Hacker News Reader

Comment by bhauer | Hacker News Reader

Amusing, but the standard Unix file system is nothing to be too proud of. I'd very much like to see a file system layout re-imagined for the 2010s.

View on HN

Binaries? Libraries? Let's stick 'em in `/usr`! Oh, but where do user files go? I know, `/home`! What about all this other stuff? Eh, `/etc` will do. Unless it won't, in which case `/var` is the obvious solution. Or maybe even `/opt`, if we're feeling especially hip today.

Historical heritage unfortuantely. /sbin and /usr/sbin are other examples...

http://lists.busybox.net/pipermail/busybox/2010-December/074...

https://news.ycombinator.com/item?id=3519952

usr doesnt mean user, it meant Unix System Resources, and a user of the system usually puts his stuff at home, what would you like it to be otherwise /users ? That would be describing the users of the system and not where their stuff is, and users are described in /etc/passwd together with their passwords, sort of.

/etc isnt called /conf or configuration because it contains examples, scripts and other small stuff as well, which is etcetera.

/var is for variable data like logs and pid files, not meant to be changed by the user directly

/opt sucks and is for losers who dont know about /usr/local or how to install stuff in ~

The Linux way makes much more sense than Windows or MacOSX way. binaries are in bin libraries are in lib, whats not to like?

> /var is for variable data like logs and pid files, not meant to be changed by the user directly

One would certainly never dream of keeping entire web sites in there!

Well thats just wrong, thats what /srv/http is for, to serve http, but historically websites were served from your home, as then they where known as home (personal) sites, and they're still known as "home sites" in many non-english languages.

Now when you have web applications they can be in /srv as well and the database is usually wrongly put in /var (looking at you mysql) but what you gonna do when you need a system user, just like a users stuff is supposed to be in home so put your database there /home/postgresql/its_datas here, as anyway you need to run the db as its own user.

Makes sense no?

My point was about Apache's default document root.

The default for apache is/was /var/www/html.

usr originally contained user files (http://cm.bell-labs.com/cm/cs/who/dmr/notes.html). I'm not really sure why or when that changed, but supposedly "Unix System Resources" is a backronym.

> /opt sucks and is for losers who dont know about /usr/local or how to install stuff in ~

I disagree. /opt is for systemwide installations of monolithic software by dumb companies that can't fit their software into the standard layout. In the Linux FHS /usr/local/ is supposed to be for non distribution installed software that nonetheless uses the standard layout with /usr/local/{lib,bin,doc} etc. In /opt you usually have something like /opt/somesoftware/ and some non-standard hierarchy underneath.

> I disagree. /opt is for systemwide installations of monolithic software by dumb companies that can't fit their software into the standard layout.

You are actually fully agreeing with what was said !

+1 for /opt. We bundled our runtimes and data, and our stuff still works 10 years later (although, yeah, security vulnerabilities, although technically none of it was remote).

Um, good luck with your packages where you relied on the OS. Let's not talk about stdlibc++.

Hint: if you really want your stuff to work for a long time, depend on the kernel and nothing else.

/opt is really where: /opt/importantstuff goes. The stuff where they bundled the libc because Debian changes too fast. Freaky stuff. Telco stuff...

/etc stands for Editable Text Configuration

> usr doesnt mean user, it meant Unix System Resources

It did mean "user" originally. Home directories were usually put there.

However, the root filesystem was typically very small. So very early on in UNIXs evolution things that were large and not needed during system boot ended up finding homes in /usr, simply because it was where the bulk of the disk space was. So you got /usr/bin, /usr/dict, ...

Eventually /usr got so full of non-user content, UNIX-based OSes started putting home directories somewhere else entirely (i.e. /home on linux for example) Because the user to home directory mapping was just held in /etc/passwd, it was much easier to move the users than everything else. Later, the "Unix System Resources" backronym was coined.

So mintplant is right: /usr is very much a historic wart on the UNIX filesystem layout.

etc isnt called /conf or configuration because it contains examples, scripts and other small stuff as well, which is etcetera.

Doesn't it stand for 'Extended ToolChest' ?

Probably a backronym like unix system resources.

It could be called misc as well in my opinion.

Ummm... the Linux way is better than the Mac OSX way?

They both use UNIX architecture...

Check out GoboLinux's file system structure:

http://www.gobolinux.org/index.php?page=at_a_glance

I like it! I suppose I am a tad disappointed that it's a distribution in its own right. But they deserve credit for taking a decent first step.

