"Can't even match GNU's coreutils"? Care to elaborate? Last time I checked OpenBSD is pretty much in line with Posix and there are plenty of us that consider many of the GNU extensions to be anti-features.
I use a Mac, so the CLIs are the BSD versions. I don't have any opinion on which is better, but it's just annoying that it's different from GNU, so a lot of my work ends up being on a Linux VM or server instead. POSIX compliance doesn't seem to buy me anything. I like the BSD license though.
That's a brand new excuse, Apple refused to distribute GPLv3 when GPLv3 was first published, 20ish years ago. How is GPLv3 incompatible with their system volume, yet GPLv2 is still fine?
GPLv3 added language which specifically requires that GPLv3 software embedded in products be distributed with "Installation Information" explaining how modified versions of that software can be installed. It is not at all clear how this would apply to software in an immutable, signed system volume. This language was not present in GPLv2.
You're entitled to your opinion, but that doesn't change the fact that the situation is "Apple can't distribute them", not "Apple doesn't want to distribute them".
Continuing to "correct" "want" to "can't" doesn't make sense, even under your interpretation of the situation. Even assuming the GPLv3 requires all software on the machine to ship under a compatible license (it doesn't), they can do that. They don't want to.
Of course, but any script meant for Mac is going to assume Mac core utils, even assume zsh instead of bash. You don't want to be the one person on the team with a non-stock setup.
> The GNU versions work fine in macOS. Apple doesn’t want to distribute them, but that’s mean you can’t use them.
We also test GNU coreutils on macOS. It is a bit tricky since the other maintainers and I don't own any macOS machines for testing. But we seem to do well enough with a shared system that Adélie Linux kindly allows the compile farm [1] to use, along with some additional CI machines.
I don't know how compatible my Mac will really be with a Free/OpenBSD machine. Last time I tested this theory, some C code I was building on Mac clang didn't build on FreeBSD clang cause I was unknowingly using some extension enabled by default. But that was about 12 years ago.
I'm not trying to. And it's not even like GNU/Linux/whatever is a superset of POSIX and I'm using the extensions. There are mutual differences in basic functionality.
"[Doesn't] get the job done"? Well, it gets my job done. You are welcome to use whatever you subjectively prefer (I use GNU's date(1) for a few scripts, but would prefer the English date grammar parsing as a tool of its own), but if you are unable to talk about anything in concrete terms there is not point in trying to discuss in good faith and I will just roll my eyes and move on.
I am not the parent commenter, but I have the same sense that the GNU ones are better.
But the only concrete example I can think of is that `cat -n a b`, if `a` doesn't end in a newline, produces something that is obviously not what the user wanted. GNU cat does it right.
I didn't report it to the openbsd people because I didn't want them to yell at me that `-n` is nonstandard anyway with no "right" behavior, so I should STFU. Which would be what I'd expect from having reported problems in the past.
I think the behavior in the GNU clone is more consistent with "we added line numbers to the concatenation tool" while whoever added it to OpenBSD's ancestry was probably more concerned with generating listings for printing.
Personally, I'd remove even the POSIX [-u] flag from all cat versions because all of them are un-UNIX-like.
Some GNU tools are particularly ridiculous with the Easter egg features I only know about because LLMs use them all the time —and I have been using Linux for 30 years.
Because you don't like the design(I don't either) but it carries information.
Someone could have thought it was a good idea and no one was willing to audit all of UCB's scripts to fix the "bug" even if it had been the result of a lazy implementation because ed(1) on a terminal.
My choice of wording is from my assumption that nobody, at any point, actually wanted that behaviour. And I stand by that being a pretty safe bet.
So I wouldn't refer to it as "the design". It's not that the methods are unsound, it's that I don't see any method at all.
As for not wanting to break existing scripts, OpenBSD is not even remotely as concerned about backwards compatibility as Linux is. I don't mean that in a derogatory way. It's a perfectly cromulent choice.
