Comment by koito17 | Hacker News Reader

Comment by koito17 | Hacker News Reader

> it’s quite an inconvenience to run your own IPFS node. But even if you do run your own node, the fact you access a website doesn’t mean you pin it. Not at all.

This has always been my major UX gripe with IPFS. The fact that `ipfs add` in the command line does little but generate a hash and you need to actually pin things in order to "seed" them, so to speak. So "adding a file to IPFS", in the sense of "adding a file to the network", requires the user to know that (1) the "add" in `ipfs add` does not add a file to the network, and (2) you must pin everything you want to replicate manually. I remember as recently as 2021 having to manually pin each file in a directory since pinning the directory does not recursively pin files. Doing this by hand for small folders is okay, but large folders? Not so much.

More importantly, the BitTorrent duplication problems that IPFS has solved are also solved in BitTorrent v2, and BitTorrent v2 IMO solves these problems in a much better way (you can create "hybrid torrents" which allows a great deal of backwards compatibility with existing torrent software).

This isn't a UX issue, but another thing that makes it hard for me to recommend IPFS to friends is the increasing association with "Web3" and cryptocurrency. I don't have any strong opinions on "Web3", but to many people, it's an instant deal-breaker.

‘ipfs add’ pins the file by default, not sure if that’s recent behaviour though.

It's not recent, been like that since at least 2015, if not earlier.

IPFS provides nice stable links to media, and there are HTTP->IPFS gateways if needed. That seems useful for embedding content on multiple apps/sites. Yeah it happens to fit NFTs particularly well, then again we all know what BitTorrent is known for. And yes I agree IPFS has some UI problems.

Would BitTorrent also be suitable for hosting embeddable content? I haven't seen that yet. A magnet URL is mainly a file hash and doesn't seem to encode a particular peer server, kinda like IPFS. But every time I've torrented Ubuntu, it's taken half a minute just to find the peers.

> IPFS provides nice stable links to media

Anyone who has tried to torrent an old movie or lesser-known television show knows this is simply not true.

I mean it's not like HTTP where all URLs are tied to a particular webserver and can even be changed on that server. If someone different starts seeding, you'll get the same data again at the same URL, with built-in checksumming.

That is something I wanted to know, does IPFS guarantee that same two files have same two IPFS URLs / hash links?

Otherwise, someone sharing same data again, because it will be in different IPFS folder won't be actually discoverable as same data.

Yes, a file's hash is only based on its contents. The way I understand it, a file doesn't really live in a directory, it's more like a directory (which is a kind of file itself) references files. So the same file can be in two directories, yet it'll have the same URL/hash. And if you "add" files to a directory, you're really uploading a separate copy of the dir that'll have a different hash.

I checked myself on this, but someone else might want to check me cause I'm not an expert.

This is generally true, though it’s possible to encode the same data into a slightly different shaped DAG to optimise for eg video streaming performance afaiu (balanced vs imbalanced). UnixFS vs raw bytes may also be different but I’m not 100%

From the fs's point of view, these are different file contents. But yeah, there's nothing stopping you from pinning something different that looks the same to a person.

Once decoded they would be the same file contents - imagine one DAG where the depth is log(n) and it’s a perfectly balance tree, and another where the depth of the left-hand branch is 1, right hand branch contains another subtle with left-hand size 1 etc etc etc.

The leaves are the same in both cases, so the file contents are the same, though the latter is quicker to stream (though not to verify) and the CIDs will be different

Yes, IPFS hashes individual "blocks" (pieces) of files. If two files have the same content, they will share block hashes.

Caveat: The other comments mention the file's contents being the only dependency on the hash, but the algo used to hash would also need to be the same. If the hash algo changes in two cases, the same content would have a different hash in those two cases.

In this case, would pinning the file make it accessible from either hash? I'd expect it to, but idk, I've only ever seen sha256 hashes on IPFS.

Kinda. Shooting from the hip based on fuzzy gatherings from IPFS usage here, but as I understand it: The leaf-level data blocks will be shared between the Merkle trees, but at least the tip (the object a given hash actually refers to) and maybe some of the other structural information will be different.

Basically it depends on specific settings that can be changed in the client as to how the individual block pieces are encoded and therefore what the resulting hash ends up being. So no there's no inherent guarantee but you may get lucky with some copies of the same file.

> Would BitTorrent also be suitable for hosting embeddable content?

Same as IPFS: gateways can exist. It's not specific to Bittorrent, or IPFS.

> A magnet URL is mainly a file hash and doesn't seem to encode a particular peer server, kinda like IPFS.

Magnet links can include a HTTP server that also hosts the content

I'm sure a BitTorrent gateway can exist, but I'm wondering why it doesn't seem to "be a thing." I've never seen one used, nor do I see an obvious public one when searching. Whereas IPFS gateways are so mainstream that even Cloudflare runs a public one.

It's because of the kind of content that is shared. BitTorrent serves a lot of content you are not allowed to redistribute, so having an open gateway immediately puts you at risk of aiding the distribution of content. But it does work, someone even made something native to browsers so browsers themselves can share content: https://webtorrent.io/. There are even fuse "gateways" to make it native to your computer and pretend the files exist locally: https://github.com/search?q=bittorrent+fuse&type=repositorie...

IPFS doesn't seem to be used for that kind of content much, it seems to be targeted more towards web-native content (html pages, images, that kind of stuff). It's probably safer for Cloudflare to run this.

> I remember as recently as 2021 having to manually pin each file in a directory since pinning the directory does not recursively pin files. Doing this by hand for small folders is okay, but large folders? Not so much.

Can‘t you just use a glob?

Reply on news.ycombinator.com