Hacker News Reader

Comment by titzer | Hacker News Reader

Just give us a web filesystem already. Stop with these shitty infantilized GUIs with no hierarchical organization and an overreliance on search to find things. I have literally hundreds of thousands to millions of files on my local machines and have no trouble remembering where I put things. Drive can't seem to scale past a few dozen before it's confusing, useless, and apparently lossy. Remember when web servers would serve up the contents of a directory as a list with hyperlinks on the file names? That actually worked ffs. I cannot understand how we could be moving backwards so quickly.

View on HN

There is a transition from a file-centric to an app-centric model, that I mostly disapprove of, but that's what is happening. Before, you had files stored in a filesystem, you picked the file and opened it with the app. On Windows, you double-click on the file, on a command line, you type "app filename". Now the trend is to first open the app and then use the app to fetch your data, that data is entirely managed by the app and may be stored in an opaque area of a filesystem, some server on the internet, etc...

I disapprove of the "new" way (that is not so new, but it is the current trend) because it favors lock-in and general loss of control. But there are also some advantages such as being able to do versioning, syncing, collaborative work, searching and general database-like operations when your filesystem doesn't support it. A lot of popular open source "not evil" software use that model, for example, IIRC, Firefox has always stored bookmarks in an opaque sqlite database while Internet Explorer stored them as files in a folder, and I don't think I need to tell you which one is considered the most evil.

On the flip side, Steve Jobs tried to make this happen for the iPad and Apple eventually gave up and built the Files app because they were trying to make it into a system you could do actual work. "All data belongs to a single app" was at odds with most work tasks other than checking your email, and sending copies of data back and forth between apps with share sheets is alright if you only have to do it once, but it doesn't really scale.

The people pushing to have all desktop programs isolated from each other by default in the name of "security" are making the same mistake.

sqlite is opaque? I've got this script to list my open tabs, and a few more scripts for other Firefox actions.

  $ grep FIREFOX ~/.bashrc
  export FIREFOX_DIR=$HOME/.mozilla/firefox/q3m93qpe.default-release/

  $ cat ~/.local/bin/ff-tabs
  #!/bin/bash
  lz4jsoncat $FIREFOX_DIR/sessionstore-backups/recovery.jsonlz4 | jq -r '.windows[].tabs[].entries[-1] | .title, .url, ""'

Ha! Of course I would choose an example based on JSON, not sqlite, but the point still stands. Sqlite is one of my favorite file formats, it presents the data in an interface far easier to manipulate for a one-off CLI command than writing another Python script with open() and dozens of re.sub().

JSON is also not very opaque. What's opaque is iOS with its locked down storage per app as you mentioned.

Some OS builders have been trying to push this model on the desktop but have not been very successful due to legacy. It only seems to be done with mobile apps running on the desktop (Windows with Android apps and macOS with iOS)

I also lament this model, I think we should always have full access to our devices. But there's a strong commercial push for lock-in and subscription models.

How do you use it?

Actually Google did made a version of Google Drive Filestream for Linux, and made it available it Google Colab, just that they never released it publicly for unknown reason. I have grabbed the binary of it. https://github.com/michaellee8/gdrivefs/tree/master/assets

There are quite a few clients for Linux, I use Syncdocs https://www.syncdocs.com/ as it has built in encryption.

Non I've seen do a proper job of mounting as a drive, NFS or SMB would be nice.

I’d take a file system that can be indexed by my local operating system (MacOS).

The last MacOS update broke Spotlight indexing for my Google Drive folders. Really annoying not being able to use Spotlight (and by extension Alfred), so the only logical thing to do was move to iCloud.

In the process of moving off of Google Drive, I also decided to move my email to Fastmail and just ditch GSuite (or whatever they are calling it today) forever. I’ll sleep better at night knowing that Fastmail knows their place and (hopefully) won’t try and force a chat function into my web email client as a means to build engagement for a product I don’t care about.

Spotlight no longer working with google drive — for months now — has been such a pain in my ass that it’s finally convinced me to shift my entire company off Google Workspace. At least with Microsoft I can get someone on the phone.

I agree with you 100%. I just want something that stores files and works with SFTP and NFS out of the box. I get so annoyed every time I have to use S3 manually.

However, there is a post on this site that has become infamous in which a commenter questions the need for Dropbox when stuff like SFTP already exists.

I’ve wondered how practical using Azure Files with SMB is. It also supports NFS.

https://docs.microsoft.com/en-us/azure/storage/files/files-s...

