Not negative at all, thanks for commenting. You're right that the answer is "nothing," and that this is a major trade-off inherent in the model. From a safety perspective, you'd need to be extremely confident that all threadprocs are well-behaved, though a memory-safe language would help some. The benefit is that you get process-like composition as separate binaries launched at separate times, with thread-like single-address space IPC.
After building this, I don't think this is necessarily a desirable trade-off, and decades of OS development certainly suggest process-isolated memory is desirable. I see this more as an experiment to see how bending those boundaries works in modern environments, rather than a practical way forward.
Well, we did have a term for it until it got dragged and sensationalized in the media. I'd tell you that's a standard "psyop" that the propaganda arm of the government often uses against communities and subcultures that they want to discredit and suppress for one reason or another but then you'd probably call me a conspiracy theorist[1].
[1] another example of a successful smear campaign
I actually just published a paper about something like this, which I implemented in both libriscv and TinyKVM called "Inter-Process Remote Execution (IPRE):
Low-latency IPC/RPC using merged address spaces".
Here is the abstract:
This paper introduces Inter-Process Remote Execution (IPRE), whose primary function is enabling gated persistence for per-request isolation architectures with microsecond-latency access to persistent services. IPRE eliminates scheduler dependency for descheduled processes by allowing a virtual machine to directly and safely call, execute functions in a remote virtual machines address space. Unlike prior approaches requiring hardware modifications
(dIPC) or kernel changes (XPC), IPRE works with standard virtualization primitives, making it immediately deployable on commodity systems.
We present two implementations: libriscv (12-14ns overhead, emulated execution) and TinyKVM (2-4us overhead, native execution). Both eliminate data serialization through address-space merging. Under realistic scheduler contention from schbench workloads (50-100% CPU utilization),
IPRE maintains stable tail latency (p99<5us), while a state-of-the-art lock-free IPC framework shows 1,463× p99 degradation (4.1us to 6ms) when all CPU cores are saturated. IPRE thus enables architectural patterns (per-request isolation, fine-grained microservices) that incur millisecond-scale tail latency in busy
multi-tenant systems using traditional IPC.
Bottom line: If you're doing synchronous calls to a remote party, IPRE wouldn't require any scheduler mediation. The same applies to your repo.
Passing allocator-less structures to the remote is probably a landmine waiting to happen. If you structure both parties to use custom allocators, at least for the remote calls, you can track and even steal allocations (using a shared memory area). With IPRE there extra of stale pointers because the remote part is removed from the callers memory after it completes.
(This paper is not out yet, but it should be very soon)
The best part about this kind of architecture, which you immediately mention, is the ability to completely avoid serialization. Passing a complex struct by reference and being able to use the data as-is is a big benefit. It breaks down when you try to do this with something like Deno, unfortunately. But you could do Deno <-> C++, for example.
I've spent years watching the same pattern repeat across project teams: the more people involved, the faster the coordination layer grows. Not because anyone wants more overhead — but because someone has to track what's changing, follow up on blockers, update tickets, and connect people who work in silos.
I built Prodigia because I think AI can now handle that layer.
Each project gets its own set of AI agents. They watch events as they happen — a status change, an assignment, a meeting ending — and proactively surface suggestions for what should happen next. No trigger needed. Suggestions appear in a feed where you approve, reject, or redirect them.
Beyond that: natural language project intake (describe a project, agents build the structure), live meeting transcription with automatic action item extraction, and per-project AI memory — preferences you set, a logbook the agents maintain over time. Agents can also handle virtually any action on the platform through conversation — update issues, reprioritize, create milestones, assign work.
It's in early access. The hardest part to build correctly was the multi-agent architecture and the tools associated to each agent — happy to go deep on that with anyone interested.
A pint in the Netherlands usually is 500ml. In very rare cases, but only in real pubs (not mass market "Irish" pubs) you get an actual pint. So you are cheated out of about ~68ml in that case. Vs the US you get a few ml more.
I'm not promoting "ChatGPT Checkout", however, just because something sucks now doesn't mean it will suck forever. I'm confident they'll iterate and improve. I say this because my 14-year-old niece's entire world seems to be ChatGPT; it's pretty much the only way she interacts with the internet. I don't think she's alone - all her young peers are like this, too. Retailers know this, so they've got no choice but to improve the experience of purchasing crap through an AI chat system.
If they can make the AI ajudicate the knowledge of the caller, I'm more than all for it.
"Hmm, this user seems to really understand network topology, better get him over to engineering"
vs.
"Hmm, the user doesn't know the difference between their router and their modem, I should help them identify the router then walk them through a power cycle".
Does OnlyFans have an extensive anti-trafficking program to prevent one typical method of exploitation? If not, it's just proving a platform for the traffickers to make money from the comfort of a safe jurisdiction.
Not sure why I got downvoted for saying none of the dozen alternatives people have suggested are anything close to Fastmail. I guess none of you have even tried Fastmail. I am a big fan of its UI, and they have recently released and official Linux client.
That made me laugh a bit as well. Definitely want to see some rigorous testing on that, I'd expect that on longer calls tha caller can make the ai say basically anything.
Americans don't claim to use imperial weights and measures; they use customary weights and measures, which were also used in the UK prior to the creation of imperial units with the Weights and Measures Act 1824.
I always assumed contributing to RFCs is about as easy as contributing to C++, which I always assumed is virtually impossible without a billion dollars or a billion citations of your academic papers.
Booking, boarding, change/gate notifications, rebooking options, customs and immigration is done via phone.
Transit to/from your the airport via Uber or a transit pass stored in your wallet.
Baggage tracking via airtags
Yeah, there's vague precedents for this stuff from the desktop computer era, but it only _really_ works when you've got an internet-connected device in your pocket.
A man still strongly emotionally attached to faded, worn out logos from many years ago is probably not an appealing signal to most women looking for a man to date.
I've been on death's door at least 3 times. Sepsis, heart failure, complications from Pneumonia. Each and every time, I've wished I had made more money and accomplished more. The money for my family after me, and the accomplishments for my own ego to feel like my life contributed to part of humanity. I have had a top 10 concrete goals since I was under 10. I have completed many, but not all.
> The same trick can also be used for the other direction to save a division:
> NewValue = OldValue >> 3;
> This is basically the same as
> NewValue = OldValue / 8;
> RCT does this trick all the time, and even in its OpenRCT2 version, this syntax hasn’t been changed, since compilers won’t do this optimization for you.
The author loses a lot of credibility by suggesting the compiler won't replace multiplying or dividing by a factor of 2 with the equivalent bit shift. That's a trivial optimization that's always been done. I'm sure compilers were doing that in the 70s.
After building this, I don't think this is necessarily a desirable trade-off, and decades of OS development certainly suggest process-isolated memory is desirable. I see this more as an experiment to see how bending those boundaries works in modern environments, rather than a practical way forward.