It's been almost half a year since we got powerful enough models, and I'm wondering, where are all those great things that were created with them? Maybe it's just me, but nothing has changed in my life so far. No improvements, but definitely, I need to be more careful when reading, listening, watching, and using things, as bad quality slowly creeps everywhere.
Local AI has completely changed my relationship with computers because now I can build utilities, tools, scripts, and get debugging help with my systems with far less friction than I ever could before. I used to do all of this by hand, but now I can spin it off as essentially an async sub-process to the AI that I can return to a few hours later and have the job done. Everything from system upgrades to custom Steam Deck launchers and ad hoc service scripts for my home lab have benefited from being able to run a two-bit quant of deepseek on my Strix Halo.
I think it's going to remake our relationship with software because it completely alters the economics of creating one-off utilities, whereas software up until this point is generally regarded reuse as one of the highest virtues.
We're not going to see that impact for a while though. It'll take years. App stores will eventually become on-demand software and we'll view the marketplace analogy (where human developers offer their wares) as antiquated. I'm not totally confident in this prediction because requirements gathering is still a challenge, but for anything vaguely common, I think it'll be a workable model.
Mostly just focusing on positives in my life, not particularly bullish on the industry generating 6T/y.
Our team migrated a chunk of internal parsers to local weights purely to kill off query costs. The cloud api bill dropped by two-thirds in a single month
I'm seeing a ton of niche apps being built that otherwise wouldn't have been. As an example, NegPy is an app that gives you much more precise control when working with film negatives, and has been built largely w/ AI tools. There are a small couple of other photography related niche apps I've seen taking off, for example, for managing your Fuji recipes. I'm working on a Lightroom replacement myself, which I never would've taken on before AI tools.
In terms of personal life, I certainly find AI tools pretty useful especially given how awful Google is now, and coding tools are indispensable at work.
You’re all confirming the point by quoting niche examples of things that wouldn’t sustain a million dollar a year market, let alone contribute to a 6 trillion dollar one.
I was not agreeing or disagreeing with that point, I was answering the question that was asked.
I think if there's a $6T market, it will come via integrations with the products and workflows of large companies, which is obviously also happening. Apparently the tech sector as a whole is worth $5 to $10T, so it's actually not inconceivable there's $6T worth of value there (especially since the value will not be constrained to the tech sector).
I wrote a golf simulator that helps visualize golf architecture with help from Gemini. When I published my first piece, I got an email from Tom Doak and the head of data science at Arccos. An absolutely insane response for my dumb blog. I was invited to the Renaissance Cup this year, almost certainly because of that post; golf nerds can tell you what a huge deal that is.
Things are happening… and quickly, even if we are not quite to the point where VC’s are ready fund prototypes of robots to fold your laundry. Most of what’s happening is nerds finally building the project they’ve always dreamed of outside the traditional funding pathways.
For what its worth, https://www.weaverobotics.com/ is one of those VC-funded robot-to-fold-your-laundry companies! I think physical world implications are coming!
I expect to see this laundry folding robot when I can take the hyperloop between LA and San Francisco.
To be fair the chances of the laundry folding robot is quite a bit higher then that of the hyperloop, but my point is that I believe it when I see it. And even when I see it, I will be skeptical that this AI lunacy we are in now is directly responsible for laundry folding robots, that is I think we could have laundry folding robots without those trillions of dollars of datacenter investments and training of LLMs.
lol interestingly my neighbor built something similar! he built an app on the App Store for doing golf trajectory things on courses when the App Store was new. (sorry; not a golfer, though i suck at topgolf on occasion!) Sold it for some cash and is now double-downing on it with Claude.
If you are building for yourself instead of buying from someone else, you haven’t unlocked any value in the economic sense. Perhaps by making NVIDIA richer, but who knows how much they have indirectly spent through AI companies these years to get your couple dollars a month back to Jensen’s pocket.
I've been using it to build things that don't exist in the market (or GitHub). Either where the product just doesn't exist, or existing products do not fully meet my requirements.
For example I got a smart cat litter box and wanted to pair it to Home Assistant. There's an existing integration for the manufacturer, but it seems that although my box has the same model number, it's a different revision with completely different data point mappings - so the existing integration doesn't work. I pointed AI at it, and it figured it out, and it works.
