Hacker News new | ask | show | jobs
by cxr 681 days ago
You could literally run the Golang compiler in your browser if you wanted to (emphasis on in your browser—not deferring to a server to do the heavy lifting) without any of that stuff.

A lot of the stuff that people think they need Chrome's proposed filesystem APIs for are not actually things they need.

Every mainstream browser accepts directories for both type=file inputs and for drag-and-drop, and that's been the case for years. So that's a non-issue. That aside, the next thing people pretend they need it for is so their app foo.html won't fail when it tries to read and write to or from ~/.foo and /etc/foo and anywhere else the author chooses, whenever they want. But apps can and should be rewritten not to expect that kind of ambient capability. The fact that in-browsernapps can't do that stuff is a good thing. There's usually nothing stopping the author from refactoring their program to either output a single binary file or .txt or whatever the main thing is that the program is supposed to output, or if the purpose is to write out multiple assets, then output a composite file format like ZIP or tar. I mean owing to its history and the context when it was created, the ZIP format already supports archives spanning multiple volumes/files. And that gets you automatic support for file versioning for free! It's the exact same version control strategy that Git uses and proved is viable: full-content snapshots—the difference being if you're doing ZIP, the opportunity to also use DEFLATE is natural and right there.

> Only thing missing really is a CLI interface.

If you want people to be able to use it from the command-line, write your app as a polyglot file foo.htm so that anyone can double click to open in their browser if they want to or run it as `nodejs foo.htm` since NodeJS doesn't care about file names but browsers do (for any files loaded from disk). If you want to "pass through arguments as [query string] params", then you're opening it in the browser and not really providing a command-line interface so there's very little difference between having them type the arguments out in their shell before pressing Enter or typing them into the app when it's open in the browser.

Again this is one of those things where nothing's stopping anyone from doing it right now today. They're just not doing it. As the Book Overflow guys said in a recent episode (the review of Wolfram's book on ChatGPT)—all the pieces are there. Or to borrow a turn of phrase from Raskin that gets down to the truth of the mess we're in[1], "none of this is necessary; it is merely customary"[2].

1. <https://www.youtube.com/watch?v=lKXe3HUG2l4>

2. <https://queue.acm.org/detail.cfm?id=864034>

1 comments

You can do a lot of things by dragging folders onto file inputs, but it is somewhat rubbish for user experience with the friction of getting files in and out of your app. Obviously it depends on what you want to do, but it seems a limiting factor. Allowing HTML application permanent permissions to file system without user input makes apps a lot more useful and portable.

Recently. I wanted to organize around a thousand short videos. I looked into software and the solutions were really integrated, media library etc. MacOS has tagging in the filesystem, but that ties me to an operating system, and didn't give me the explore related function. I wanted something that could let me browser a folder with added metadata, explore related clips etc.

After looking at all the options I could find, I realized it was easier to just build myself a little Node.JS app using sqlite for metadata, it worked great. I plugged in local ollama and whisper for transcriptions and automatic tagging, which worked suprisingly well. It's annoying needing to run a server for such a simple app, packaging it is just so heavy ... and I thought, it's really not that far off being something I could do in a single HTML file, it was pretty much just the filesystem access - then my app/content is just a folder that does what I want to on any platform without installations or configuration.

Thanks for linking the Joe Armstrong lecture, I've watched half and its wonderful.

> it is somewhat rubbish for user experience with the friction of getting files in and out of your app

It's not worse, generally, than the average experience of using a NodeJS program that won't run because the dependencies didn't install correctly or it's the wrong version of the runtime, or... (cf Armstrong; or, to borrow a saying by Ousterhout (with a small tweak): the greatest improvement to a program of all is when it goes from not-working to working).

The two worst rough edges for file saving for browser-based apps are:

- if you consider, say, a paint program that you're using to work on an image, it can't implement the ordinary save behavior where some keyboard shortcut or menu items saves (overwrites) the current file you're working on; to do the same in the browser requires the user to do the equivalent of a Save As every time (instead of a Save)—i.e. complete with a prompt to double check that the user really wants to overwrite the previous file

- some browsers have changed the default download behavior so it automatically saves to a downloads directory without asking the user and allowing them to change the path/filename with a conventional filepicker; this setting has to be changed by the user themselves to get the old, circa 2005 browser download behavior

These are both issues that can be fixed and don't require something as expansive as Chrome's filesystem API proposal.

re alternative options to your NodeJS-powered video tagging application server: check out <https://redbean.dev/>

I agree, file access via upload/download concept is friction on the user experience. I'd love to see serverless offline HTML applications with permission based file system access, because you can do so much very quickly in the browser with access to GPU, bluetooth, USB even, but I'm not holding my breath because of the security implications

Redbean looks good, I didn't know about the APE format, but unfortunately doesn't solve the file system access problem without presumably rewriting a backend in Lua. Docker, reluctantly, is better than nothing.

I suspect Justine will eventually do what the Cloudflare Workers team did, which is after having originally only supported Lua, to start supporting JS (albeit via quickjs in the redbean case, rather than V8) and then see wild uptake.