Re: [Feature Request] UUID in fstab.

["S.P.Zeidler" <spz%NetBSD.org@localhost>]

Thus wrote Utkarsh Anand (utkarsh009%yandex.com@localhost):

> >There is no reason whatsoever not to automatically create
> >useful NAME labels that are also human readable.
> >a) You are aware that uuid collisions may be rare but not impossible?
> Putting the above statements together, do you now see a fallacy in your 
> argument?

I do not see a fallacy in my argument. There may be a fallacy
in what you infer my argument was, but that's a whole different
kettle of fish.

One result of "uuids may collide" is that you need to check
for duplicates with uuids too.

>  Useful name labels are more likely to collide, if you're 

They only need to be unique on that machine (or set of machines if you
have removable media).

> picking words out of a dictionary and using those chunks instead of 
> alphabets.

If you have 32 hex-numbers, you have
340282366920938463463374607431768211456 options. If you use 4 random
words out of /usr/share/dict/words, you only get
3100519963322993491281 variety, but if you really are worried that's
not enough you can use 8 words and get
9613224042964416904375938680615889423020961 options.

>  Even if you just go on numbering them from 1...whatever the 
> last disk number is,

That would be just about the most stupid thing you could be doing.

>  you are more likely to end up with a collision 
> later on, while adding more disks manually.

The implementation you are envisioning is breathtakingly stupid.
We may only implement basic competence when we copy Linux?

> BTW, you'd be generating labels only when they don't already exist. But,
> in case of UUID, you'll always be regenerating it to ensure that there 
> are no collisions.

So if you are installing with removable media present, you get to fix
its fstab entries on every other computer in its usage set? That does
not strike me as a desireable trait (note: if you don't reformat a drive
enterprisey Linuxes don't mess with its partition UUIDs either).

> >b) In NetBSD, the names of disks are not in fstab, the relevant
>    item is the name of the partition.
> So, what? Readable names are more likely to collide (as mentioned 
> earlier).

This assertion needs proof.

> >Why is getting a "you have a name conflict for partition a to be 
> >mounted on /foo and partition b to be mounted on /bar, please rename 
> >either" a problem in the partitioning stage?
> Not a problem, just more cumbersome for an average user. You wouldn't 
> assume the target audience to be sysadmins, or professional users, would
> you? Think from an average user's perspective.

The average NetBSD user is not a complete dummy.

> >Automatically generating a name from diskname and mount path is trivial
> >, and likely to be unique enough for one system. (IOW: how many /usr
> >do you expect to have?)
> So, you want to rename the labels every installation? I believe labels 
> are for the users to decide and not system.

let me pseudo-code that for you:
if (label == empty)
	label = generate_label_recipy();

> >If that is not random enough for you: picking 4 words out of dict is
> >still more human friendly than a long hex number.
> Like I said already, much much more chance of collision for the same 
> total length of the string.

This assertion needs proof.

12345678-abcd-abcd-abcd-12345678abcd
a-dog-ate-five-penny-worth-pork-meat

consider that one takes its positions from [0-9a-f] and the other
from [0-9a-z]

> >"We must make everybody suffer under a user-hostile default because
> >someone MIGHT be an idiot" is a very bad argument for a Unix system.
> >Unix traditionally has "we don't stop you doing something stupid
> >because that might stop you doing something smart".
> Most users wouldn't even care about /etc/fstab or UUID. Sysadmins and 
> professional users are smart enough to handle that.

To clarify:
I was referring to users in the sense of not-the-developers, but those
that use the software; that includes sysadmins. And yes, I am "smart enough"
to use UUIDs even when I cannot handle them in the only reasonable manner,
which is cut & paste, but those are four hours of serious aggravation
I am never getting back and do not want inflicted on anybody else
(nor my future self).

> >(Also: not trying to be Linux. Linux already exists, there is no need
> >to create a copy)
> I think the you should focus more on device drivers and kernel for 
> setting yourself apart from linux.

Thank you for your advice on how I should spend my free time.
I will, however, continue to use it as I see fit.

> I could also say that you're just trying to be anti-linux and do just 
> condemn every decision they take.

Again, you take what I wrote and galopp off into the blue with it.

I use Linux, quite a bit, where it is the appropriate fit for a very large
number of parameters. I also use other systems where they are the best
solution to the problem and the environment. I even use Windows where
that is the rational choice.

RedHat does not need to be Suse, because Suse already exists and when
I need Suse I can get Suse. Linux does not need to be Solaris etc etc
And likewise there is no need to make NetBSD indistinguishable from
Linux because Linux already exists, and when you want or need Linux
you can readily get it.

Given that, it is especially daft to reproduce -errors- just because
another system makes them. Learning from others oopses is how we get
smarter faster. Using an identifier to identify filesystems to mount
is a good idea (that was already present in AmgaOS way back when,
and I'd not be surprised at even earlier art). Using a label that is
distinctly human-unfriendly without need, otoh, is not the smart choice
unless you intend to annoy sysadmins.

regards,
	spz