Could I have done it? Sure, if I'd spent an entire afternoon doing it. I don't care that much about my cat litter box though.
GDP is known to have shortcomings. The example from HN a bit ago is mowing the lawn. If I mow my own lawn and my neighbor mows there, the GDP doesn't move. If I mow my neighbors for $200 and they mow mine for $200, $400 gets added to the GDP. So the software I wrote for myself won't show up in any GDP calculation. How happy (or not) I am over that software doesn't show up in the GDP. My subscription to the AI provider is the only thing that will. Now, will that, multiplied over millions of users reach $6T? I have no idea. GDP is a metric like any other and is just as subject to Goodhart's law.
You explained the problem brilliantly.
Frontier models are great at making you tinker and come up with shit you wouldn’t have done because there is zero value in it and you would have had to bet a value N (time, money) to obtain before this investors-subsidized almost-free software existed.
This is deluding individuals and CEOs and everyone who seems incapable to abstract from the sycophantic personal experience into systemic thinking.
To be fair, LLMs have been engineered to give this illusion, so the fault lies with the grifters controlling them.
Now are there useful applications of these tools (not gods) that might justify a sizeable new tech market?
Absolutely yes!
Is that market ever be going to be worth trillions a year? LOL hell no.
> Have you built that one application you always wished existed but it wasn't available?
Yes, more than one and I use them a lot. Not being good at design and not liking front-end, among a few other things, work always stopped from from building them before,
> So you've unlocked tremendous value, but you're not earning any money?
No, I also don't make money from my coffee machine, but it provides value.
I'm de-FAANG-ing my life. One AI subscription can do the lift of a lot of other subscriptions when it comes to putting OSS tools on my homelab, and I truly don't need _everything_ when I'm on the go.
I think a somewhat plausible scenario is that even if you made a great product that’s heavily AI aided, you don’t really gain a lot by trumpeting that fact. There are a lot of downsides to saying it’s made with AI and almost no upsides. So unless AI is somehow central to your product or you’re trying to ride the hype cycle, you’re probably better off not mentioning anything about AI.
Anecdata to be sure, but my personal experience with AI has been that I've used it to do stuff that is certainly useful but far from significantly transformative. It definitely helps make certain tasks easier and faster, but it doesn't do so consistently enough with enough of those tasks to make my overall life experience feel that much different.
Maybe I'm using it wrong, or maybe the tooling around AI hasn't yet caught up to the capability of AI models. It feels somewhat wrong to me that the benefit any individual person gets from AI is derived from their own prompting rather than autonomous systems that run in the background for them. But either way, the massive investment in AI compute seems very disproportionate to the actual benefit the average person, or even a tech savvy person, is seeing day-to-day right now.
I remember making a comment here on HN a couple weeks ago.[0]
One of the things that was promised in 2024 was a "deflationary spiral" for physical goods and the economy in general because of insane efficiencies that AI would supposedly bring. I believed that whole-heartedly and I was a huge AI fanboi even as my artists friends and Lib Arts friends were complaining.
We're in almost October 2026 now and they're nowhere in sight. We have a bunch of unreliable agents that do clerical and design work here and there, but where are all the industrial productivity boosts? If AI were so good, large industries and factories should be adopting them as a never-before-seen scale. Where are they?
Patience. I know it's hard being a huge AI fanboy among artist and Lib Arts friends who are in a cult of perma doom and hate. But I see artists using AI to make AI art that is fantastic and doesn't waste fossil fuels unlike Christopher Nolan's performative movie, The Odyssey.
Meta Muse, the #1 downloaded iOS app in the world, is only a few weeks old.
Opus 5.5, Sonnet 5.5 and Astra 6 are only a few days or weeks old.
Get new friends. They are holding you back and they will never trust you again anyway because you "betrayed" humanity by being a huge AI fanboi.
How much do you think Meta paid for that top advertising spot, or do you really believe that most normies in day to day life downloaded Muse to their phones?
