TomaszZielinski

Yeah, I forgot there’s the intermediate VM level, and user folders are shared there so that folders could be mounted to the individual containers using host paths.

My personal take is that the only way to be reasonably sure you're OK is to install as few apps as possible and then as few plugins as possible (and ideally stick to the bundled ones only). I don’t think it’s controversial, but for some reason this is not how many people think, even if in the real world you don’t give keys to your place to everyone who says they’re cool :)

I „love” such sandboxing defaults. Apps like Docker Desktop also share the whole home by default [1], which is pretty interesting if a big selling point is to keep stuff separated. No idea why node_packages need to have access to my tax returns :). Of course you can change that, but I bet many users keeps the default paths intact.

[1] https://docs.docker.com/desktop/settings-and-maintenance/set...

I believe LS has some protections against this. Never tried them, but there are config related security options, incl. protection against synthetic events. So they definitely put some thought into that.

I treat LS as a privacy/anti-telemetry/anti-accident tool, not as anti malware.

Obviously it can detect malware if there’s a connection to some weird site, but it’s more like a bonus than a reliable test.

If you need to block FS access, then per app containers or VMs are the way to go. The container/VM sandboxes your files, and Little Snitch can then manage externa connectivity (you might still want to allow connection to some legit domains—-but maybe not github.com as that can be use to upload your data. I meant something like updates.someapp.com)

Ahh, this brings memories - back in the day GORILLAS.BAS was a perfect game, and also a mind blowing piece of code (for a kid, ofc)!

As a kid I had Atari 520ST(M) and GEM was like a… window to a magic world. It was so different from anything I had seen before (older Atari, ZX Spectrum, C64).

Funny thing is that it was also my window to Turbo Pascal, because there was a PC emulator (8086 on an 68000!). It run very slowly, but fast enough to be usable.

The contrast between the magic of GEM and the crude text mode of DOS was another thing I remember - I think it made DOS much more exciting than it was in reality :)

A few months ago there was a seemingly similar crash in Sonoma 15.4, on some M1s [1], and AFAIR it was fixed in 15.4.1 ~2 weeks later.

[1] https://discussions.apple.com/thread/256030581

Safari is interesting. It's been separate, except for major macOS updates, which had it bundled. But if you had a newer Safari on an older macOS, and upgraded macOS to anything else that the latest version, then your Safari was downgraded, often causing data loss..

In Sonoma or Sequoia they started bundling all Safari updates with macOS, but right now Safari 26 appeared as a separate update in Sonoma/Sequoia—-and it will likely stay that way.

Each thing separately can be explained, but when put together it’s somewhat messy..

Let me play devil's advocate for a second, because I feel it’s more nuanced.

Pros:

- [for users] 15.8.5 patches one high-profile bug.

- [for Apple] minimal effort which translates to longer perceived support time

Cons:

- It leaves unpatched multiple bugs fixed in iOS 16-26, and so it might give users false sense of security

I'm on a fence here, especially without real numbers

Yes, I also have fond memories of quite a few TUI apps for DOS. Not sure if it’s pure nostalgia, it might be. But then it feels like dark magic that you could have 40kB .COM or 100kB .EXE doing so many things and looking so nicely..

Yeah, that's what I meant by hardening the existing stuff.

It's very good they are doing it and we're all more secure thanks to that, but it’s not a caps-based system designed from scratch.

Well put! But I think there's an interesting exception—APFS seems to be very reliable. It's been quite a few years since the very successful silent auto migration and it’s pretty quiet about it, which is a good thing for filesystems.

Ah OK, I see it now—I focused too much on the „magic syntax”.

As for hooks, I guess I mentally treat them as magic that lives in its own lane, and so I've never really thought that I could do anything with them, other than what the manual prescribes.

I don't know much (if anything) about it, but it can be turned into an interesting thought experiment.

Let’s use Apple as an example, as they tend to do major transitions on a regular basis.

So, let’s say that the top tier already approved the new security mode(l).

Now, how to do it?

My understanding is that most if not all APIs would have to be changed or replaced. So that's pretty much a new OS, that needs new apps (if the APIs change, you cannot simply recompile the apps).

Now, if you expose the existing APIs to the new OS/apps, then what's the gain?

