cross-posted from: https://discuss.tchncs.de/post/13814482
I just noticed that
ezacan now display total disk space used by directories!I think this is pretty cool. I wanted it for a long time.
There are other ways to get the information of course. But having it integrated with all the other options for listing directories is fab.
ezahas features like--git-awareness,--treedisplay, clickable--hyperlink, filetype--iconsand other display, permissions, dates, ownerships, and other stuff. being able to mash everything together in any arbitrary way which is useful is handy. And of course you can--sort=sizedocs:
--total-size show the size of a directory as the size of all files and directories inside (unix only)It also (optionally) color codes the information. Values measures in kb, mb, and gb are clear. Here is a screenshot to show that:
eza --long -h --total-size --sort=oldest --no-permissions --no-user
Of course it take a little while to load large directories so you will not want to use by default.
Looks like it was first implemented Oct 2023 with some fixes since then. (Changelog). PR #533 - feat: added recursive directory parser with `--total-size` flag by Xemptuous
I just tested this and the reported sizes with
eza -l --total-sizeare wrong for me. I compare it todu --human-readable --apparent-size --all --max-depth 1and with opening properties in my Dolphin filemanager. Some are way off. In exampleduand Dolphin report for a certain projects folder of mine "149M", whileezareports "184M".this looks like one is using the SI 1000-based units, instead of the binary 1024-based. im pretty sure
duhas a--sioption.the
B(for bytes) is omitted, so it each is ambiguous to whether itsMiB(mebibytes -- binary) orMB(megabytes -- SI).i may be wrong on the technicals but u get the jist.
The difference is too large for that. 184 MB is 176 MiB not 149.
No, the difference is way too high to explain it like this, there is no way that 1024 vs 1000 base could explain an increase of approx. "35M" for a "149M" directory. Other folders are much closer like "20K" and "20K" or =or "44M" vs "45M". Also as said Dolphin filemanager reports the same output as
du. I even testedduwith--sioption, which power of 1000 instead 1024 (I'm pretty sureezadoes it correctly with 1024, so this is not necessary option to compare anyway).No, @[email protected] is correct.
I just did a test using
dd- I created 100 files of exactly 1 MiB each (1048576 bytes).dureported the size as "100M" as expected, whereasezareported it as "105M" - which is what you'd get if you divided 104857600 by 1000000 (= 104.8576 or 105M if you round it off).He is wrong, as I explained it multiple times that this is not the issue here. Install
ezaand compare toduand possibly some other application that reports the directory size. The difference in filesize cannot be explained by 1000 vs 1024 base. Do the math if you don't believe me.ezais reporting false directory size for me, unless there is an explanation.And for reference Dolphin the filemanager of KDE Plasma reports
149,1 MiB (156.366.443), which aligns withduwithout using--sioption. Even the one folder "target" is at183Mwitheza(which is the biggest folder in that directory anyway).I was talking about the 1000 vs 1024 issue, do the dd test yourself and it's easy to verify that he was right.
As for the specific descrepancy that you're seeing, lots of things can throw off a file size calculation - symlinks, sparse files, reflinks, compression etc. Since you're the only one with access to your files, you'll need to investigate and come to a conclusion yourself (and file a bug report if necessary).
Could it be this AND block size vs actual used size?
hmm I didn't think to actually test the results. But now that i do, I get same sort of descrepency.
How about this?
that works in a couple test directories with the column
Blocksize.Also it might (??) be ignoring according to your
gitignoreif that is relevant? Or behaving differently wrt symlinks?Seems like the default behavior should be whatever is most expected, standard and obvious. Or else give user a hint.
I find this in the repo, is t relevant?: bug: Inconsistent Size Display in `exa` Command for Large Files (1024 vs. 1000 Conversion) · Issue #519.
don't forget
eza --version. I find it is not updated quickly in every distro. See changelog; it looks like there might have been a relevant update as recently as[0.18.6] - 2024-03-06. Actual my system is only updated to0.17.3now that I check this too.With
--binaryoption I get size of174Miineza. Experimenting with some other options didn't help. If something is ignored (maybe gitignore), then it would be thatduAND Dolphin filemanager would ignore those files, andezawould not. Which its hard to believe for me. I also deleted the .gitignore and .git files/folder to see if it makes any difference and no, it did not.The only thing I can think of is maybe something going on with link files, but no idea how or what to test for here.
well I guess a way to test would be to create a new directory and copy or create some files into it rather than using a working directory where there are unknown complexities. IIRC
ddcan create files according to parameters.Start with a single file in a normal location and see how to get it to output the correct info and complicate things until you can find out where it breaks.