I bet not a single one of my friends or coworkers, tech savvy or not, have downloaded muse.
> I know it's hard being a huge AI fanboy among artist and Lib Arts friends who are in a cult of perma doom and hate.
aw come on. no need to be insulting ... and I dunno about "doom", but a lot of the hate from artists is actually kind of justified, that didn't go away.
and tbh if you call yourself a "fanboy" unironically, kindly go easy on calling other people's behaviour a "cult".
We have replaced multiple 100k contracts from SaaS vendors with fully AI created alternatives. AI is a tool. In the hands of talented and efficient workplaces, it is like any tool. But, if you give the latest DeWalt OMT to a person who doesn't know or is scared of it, and put them in an organization that is full of politics and bureaucracy, then that tool is going to be useless.
You might be one of the rare few doing that. According to Stripe Economics, the supposed SaaSpocalypse brought about by AI actually resulted in accelerated revenue growth for SaaS. Companies just aren't building these custom SaaS-killers like HN would have us believe.
I've heard from my community that folks in large enterprises are blowing billions on AI with nothing to show for it. They have committees that decide how AI is used, and most of it is a fill out a form and go through tons of bureaucracy and needs buy in from multiple levels of leadership. By the time you do all that, you're better off just getting funding for a team to build you that internal (or external) tool internally or it's a SaaS that can be bought.
Like I said earlier with my tool metaphor, AI isn't a magical tool that can just be sprinkled anywhere for efficiency gains. Human systems are still a huge hurdle to efficient workplaces. That problem has to be solved first, and no one has figured out how to solve that in an enterprise setting.
We had at least 2 low code/no code tools that are pointless in the AI world. I honestly don't know the future of these companies because what is the point of lo-code/no-code tools when that's what AI does? I digress, but that was our reasoning, and its been almost 1/2 a year and no one regrets the decision.
computer scientists/software engineers and their completely messed up sense of how long it takes for things to actually take a hold in real life is always a bit amusing to me.
Half a year is not a lot of time when it comes to meaningful change in directions that matter.
Sure but the core point of commercial LLM’s pitch is tremendous speed up in development. It’s been speeding up for years now, yet like the commenter above I too cannot notice any improvements as an end user.
also, the part that is making lenders and such nervous is that they like to be paid on time on a schedule, and it's unclear how they will keep doing so without growing an increasingly large amount of debt.
Most big companies are performance theaters see token maxxing as one prominent example.
But go and talk to small companies. What they are able to pull off is just incredible. More features in shorter time. Improved UX and client satisfaction.
On a society level 6 months is nothing. This will take about 10 years to penetrate all organizations I assume.
Do people need this "more features in a shorter time"? I can say that, for me, I stopped reading changelogs because some software that I use started to update too often. If there are any new features that were added, I probably won't notice them.
And I remember times when people got angry at companies for "improving their UX" and breaking the daily routines of their users.
It's actually is an interesting point. On personal level I've built a few products but I did spend a ton of time exploring things I was really curious about that will never amount to anything practical beyond satisfying my curiosity.
> where are all those great things that were created with them?
You will have to wait great amount of time for great things just as you did before.
LLMs are not AGI, if so they are not foundational technology, if so we can't really built upon them - we can just use their wrappers as productivity tool, if so we will be bottleneck by what we were bottlenecked before: corporate bureaucracy, slow physical world innovation pipeline, supply chains, just a small number of actual problem solvers/risk takers around, etc.
There will be of course a lot of attach LLM wrapper and improve productivity here / there underwhelming weak fart products, ex: Meta Muse.
I've been working on my project, a paid mobile app, for five months now.
I'm pretty confident I can release it in October, after some 500 hours of work.
Usually in these discussions I get "so it has zero users then".
One time someone on here told me that for AI to prove its worth for software engineering, my application would need to be in use already for at least ten years.