And if you don't expose them, then you basically need a VM. I mean, I don’t know Darwin syscalls, but I suspect you might need new syscalls as well.

And so you end up with a brand new OS that lives in a VM and has no apps. So it's likely order(s?) of magnitude more profitable to just harden the existing platforms.

It's not my experience, but then I have a „living” GEMINI.md doc where I add/clarify/tweak what to do and what not to do. And it's possible the initial revision already contained the correct spell :)

I can imagine that some race condition slips into production, but what's hard for me to process is why it’s 5 or 6 years later and it’s still not fully fixed.

I mean, even if you have no idea what's the cause, you e.g. stuff counters everywhere and when they don't match you send the telemetry with the details. Privacy is preserved and over time you get an idea what to look for.

I admit I have no idea how mail client works, but clearly there must be some way they could pinpoint it, with such large userbase..

I've found that LLMs help me find the boundary.

First I ask a bunch of questions in a chaotic manner. I explore the topic, check references, etc.

At some point the dots start to naturally connect. Then I start paraphrasing what I learned and the LLM either confirms it or clarifies where my understanding falls short.

At some point I feel naturally satisfied with the level of understanding that I have—it's likely because there's no „one more page of Google search results” trap there.

One thing to watch out is the „GOAT trap”—for instance, the default ChatGPT tends to reply with sth like: „You are the GOAT and your understanding and insight are unmatched. Let’s just clarify a few minor points”, followed by a destruction of my line of thinking, but worded in such a way that you're happy for the upcoming trip :). So you need a system prompt like „be very blunt”.

I mean, your router is the single key to your kingdom—-your local network. If you don’t treat all your local devices as hostile (which is a reasonable thing to do but almost no one does it), then having your router in shape is somewhere in the important to critical range.

Re 1., based on [1] it seems that some data loss bugs are getting fixed. Asymptotically :)

[1] https://mjtsai.com/blog/2019/10/11/mail-data-loss-in-macos-1...

Also things like:

    _('There are:') _('%d items', count=len(items))

—-which look correct until you want to translate them into a language with a different order of words in a sentence.

My favorite reply is something like: „You’re The Real GOAT!!! And now let’s just quickly clarify some minor points”, followed by a complete destruction of my arguments :).

As a rookie with gray hair I completely agree :)

I can’t recall a single time where I would want to put JSX in an „if” statement.. In fact I can’t recall a single time where I would want to use JSX in a different context than the return value of a render phase.

So from my personal experience the things you mentioned are non-issues in practice. Do you have a different experience?

Yeah, well put tech wise (the ad personam part puts me off a bit, though). Personally I like hooks quite a lot, but e.g. I feel I have to check once a month or two if a memoized component (that’s „functional” in name) re-renders if its props stay the same and the only change is the state..

Another vote here against using LLM. Because if you do ERP incorrectly then you can even worsen the fear and make your life even harder.

Example:

You are afraid of dogs. LLM tells you to approach one to habituate your anxiety. You do that but the fear overwhelms you and you run away.

Brain’s diary: I avoided death, that beast would shred me to pieces in a moment.

(Note this is about „normal” anxiety and not OCD, but the idea translates.)

If you dont want (or e.g. cannot afford) a proper therapy, IMHO a book by a competent author is a much better choice than LLM. Or a few different books.

Technically you’re right, but it’s like some antiviral therapies—technically the virus is still there, you are on pills forever, but your quality of life gets soo much better.

Pretty handy, right :)?

And seriously, managers like 1Password clear the clipboard after some time. I would guess that there’s some clipboard API that allows managers to exclude copied passwords from being permanently added to the history.

Still, there are pieces of data that one might not want to store in such unobvious place as clipboard history so it’s good to know about it.

I click one icon, then another. It takes say 2s. Typing two letters and pressing enter would take 10x faster, so 0.2s. Given that I delegated work to AI agents, that’s 1.8s less of waiting :))

Usually I just go with the flow, because what else I could do :)?

But somehow the missing App Laucher made me bit sad (well, to the extent software can make one sad :)) - even though I can always switch to Finder to browse apps, App Launcher has some nice visual quality to it that makes it more pleasant to use for me..