That's what I would do, but maybe a dev would have a more sophisticated method. Might be worth while to read the PR where the feature was introduced.
Also kind of a shot in the dark but do you have an ext4 filesystem? I have been dabbling with btrfs lately and it leads to some strange behaviors. Like some problems with rsync. Ideally this tool would be working properly for all use cases but it's new so perhaps the testing would be helpful. I also noticed that this feature is unix only. I didn't read about why.
Although only 1 of various potential causes, I don't think it is implausible on its face.
duprobably doesn't know aboutgitat all right? If nautilus has a VCS extension installed I doubt it would specifically ignore for the purposes of calculating file size.I have found a lot of these rust alternatives ignore
.gitand other files a little too aggressively for my taste. Bothfd(find), andag(grep) require 1-2 arguments to include dotfiles,git-ignored and other files. There are other defaults that I suppose make lots of sense in certain contexts. Often I can't find something I know is there and eventually it turns out it's being ignored somehow.About the gitignore stuff of Rust tools: Its the opposite for my results, in that eza has bigger size. And the fact that the independent program Dolphin filemanager aligns with the output of the standard du tool (for which I don't have a config file I think) speaks for being the more correct output.
Ok so I found it: Hardlinks
I went down into the directories and compared some outputs until I could circle it down (is it called like that?). Look at the number
2, which means those files are hardlink. Their md5 checksum are identical. So its what I was thinking all along, some kind of linking weirdness (which in itself is not weird at all). Soezais not aware of hardlinks and count them as individual files, which is simply wrong, from perspective of how much space those files occupy. The file exists once on the disk and requires only one time space.EDIT: BTW sorry that my replies turned your news post into a troubleshooting post. :-(
For my part I think all this troublefinding and troublesolving is a great use of a thread. :D Especially if it gets turned into a bug report and eventually PR. I had a quick look in the repo and I don't see anything relevant but it could be hidden where I can't see it. Since you've already gone and found the problem it would be a shame to leave it here where it'll never be found or seen. Hope you will send to them.
I also reproduce the bug by moving an ISO file into a directory then hardlinking it in the same dir. Each file is counted individually and the dir is 2x the size it should be! I can't find any way to fix it.
The best I can come up with is to show the links but it only works when you look at the linked file itself:
If you look further up the filetree you could never guess. (I will say again that my distro is not up to date with the latest release and it is possible this is already fixed.)
This should be an option. In
dua-cli, another one of the other rust terminal tools I love, you can choose:BTW I actually did a bug report. :-) -> https://github.com/eza-community/eza/issues/923
So nothing wasted. Without your post I would not be curious to test this and who knows, maybe it gets fixed or an option to handle it.
Nice! I'm sure they will appreciate your thorough report.
I wonder if they also plan to make an option about crossing filesystem boundaries. I have seen it commonly in this sort of use case.
Maybe all this complexity this is the reason why total dir size has not previously been integrated into this kind of tool. (Notable exception:
lsdif you are interested.) I really hope the development persists though because being able to easily manipulate so many different kinds of information about the filesystem without spending hours/days/weeks/years creating bespoke shell scripts is super handy.I used
lsdbefore switching toexa. BTW I was the one who suggested to integrated hyperlink option tolsd. :-) Not saying it wouldn't be added otherwise, but at least it sped up the process back then.^^On the topic of filesystem boundaries, this is something I always have in mind. Hardlinks in example cannot be on two different drives, by its nature. It's an option I use with my own
duscript often:-x. It helps to keep things in check and simplifies some things, as I have a complex of multiple drives that are mounted into my home from various places, plus the symbolic links from Steam and so on. Avoiding other filesystems is part of my "workflow", if you so will and would like to see such an option as well.I just noticed
exahas an option-Hor--linksto list each file's number of hard links. So at least it is aware of hardlinks and could calculate the total size correctly.