The responses you’re getting are quite telling. LLMs have unlocked a whirlwind of shit where we can no longer trust most things, communities are suffering with noise, lack of water, increased energy prices, mass surveillance is rapidly increasing, private information is being leaked and sold…
But hey, you can now close yourself off even more, not interact with anyone or develop a skill to get a dumb bash script for a menial task you wouldn’t miss that much anyway if you didn’t have it. Woopie, totally worth it!
I formally proved an important 2025 CS paper (which I did not write) in Lean last month, it took like 2 or 3 weeks of intermittently poking at claude to keep going. AFAIK, it is the first rust-like borrow checker completely formally proven in Lean.
I'm currently using it as a basis for building a systems language with Rust's memory guarantees, but with new features like generators and co-routines, and effects instead of colored functions. The fact that the borrow checker's properties are formally proven means I can trust more of what Claude is doing than you normally would be able to do.
In early May, I broke my leg. Doctor eventually said get on a bike trainer and recover. So, I bought a trainer, didn't like the software that came with it, "wrote" my own with AI. Since it is just written with AI, I made it open source and free for everyone. I'm now months ahead of scheduled recovery. I'd call that a success.
Because you've hosted the code on github and the Readme is for humans but you've even slopped that.
I mean who TF does that? Rarely see something so moronic while browsing someone's github project.
Wtf. =)
I suspect the plan is not “make 6T revenue by having AI add >6T value in new and improved software”, but rather it is more like “make 6T in revenue by having AI substitute for ~6T cost or human labor”.
Not true due to induced demand. Traffic increases when more lanes are added to a highway. The alternative to $200 in AI costs is not $20,000 in labor costs. The alternative is nothing because the work was never worth $20,000. So the AI did not eliminate $20,000 worth of hypothetical labor.
AI creates demand for things that didn't exist previously
I just invented a simple upgrade script for my father's ancient Fedora install. I've gone from version 33 to the latest(44) with a simple utility I one shot in 5 minutes.
Since Fedora moves old releases to some mirror it's a nightmare to update old installs.
Now I did it very conveniently.
Did I build a space ship with the help of ChatGPT? No.
Is it useful to me? Yes.
I'm doing many much cooler things that I haven't released yet but I am launching projects I never dreamed I could do as a solo dev in 2-3 months.
I did some projects, but there are some things that I don't like. For example, when I wrote a program for my old ThinkPad, I didn't only create a useful utility for myself, but also learned and practiced Qt/C++. Now, I don't learn anything new, nor do I practice my skills when I "create" something. Maybe, the things I'm learning are how to prompt LLMs, add necessary skill files, and so on.
If you work in cybersecurity, it has changed you life. The pace of everything has increased exponentially. AI both makes the job easier, and harder, at the same time.
> I'm wondering, where are all those great things that were created with them
Maybe you are not looking in the right place. Maybe your expectations are not right. Maybe the revolution isn't a super app but a mini app. Not an app with code written so beautifully it could win a Pulitzer but an app that will be used a couple of times and then discarded.
I am trying to offer an alternate viewpoint (I don't necessarily support it but I thought it was needed).
I think it's more, let's layoff 20% of the people, give the rest some of the money saved in tokens, deliver features a lot faster (or at least expect them to). Jury's out how long this can continue before the "nobody understands the code anymore" chickens come home to roost.
A half a year! I was an arpanet(->Internet) user starting in the 70s and yes it was a massive research accelerator (for a tiny set of domains) but the real use (human contact, mainly mailing lists back then), while already strong, took decades to reach mass adoption.
I run my own custom version of BlueBubbles and a locally run clone of Wispr Flow.
In addition, I've vibe-coded lots of nonsense throwaway apps that helped me with a one time boring task.
thing is, I'm creating great things for our internal operations _from_ local models.
The idea that the commercial value will all be capitalized is a farce. Qwen3.8 seems to me to have hit the threshold to these things being good enough without every buying a single cloud service.
That should scare cloud AI more than it obviously does.
Among a few things ive boticed one notable is devops has been less of a blocker / faster to address issues. Most of the time the asks are trivial but incompetent Devops couldn't figure it out. Now they can ask ai and through their elevated permissions quickly solve the problem.
