Login
You're viewing the functional.cafe public feed.
  • Nov 9, 2026, 9:52 PM

    @dymaxion @rob But the end user device vulns that actually have serious impact on the user have not been that numerous, and nearly all were preventable.

    A bunch were MMS apps irresponsibly accepting and auto-decoding media content from untrusted sources.

    A bunch more were browsers doing ridiculous JIT shenanigans rather than just having a simple and safe interpreter.

    Aside from that kind of stuff, there basically isn't attack surface on normie end user devices. At times the vulns have been treated as "in the OS" and requiring OS update, but that's only because stuff was gratuitously bundled into the OS. Which was entirely avoidable; the reason it wasn't avoided was monopolist behavior.

    💬 2🔄 0⭐ 0

Replies

  • Nov 9, 2026, 9:57 PM

    @dalias @rob Your device is almost nothing but attack surface. Every piece of code that processes off-device input — every piece of radio firmware, every format parser, every piece of app code that interacts with any data from off the device, it's all attack surface.

    Yes, the iMessage bugs were bad. They not only fixed the bugs but fundamentally overhauled how external media is parsed and literally gave users a switch to halt processing of media from anyone they don't know — they actually built the thing you wanted them to build. It's called lockdown mode, and it exists specifically for iOS users targeted with nation state malware. The apple security team, for all the insane nonsense the rest of the company is doing, is fighting this fight at a higher level, frankly, than 99.5% of the rest of the industry.

    💬 1🔄 0⭐ 0
  • Nov 9, 2026, 9:59 PM

    @dymaxion @rob Yes, Apple is (or at least was) moving in the right direction on that. Except only doing it in lockdown mode, that also forces you to make a lot of other tradeoffs. Media attachments should just never be decoded by default except for existing trusted contacts with e2ee channels.

    💬 2🔄 0⭐ 0
  • Nov 9, 2026, 10:03 PM

    @dalias @rob Cool. I look forward to seeing the system you go build — my calendar is looking pretty open in 2067, so just let me know when the release party is.

    The rest of us are going to go keep fighting for users in the real world.

    I think I'll end my participation here now — I was really hoping you'd come back with a conversation more connected to the world we live in, but I'm not sure there's much more to say.

    💬 0🔄 0⭐ 0
  • Nov 9, 2026, 10:03 PM

    @dymaxion @rob And the radio firmware is not attack surface if you architect the device correctly. Doing it wrong, like they do, is a commercial choice. One that's pressuring them to make the other bad and unethical choices I'm talking about.

    If you isolate the radio entirely from the host computer with just a generic high speed data channel in between, you don't have to worry about attacks on it. At worst they're DoS.

    💬 0🔄 1⭐ 0
  • Nov 9, 2026, 11:11 PM

    @dalias @dymaxion @rob

    A bunch more were browsers doing ridiculous JIT shenanigans rather than just having a simple and safe interpreter.

    Worth noting that we know how to do JIT properly and safely.

    They deliberately did not, for the sake of expediency & "performance".

    💬 2🔄 0⭐ 0
  • Nov 9, 2026, 11:25 PM

    @lispi314
    @dalias @rob We know how to do JIT in ways that formally guarantee that bugs impacting security are impossible?

    Yes, they made trade-offs. Were they the right ones, given a few billion users with low resource machines? I don't know. And yes, there are other places where they could easily have avoided perf bloat, but seeing teams as monolithic isn't really helpful when you're trying to guide engineering work — I know browser perf teams have genuinely been trying hard.

    💬 2🔄 0⭐ 0
  • Nov 9, 2026, 11:32 PM

    @dymaxion @lispi314 @rob Browser perf is like highway construction. It's a no-win game. As soon as you increase perf, adtech malware and bad web app frameworks will just expand to consume all available cycles until performance becomes intolerable again.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 9, 2026, 11:35 PM

    @dymaxion @dalias @rob

    We know how to do JIT in ways that formally guarantee that bugs impacting security are impossible?

    We know how to do JIT and assembly with full respect to the type system, contract mechanisms, and other language semantics of a higher level language, yes.

    So if the software isn't broken without JIT, JIT done in such a way does not break it either.

    Some useful keywords for relevant papers here would be: "typed assembly" and "proof-carrying code"

    Were they the right ones, given a few billion users with low resource machines?

    They weren't. The resource issue is entirely addressable by better system design and not insisting on dead languages.

    This applies to much more than just browsers, of course. I would apply it to the entire system.

    Anything short of the most constrained of microcontrollers would work then.

    Smalltalk ran in graphical mode on computers with less resources than several Arduinos.

    💬 1🔄 0⭐ 0
  • Nov 9, 2026, 11:42 PM

    @lispi314
    @dalias @rob Yes, in theory, a perfect JIT written with all the defensive techniques we have should not introduce new vulnerabilities. However, even if we do all that, what we're going to have is an actual JIT, and it will have bugs.

    Assuming any piece of software will not have security bugs has been comprehensively shown to be a mistake for anything dealing with real systems. "I have merely proven this code to be correct, not tested it" etc.

    Yes, we can do better. I've spent twenty years arguing we need to. I also have to work with real engineering teams to get them to do better in practice, and I've never seen a team in a position to do half the things we know we should do.

    💬 1🔄 0⭐ 0
  • Nov 9, 2026, 11:58 PM

    @dymaxion @dalias @rob

    However, even if we do all that, what we're going to have is an actual JIT, and it will have bugs.

    I believe that the size of the JIT is sufficiently constrained that full formal verification is possible.

    "I have merely proven this code to be correct, not tested it" etc.

    In all of those cases, the proof was clearly not complete and ignored various interfaces.

    The main case where this is false is when physical interference is a consideration, as this requires modeling correctness beyond software/logical interactions.

    This is usually supposed to be handled by the hardware. ECC is an example.

    Mainframes and aeronautics have various more examples. Software design is still affected, but it's entirely insufficient on its own to handle all the issues they purport to address.

    Assuming any piece of software will not have security bugs has been comprehensively shown to be a mistake for anything dealing with real systems.

    Indeed, even when the program has been correctly compiled to native-equivalent execution, there is zero reason why it should have ambient authority or universal capabilities.

    The problem domain here was "JIT should not introduce further hazards", but even in the unlikely event that it does, those hazards should still be subject to all the other containment that affects every piece of software on the system.

    I also have to work with real engineering teams to get them to do better in practice, and I've never seen a team in a position to do half the things we know we should do.

    And now we really do get back to "is it end capitalism [...]?" as you suggested in your other post.

    I'm still not convinced it's strictly required, but certainly one cannot expect zero-liability corposcum do to proper engineering in any way that resembles what I suggest.

    If they had the kind of liability other engineering fields have for fuckups that cause harm, grievous harm or death to users or third parties?

    Now that might change a bit.

    💬 0🔄 0⭐ 0
  • Nov 9, 2026, 11:47 PM

    @lispi314 @dymaxion @rob Even if that's true on a theoretical level, I'm not very okay with the concept of moving code out from the embedded-language execution environment we control to native code on the embedded-language host on the basis of trusting it's correct/safe. Doing that is inherently removing a layer of defense.

    💬 1🔄 0⭐ 0
  • Nov 10, 2026, 12:00 AM
    @dalias @dymaxion @rob It is, indeed, a tradeoff.

    One that should not be mandatory either.

    Various Lisp Machines were fully capable of using interpreted drivers (with harsh performance degradation) without native compilation.

    It was actually a semi-common user-error to do so and forget to compile them, then wondering why the performance wasn't matching up expectations.
    💬 0🔄 0⭐ 0