4 ms·
We have an ecosystem where running nontrusted software can be relatively safe... it's just on the web. Legacy desktop OSes don't have the permissions models th
by solardev 4y ago
We have an ecosystem where running nontrusted software can be relatively safe... it's just on the web.
Legacy desktop OSes don't have the permissions models that make this easily achievable outside the browser. Even the mobile platforms are only slowly evolving (I want this app to access ONLY this folder or that picture, not all my user files).
Frankly I would never trust users to be able to differentiate between sane and unnecessary permissions. If you make everything super granular or repetitive it's just going to lead to banner blindness (yes, damn it Google, I want this hiking app to know precisely where I am... I told you the last seven times you asked). Or on iOS, having to enter my longass Apple password just to download a free app whenever the fingerprint scanner doesn't work (which is usually). Even without security concerns, native apps are a pain in the ass.
The web is nice in that most of the sandboxing is invisible unless you need special sensor permissions (location, mic, camera, etc.) and so the user never gets bugged about it. Or has to worry about platform idiosyncrasies, disk space, versions, etc. In many ways the web is a superior app delivery platform. Why bother installing "safe" software when you can just run it in a sandbox without any installation or prompts at all? What Java and .NET tried to do back in the day now just happens, sight unseen, in browser windows.
- zzo38computer 4y agoUnfortunately there are problems with running it on the web (and many problems with its design, and problems with implementations), as well as problems with native permission models of some systems, too. For example, it is not very good working for: command-line with pipes, non-Unicode character encodings, non-USB devices, non-HTTP(S) protocols, working local files without internet connection, interaction with other local programs in a good way, etc. Permissions need not be asking every time (if user will configure it to always allow or deny or other settings), and need not only be "allow"/"deny"; for example, if it requests the camera access then the user might enter a command to use instead, which might access the camera (possibly with filters such as fault simulation), or do something else such as returns a still picture (which you can use if you do not have a camera, for example), and this can be "wired" by the user configuration if not wanted to enter every time. Disk space is going to matter for any program that stores files, although one which is designed well will allow the end user to specify a disk quota for this program if desired, and will also allow specifying default disk quotas in the manifest, in order that you can use it without needing to know about these things and manually set them up, too. I had tried to make the specification of "VM3" which is meant to, among other things, solve these problems. A program can be install or just run, and all I/O must use extensions, which makes both extensible capabilities and highly user configurable capabilites. The same is true for program entries (e.g. command-line, GUI, etc). There are also some other minor things I had done differently due to I think being better than what some other designs are working.
- solardev 4y agoI think that kind of computing will become more and more the realm of specialists (devs, computer scientists, whatever) doing their work, with specialized tools and permissions models. For 99.9% of regular people, there is no reason to expose them to so much unnecessary complexity. Just like with supply pipelines, most people don't need to care or know about information pipelines... they just want to consume the thing they're there to get, whether it's refined petroleum or some unit of information. All tools are built up from layers of primitives, like a car is made up of components and nuts and bolts and such, but drivers don't need to know or care how its ECU or ICE works. I think the web / mass-market computing is similar... it just doesn't (and I'd argue shouldn't) matter to most people. It's the difference between engineering for other engineers and designing for end-users, two related but ultimately separate concerns.