Ick, capitalization in a case-sensititive file-system? Of course, as far as I'm concerned case-sensitivity is generally a misfeature in anything user-facing.

If I remember correctly, their scripts automatically convert lower case letters to uppercase when you are trying to autocomplete typing with TAB. It only does it for the gobolinux-specific uppercase directories

Agreed! I feel retaining case-sensitivity in file systems in 2013 is basically trolling users. I can think of no practical upside; only the opportunity for annoying jokes and tricks.

But what of users (as they often do) descriptively naming files?

I've seen plenty of "Accounts for Proctors' Attention.xlsx" style files. Case insensitivity would lose the case there. :/

There's a difference between case sensitivity and case preservation. FAT32 and HFS+ are examples of case-insensitive, case-preserving file systems.

That there is, but I'm inclined to suggest this behaviour to be more confusing than simply maintaining name identically to input. :/

Not least this can sometimes result in some odd corner cases.

shrug

It seems like the ideal case to me; generous on input, strict on output. In most cases, users won't even know that the insensitivity is there, in which case it will almost never affect them (since users very rarely want two files with the same name, differentiated by case). For users who do know about it, it's not confusing.

What corner cases does it lead to?

Other languages have capitalization rules that are far, far more complicated than English, involving additional characters that either denote capitalization or a semantically different word depending on context.

Personally, I'd aggressively pursue case-insensitivity and treat the problem of "oops the user can't create two files that have nearly the same name" as a much smaller bug than "oops the user can create two files have semantically the exact same name".

Retain the case, but don't use it when testing for uniqueness. That way, I can type "accounts for proctors' attention.xlsx" and the filesystem will know what I mean. But when I list the contents of a directory, it would show up with its original casing.

Case insensitivity is locale-dependent, so you have to either pick a single locale and confuse users with other locales, or deal with the possibility that you could have two files in your filesystem that have the same name in your locale.

It's not as big of a deal when you're using ZSH autocomplete... but I agree, I like my filesystem lowercase and devoid of spaces.

Even that has some warts: For example, Why not "MountS", since other folders are plural: FileS, ProgramS, UserS

Perfect is the enemy of good enough.

I am agreed. Although the last time I heard about a major attempt for a modern file system was in Longhorn and all we know how that ended.

I guess there are still some fears to try it again, considering that current file systems in mainstream OS' are "good enough".

Indeed. I think people worked out that metadata indexing over the top is probably better.

I found not just the naming aspect funny, but the whole munging of 32/64 bit files good, too. That's beyond naming atrocities.

Agreed.

To me, it's even more funny when you realize "WOW" actually means "Windows on Windows." Say what??

Makes sense to me, then again I know NT inside out...

It's NT's syscall translation layer, CPU mode switching backend and entry point front end. It was originally used to run 16-bit apps on 32-bit platforms but is now WOW64 so is to run 32-bit apps on 64-bit machines.

Basically it's a very light weight virtualisation layer.

And it works well. My circa 1996 RPN calculator I wrote in VC++ works perfectly on 2013 x64 machine.

Don't get me wrong. I know it makes sense once you know what it is. And the result in terms of compatibility and stability is impressive.

Still, when I first heard the phrase "Windows on Windows" I had a chuckle.

It reminds me that I enjoyed the way x86 apps ran on my old Alpha NT machine. If I recall correctly, the process list included "fx32" beside each.

WOW doesn't stand for Windows On Windows. WOW64 stands for Windows On Windows 64 [bit].

Oh, that I didn't fully appreciate. That actually makes it more reasonable. (Not being sarcastic!)

What are you trying to do? Make me feel bad for having a good laugh at the name when I first looked it up? :)

>Still, when I first heard the phrase "Windows on Windows" I had a chuckle.

The Windows 2000 loading screen had the tagline "Built on Windows NT Technology." They left it there through multiple SP rollouts IIRC.

I always assumed that was to placate business customers who might have (justifiably) worried that Windows 2000 was related to Windows 98, rather than descending from NT 4.

NT stands for "New Technology".

For everyone worried about Win98 relation simple "Built on Windows NT" would work.

That's really what it is. 32-bit Windows applications on 64-bit Windows are running in what is essentially an emulation layer. It's a fairly thin layer now, but remember that the original 64-bit Windows ran on Itanium. The original Itanium had an embedded (slow) x86 chip for compatibility. I'm not sure if WoW ever used it, but certainly the WoW for Itanium 2 didn't, since it was dropped for that processor.