Strongly disagree, and I am one of those who believe this is a bubble.
At home, Claude has been a god send to make it easy to learn using linux, and all sort of technologies. My personal learning curve has gone steep at an age where people start flattening out (40s).
At work, we use AI to automate 50-75% of our work. Better quality output, more productivity. I work in a bank. But the average plumber or electrician benefits too. In minutes they can become an expert into some tech they never touched before.
I thought crypto was bullshit, web 3.0 [corrected] was bullshit, meta-universe was bullshit. AI is bubbly, not sure how you can justify multi-trillon valuations for companies with 80% operating margin in a competitive market where you can switch to the competitor with a drop down. But the productivity gains will be massive for the economy. I think the economy will ultimately reap the reward, not so much the labs. But I may be wrong.
> At home, Claude has been a god send to make it easy to learn using linux, and all sort of technologies. My personal learning curve has gone steep at an age where people start flattening out (40s).
Your own personal learning curve has also gone up when, if your thesis is correct, it is useless for it to have gone up because it is a skill that we may not need anyone to use, and we definitely won't need someone who learned it last week to use.
That's more of what people mean by no revenue. I'm actually bullish on some revenue coming in from language learning, because in-ear translation will always have a delay and involves constantly wearing equipment, but Linux learning seems like a big zero if models are supposed to be good enough to essentially replace OS UI.
> But the average plumber or electrician benefits too. In minutes they can become an expert into some tech they never touched before.
They're already fine at learning from the manual. They have a lot of background. The AI will 1) only annoy them with how much it gets wrong, and 2) not give them any practice with the things they have to learn with their hands. They may be helped by AI powered robotics.
I don’t think all of the data centers will ever be built. In that respect I see this bubble like the railroad building bubble of the 1800’s. A lot of data centers are going to end up in bankruptcy and sold for losses to a second round of investors.
I expect Open AI or Anthropic to make it through. XAI will go back to being a side hustle for SpaceX. And some open weight model will run on your own hardware or AWS. Copilot will remain a piece of trash and Gemini will be an untrusted 4th option.
I hope the chips are at least procured. When the bubble pops I would like to make a home rig which can run a 1tb model easily and then play games as well.
> But the average plumber or electrician benefits too. In minutes they can become an expert into some tech they never touched before.
No one “becomes an expert in minutes” at anything more complex than tic-tac-toe, that’s just absurd. If you don’t understand the thing without needing a chatbot, you’re not an expert.
To add to that, I don’t think many employers will be happy about their legal exposure to inadequate training and supervision when things go wrong if this is the sort of training people are resorting to. Especially if insurers are excluding coverage for losses arising from the use of generative AI, as seems to be happening.
What we've seen in the past with generally applicable technologies generally take 20 years to show up as productivity.
There was a Bank of Korea study that showed AI saved workers about 1.5 hours a week, with with no increase in output. Just because I'm more productive at a task, it doesn't mean the people up or downstream of my work benefit. And my time probably gets sucked into meetings. (In the study, solo entrepreneurs did benefit--probably because they own everything end to end)
In term of cool little projects, there's all sort of things that people are doing to improve their own lives or make interesting things. No shame if you're incurious about those.
I mean, I don't know that there is some sort of AI equivalent of the lifechanging magic of rural electrictrifcation. But I don't know that it was ever likely in the near term anyway.
And I'm personally not happy with that. To manually test something, we have QA, to shape business requirements, we have a business analyst and a product owner. Now a developer needs to be a 3-in-1 person?
For software within a certain complexity scope, this is true. Like the average SaaS CRUD experience that many web developers were building before AI - a lot of that can be handled almost entirely by agents, with very little human in the loop activity. But past a certain point of scale/complexity, this breaks down rapidly. In the past 8 months I've seen this occur several times, and if you only rely on agents to get you to that point, then you're in a heap of trouble when you do reach it, because it's very hard to backtrack.
AI has mostly been improving faster than the complexity of the projects people are capable of handling has increased. The lines are going to cross soon.
