> If anything, knowing whether the app is installed or not is kinda important? If you open a file shared with you in the browser, the option to "Open in Desktop" versus "Install Desktop App" actually works correctly?

This is not an approach any other app on any platform has historically used, and it doesn't seem sustainable if every app you install has to modify your hosts file to use a hack like this to detect whether it should handle files or not.

If you want the browser to be able to give the OS a file handler and have the OS present an option to install the app if it's not installed, that should be handled at the platform level, not on the website using a hack like this.

Why can a file not simply be downloaded with a page displayed showing a link to install the app and also instructions to open the file, trusting the user will know if they already have it installed? At best, you're talking about a very small UX optimization. Emphasis on the "kinda" in "kinda important."

▲

naniwaduni 3 hours ago | parent | next [-]

> This is not an approach any other app on any platform has historically used, and it doesn't seem sustainable if every app you install has to modify your hosts file to use a hack like this to detect whether it should handle files or not.

How many apps are you installing that it becomes "unsustainable"? Host file entries are extremely cheap, and it's not like the app needs more than one. Of all the arguments against this, sustainability is a comically weak one. If anything, it's using less contested resources than the "hitting random ports on localhost" approach...

▲

jacobgkau 3 hours ago | parent [-]

The "sustainable" comment wasn't about the hosts file ballooning to the point of causing performance problems. It was more about the engineering effort required for every program ever (or at least every commercial program that might want this sort of analytic) to have to parse and edit a text file on both installation and removal, without messing that important text file up.

Do you really not see scripted editing of shared system-wide text files as a step back compared to the general containerization that app development has moved towards? This sort of approach would be explicitly incompatible with sandboxes. Adobe can only get away with it because they're already very entrenched with their own app store on their users' machines.

	▲	gjsman-1000 11 minutes ago \| parent [-]
		> Do you really not see scripted editing of shared system-wide text files as a step back compared to the general containerization that app development has moved towards? Sir, this is Windows. This is not Android, this is not iOS, this is not macOS. Wait until you learn about the registry.

▲

gjsman-1000 3 hours ago | parent | prev [-]

Actually it's completely sustainable. DNS was invented a decade after hosts files. The idea of your host file being almost completely empty is a modern aberration from the days it used to be thousands of lines long.

Do I wish there was a better mechanism? Sure. Would HN ever agree on a OS-level app-detection API for the browser? Never.

> Why can a file not simply be downloaded with a page displayed showing a link to install the app and also instructions to open the file, trusting the user will know if they already have it installed? At best, you're talking about a very small UX optimization. Emphasis on the "kinda" in "kinda important."

A small UX decision, adding up to tens of millions of times per day, affecting 99.9% of people who don't give a darn - versus a matter of slight software engineering principles of "we just don't do it that way." Easiest decision ever.

▲

jacobgkau 3 hours ago | parent [-]

> Would HN ever agree on a OS-level app-detection API for the browser? Never.

There already is one. It just asks the user whether it's okay before it tells the website, as you acknowledged: https://news.ycombinator.com/item?id=47664546

What you're arguing for is not good UX. It's lack of user privacy & control. You just think you're being hip or whatever for being blasé about it.

	▲	zamadatix an hour ago \| parent [-]
		> There *already( is one. It just asks the user whether it's okay before it tells the website The current implementation defines a way to launch (w/ the user's approval) but it lacks any signaling of success or failure of the request. Without such feedback, it falls short of being a detection API.