A Linux equivalent acronym would be childish.

It's an API compatibility layer to allow 16-bit Windows programs to run on 32-bit Windows -- or, more recently, 32-bit Windows programs on 64-bit Windows.

Recently Microsoft went full Keanu with Windows on ARM or WOA -- but eventually they changed that to Windows RT (which might be actually worse).

Actually it stood the test of time quite well. System directory names are lowercase, short and mostly-standard. Granted, there is the /bin/, /sbin/, /usr/bin/, /usr/local/bin/... stuff, but at least /home/ directory is standard and applications really use it since forever, as opposed to Windows (apps saving user settings to registry, IE6 saving bookmarks to its "Program Files" dir - not sure how it is done now).

One could argue that someone who knew Unix 30 years ago could still use command line in today's Linux without much problems.

At least there are no spaces in the directory names.

Yes and UNIX is really better!

   $ mkdir ".. "
   $ ls -a
   .  ..  ..

Fun eh?

Now go and fire of "rm -rf ..". Now we're talking fun.

Exactly my point!

You can't plan for user stupidity. The system paths do not have spaces in them.

No but allowing leading and trailing spaces is not user stupidity. That is design stupidity.

You know like recording numbers as string in the format: "0000000000000000000000000123.23210000000000000"

The shell by default ignores spaces, you have to specifically force it to acknowledge them by enclosing the name in quotation marks or escaping with a backslash. In that situation, a command line tool should do what it's told, not disobey a direct command. Leave the rubber padded tools for the GUI land.

Rubber padded I do not expect. Retarded I also do not expect.

Explain the validity of:

   mkdir " "
   mkdir "  "
   mkdir "   "

It at least violates the principle of least surprise

Validity? It's a string like any other. It's not mkdir's job to second-guess what I tell it to do.

And I don't see how it violated such principle. You tell it to create a file with spaces - you even emphasized them with quotes -, and you're surprise it did so?

It is merely an illustration. The reality is stuff like that occasionally does happen when you use variables and parse results from other commands.

No it's not. If you type

    mkdir ..<space>

it tells you "cannot create directory". You have to explicitly tell it you want a space in by using quotes (as you well know).

Oh, and it's my computer. I don't want my shell telling me what I'm "allowed" to do. If I explicitly tell it to do something, it should damn well do it!

What if you do the following?

    VARX=".. " # by accident as the result of a command
    mkdir $VARX

Sure, but a non-ambiguous external representation should be mandatory when a semantic value conflates with a presentation one.

Here ' ' is part of both worlds: . .. ..

A first idea: . .. ..\

ls may also take inspiration from bash completion, where dirsep serves as a better delimiter than ' '

    noob ~ $ mkdir -pv lsb/wat
    noob ~ $ cd lsb/wat
    noob ~/lsb/wat $ mkdir ".. "
    noob ~/lsb/wat $ ls -lah
    total 12K
    drwxr-xr-x 3 noob noob 4,0K august  28 16:52 .
    drwxr-xr-x 3 noob noob 4,0K august  28 16:52 ..
    drwxr-xr-x 2 noob noob 4,0K august  28 16:52 .. 
    noob ~/lsb/wat $ cd ..<TAB-COMPLETION>
    ../  .. /

I once accidentally did:

    $ touch ./~
    # Much later I noticed the "~" file in the directory
    # so I mindlessly decided to cleanup:
    $ rm -rf ~

Luckily $HOME was on a _really_ slow NFS mount so I didn't manage to remove anything too important before I killed it.

Ha that's a classic.

A friend of mine did a "rm -rf ~ /tmp" rather than "rm -rf ~/tmp" on a fast SCSI local disk on his Sun workstation.

His backup strategy of tarring up everything into ~/backup went with that one too. I lectured his ass on that one.

However...

The same guy, no less than 8 years later carried around a DVD-RAM with his single copy of his life's work on it (I assume this was only 8 years of his life) and jammed it in one of those dodgy Pioneer slot loader drives and proceeded to mount it.

bzzzt ... wheee ... clunk ... BANG! DVD span to full speed and promptly shattered taking the drive fascia off and spraying everyone with bits of "I told you so".

He now puts up scaffolding which worries me even more...

Luckily I was on an IT maintained machine that takes nightly snapshots of home partitions so I was safe in the worst case. Had I done that on one of my personal machines I wouldn't have been so lucky.

Reply on news.ycombinator.com