It’s more like goroutines or other lightweight concurrency mechanisms. If threads are OS-level concurrency primitives, fibers are scheduled within Ruby itself, which makes them much more efficient than threads.
In fact, I got the following results in HTTP benchmark tests:
Go:
* Latency under stable load: p95 0.25–5.32 ms, p99 2.20–9.92 ms
* Memory: 23–31 MB RSS across HTTP scenarios
Ruby:
* Latency under stable load: p95 1.03–6.45 ms, p99 2.32–8.30 ms
* Memory: 84–295 MB RSS, depending on the scenario
Fibers can also handle WebSockets better because WebSocket workloads involve more I/O waiting.
A typical Falcon setup uses N workers, one thread per worker, and many fibers. Since the fibers are cooperatively scheduled within a single thread, this avoids much of the context-switching overhead associated with OS threads. Multiple workers can still run in parallel across CPU cores.
To dig a bit into why fibers are more efficient than threads.
The OS has a rather opaque view on what happens in a thread. It doesn't know how the memory is used or what memory is efficient. The OS can block a thread when it runs into IO, but it doesn't know or care about what other threads in an application it can or should activate.
Fibers bring the threading into the application layer. While the application can't choose which and when a thread runs, it can make choices about which fibers to run. Further, the application knows intimate details about things like the stack of a given thread. When it goes to park a fiber, it knows just how much memory should be saved off so the fiber can resume and it doesn't have to save off all the memory allocated to a thread's stack. Further, because so many programs are stack based an application can pretty smartly save and reuse segments of the stack which are common amongst fibers. So, for example, if you spin off 1000 fibers at one location in code, the stack for those 1000 fibers will be identical right up until the fiber starts executing.
The main drawback of fibers is they can't implement things like fair scheduling. Applications have few ways to park a currently running fiber to let another one run if, for example, the app wants to make some progress on all the fibers alive. The app has to wait for the fiber to hit some sort of IO point in the code or insert explicit park checks (The JVM actually does this for GC purposes. It creates "safepoints" which application threads make a quick check to see if the JVM wants to start a GC). The OS has more power here, it can simply interrupt the thread and start running something else for a given quanta.
I'm unclear on the relevance of a comparison between Go and Ruby here. Ruby is radically slower than Go. It's down there with Python, if not a touch behind [1]. From the outside, I'd expect that it's possible that slight improvements in the context switching time will be dwarfed by the generally slow execution of Ruby itself; that is, if Ruby is going to take 100 microseconds to do something, whether it context switches in 1 microsecond (best-case OS thread cost) or .2 microseconds (best-case goroutine switch) is of somewhat less consequence then it is for a compiled langauge that can complete that task in 2 microseconds. That ratio of 50x is not just something I made up, it's about what you can expect in general. I'd need to see the actual Ruby benchmarks to come to any conclusions as to whether or not I'm right.
The other problem with this sort of benchmark, which is a mistake I also commonly see made by Node developers, is that the Ruby HTTP stack has significant native code in it, like: https://github.com/puma/puma/tree/main/ext/puma_http11 This is a good and proper thing that brings benefits to all involved; it's not like it's "cheating" or anything, it's a real performance benefit. But it does mean when you're benchmarking a simple HTTP server, you're benchmarking Ruby qua Ruby a lot less than you think you are, and so the relevance of such benchmarks to codebases that have actual Ruby in them will be less.
It was mostly HTTP/JSON tests with some sort of validation, basic auth and logging. I think Ruby is used the most for this. I see many companies switching from Ruby to Go or FastAPI for this basic web services stuff, and I have no idea why. Ruby needs to improve its memory management, but the speed is pretty good.
How much load? How many concurrent connections? What machine? Don’t get me wrong, benchmarks like these are useful to help ballpark performance floors, but really only relevant for a given load scenario. They don’t mean a lot without context. Not an attack by the way, just feedback.
That ruby fiber vs go goroutine benchmark is interesting. The 4-10x memory use doesn't surprise me, but the near performance does. I'm guessing there is more of a gap with the p50.
reply