fix: ipfs file ls: list node size for directories

[schomatis]

List the size of the underlying node for directories to honor the unix API.

If a test is needed: only the JSON encoding outputs the file size and I'm not sure how to process that with the sharness framework.

Fixes #4580.

[schomatis]

I was looking at the test file t0250-files-api.sh when I should have checked t0200-unixfs-ls.sh, I'll fix the second test.

[ghost]

It would be good to take directory sharding into account here too -- a directory is a unixfspb.Data_HAMTShard in these cases. We'd probably want the combined size of all shards.

[Stebalien]

It would be good to take directory sharding into account here too -- a directory is a unixfspb.Data_HAMTShard in these cases. We'd probably want the combined size of all shards.

Really, directories should list their sizes. Recursively reading the size of the hamt is going to be really annoying.

[ghost]

Not the size of the contents, just the raw size of the (sharded) directory object. Or do you mean that's something to precompute and stick into the object?

[Stebalien]

Not the size of the contents, just the raw size of the (sharded) directory object.

That still involves traversing the sharded directory. Not that bad, just annoying.

Or do you mean that's something to precompute and stick into the object?

Yeah. Ideally, everything would have a filesize.

[whyrusleeping]

Yeah. Ideally, everything would have a filesize.

Add that to the ipld unixfs requirements listing

[Kubuxu]

SGWM, we can't simply resolve the issue of sharded directories without defeating most of the benefit of sharding.

[Stebalien]

Add that to the ipld unixfs requirements listing

ipld/legacy-unixfs-v2#7

[schomatis]

I got lost after the sharded directory comments, should I modify something of this PR?

[Stebalien]

@schomatis so, we've been discussing this:

• Record the FileSize in directories as well ipld/legacy-unixfs-v2#7
• "Size" fields to include ipld/legacy-unixfs-v2#9

And I'm now less convinced we should actually include the size. Normally, filesize tells you how big a file would be when downloaded. However, the directory size (in bytes) won't really tell you anything useful. Thoughts? I don't really have any strong opinions. My initial opinion was "follow unix" but I'm now less convinced.

[kevina]

follow unix

Is not a good guide on this. Unix uses the physical directory size (that is the list of inodes), not the size of the contents, or even the number of entries. In some ways this makes sense and it is also easy to implement, but it isn't very useful and it doesn't really make sense for us.

[schomatis]

As suggested in the previous comments, this PR doesn't seems to add much value, closing.