my top-level comment for a link to git’s data model. A data-first approach with blob, tree, commit, and tag can be enlightening. The on-disk format tries to balance integrity, easy manipulation, disk space, and incremental updates; it’s also weakly monoidal, enabling distributed development. Look up the history of Bitkeeper and git; this is “a version control system [designed] from the ground up with documented architecture from the start”! And there are many non-C implementations as a result, like pure-Python dulwich.