SMB is absolutely terrible over latent connections.

I don't understand how SMB could work unless there was geographical replication to nodes closer to you, with some kind of load balancing to change the endpoint you're using to be closer to you.

Yup, hence the wondering. They keep improving the protocol and I am hopeful that one day I might try it and be pleasantly surprised.

https://docs.microsoft.com/en-us/windows-server/storage/file...

I tried it a little less than 2 months ago across a suite of services, Azure, self hosted, AWS EC2 instances, and all of these same options again with WireGuard over the top (well, with Tailscale in this instance).

Tragically it was still trash.

Thank you for your sacrifice

You want https://rclone.org/commands/rclone_mount/. Yes it should really just be a straight up product, but for the moment you can hack it together fairly effectively.

They should take inspiration from nginx and Apache directory listings. Doesn't take 4s to load and uses actual links so you can open in new window or copy the link.

> Just give us a web filesystem already.

WebDAV is kinda that. I subscribe to a (super cheap) hosted version from Hetzner and I'm happy with it even though the server is on a different continent. As mentioned somewhere in another comment, I also use rclone mount to use it, but there are other options as well (the client built into GNOME is reasonably good, the client built into Windows 10 is terrible).

Agreed. Every time I go to save a file I have to stop thinking like a techie and put my 'think like Google' hat on. My wife is the average, basic user. Windows and Linux filesystems she navigates with no issue. Google Drive? Confuses the heck out of both of us and I've been a techie my whole life.

You’re still free to use a webserver with directory listing turned on..

I agree with all of your complaints, but will say Gnome nautilus integration is working pretty well these days, and allows me to interact with it more like a filesystem than a shitty webui.

... until you try to open the files with a non-gnome application. Mounting a filesystem was supposed to be a solved problem.

Examples? Maybe I just don't have enough non-Gnome applications tbh.

I just tested with Slack, VSCode, and GIMP, and all three allowed me to browse and open from there.

From the top of my head, QOwnNotes (hierachical markdown notes) and KeePassXC (password manager). I remember those two because I had to use DropBox, they can't see the Gnome Google Drive path as a regular path.

I ended up buying InSync because it was 50% off in easter, and its Google Drive/Dropbox paths works perfectly even in the terminal. Unfortunately it lacks Google Drive streaming.

Consider paying for S3.

S3 is not a filesystem replacement unfortunately.

Not a filesystem, but rclone works pretty well

rclone also works with Google drive, if that's all you're after.

Why not? It's a cloud service that stores files, can pretend to have directories, and for which there exist multiple utilities to mount the thing as a local file system. What's the shortcoming?

S3 is object storage not block storage.

https://www.digitalocean.com/community/tutorials/object-stor...

Object storage isn’t random access (the ability to read and write any part of a file) the way a disk drive is. This causes complications for software that expects to be able to randomly access a file like they would on a disk.

WebDAV is also an object storage protocol but it can do random access. Well, almost.

Partial reads are supported everywhere out of box just by using the Content-Range HTTP header.

Partial writes aren't standard, but some implementations e.g. SabreDAV can do those using PATCH with X-Update-Range header (and e.g. rclone supports this).

WebDAV has its warts and limitations but it's a very underappreciated protocol.

As someone else pointed out, it's object storage not block storage meaning things like partial reads and partial writes don't work. That may not be a problem for you, but it _is_ a problem for me (particularly when dealing with remote data).

> for which there exist multiple utilities to mount the thing as a local file system

Which all have their own sets of intricacies. The most obvious one being listing a directory is either impossible, costly, or as slow as the heat death of the universe. Other (far riskier) ones (from s3fs-fuse, which I have the most experience with[0] ) are the lack of guarantees about atomicity and concurrent access.

[0] https://github.com/s3fs-fuse/s3fs-fuse

Why pay when https://min.io/ exists?

I'd rather that didn't happen. The current web ecosystem of adtech companies and data brokers makes the web one of the last places where I'd want to entrust my files with. And before anyone mentions permissions, I don't see that putting a stop to data harvesting. Because if that even worked, social media would not have been a thing. It has been shown that people just "willfully" give up their data without giving as much thought to the consequences. So if this web filesystem plan goes through, entire services would be built around users giving companies access to their files, and those would go straight to the ad networks. Less friction in the browser means that we'd see an exponential increase in those type of services. So the current friction is better for user privacy.

Reply on news.ycombinator.com