bug#26741: Numeric sort in ls

[Eric Blake]

tag 26741 notabug
thanks

On 05/01/2017 07:50 PM, Tony Malykh wrote:
> Hi all,
> 
> I am wondering if there is a way to do a numeric sort in ls?

Thanks for the report; the quick answer is that sort can already do what
you want, so post-process your ls output with sort.

> Here is a concrete example.

Thanks, that makes it much easier to reproduce what you want, and
demonstrate along the way what is going on.

> 
> Here are some files
> $ ls -1
> 1.py
> 11.py
> 2.pl
> 2.py

I'll compress that set of data to:

$ printf %s\\n 1.py 11.py 2.pl 2.py

> 
> Desired output:
> $ ls -1 --<magic-options>
> 2.pl
> 1.py
> 2.py
> 11.py

If you are okay with:

ls -1 | sort <magic-options>

then we are good to go. In fact, in that configuration, you can use
plain 'ls' rather than 'ls -1' (since POSIX requires -1 to be the
default behavior when stdout of ls is a pipeline).

> # Note that 2.pl goes to the top because of its extension.
> # Note also that the *.py files are sorted numerically and not
> lexicographically.

So that says that you want your primary sort key to be the extension
field, sorted lexically; and your secondary sort field to be the rest of
the name, sorted numerically.  How about the following?

$ printf %s\\n 1.py 11.py 2.pl 2.py | sort -t . -k2,2 -k1,1n
2.pl
1.py
2.py
11.py

Again, with your directory layout, you could use:

$ ls | sort -t . -k2,2 -k1,1n

for the same results.

Now, is that entirely robust? Not really - all it takes is one file with
no extension, or one file with embedded '.' in the name beyond the
extension, and you are no longer able to reliably specify which sort
field is the extension.

$ printf %s\\n 4.4.py 3 1.py 11.py 2.pl 2.py | sort -t . -k2,2 -k1,1n
3
4.4.py
2.pl
1.py
2.py
11.py

But never fear - we can use the decorate-sort-undecorate pattern to
temporarily swap things around:

$ printf %s\\n 4.4.py 3 1.py 11.py 2.pl 2.py \
 | sed 's/\(.*\)\.\([^.]*\)$/\2.\1/' \
 | sort -t . -k1,1 -k2n \
 | sed 's/\(^[^.]*\)\.\(.*\)/\2.\1/'
3
2.pl
1.py
2.py
4.4.py
11.py

(although when you start getting that complex, awk, perl, or a dedicated
C program start to sound more appealing).

> 
> Attempt #1:
> $ ls -1Xv
> 1.py
> 2.pl
> 2.py
> 11.py
> # Note that the files are sorted numerically, but the extension sort
> (-X) option seem to have been ignored, since the 2.pl file is among
> the *.py files.

That's because 'ls' has exactly ONE level of sort. You cannot specify a
primary and secondary key to ls, but rather the last sort type requested
overrules all earlier requests.  The only program with multiple levels
of sort is, not surprisingly, 'sort'.  'ls -Xv' is identical to 'ls -v'.

> 
> Attempt #2:
> $ ls -1vX
> 2.pl
> 1.py
> 11.py
> 2.py
> # Now the extension sort is working, but the files are sorted
> lexicographically and not numerically.
> 

'ls -vX' is identical to 'ls -X'.

It appears that your request is to modify ls directly to subsume the
ability that sort already has to list multiple sort keys, and break ties
under one key by resorting to the next.  However, the bar is VERY high
to add any new features to ls, and unless you can point to existing
practice of some other ls implementation that does the same, we are
probably going to leave it at requiring you to post-process the data.
Besides, you'd have to wait for a new version of ls to be build and land
in your distro, while you already have post-processing tools at your
disposal that are more portable.

So, for now, I'm closing this as not a bug, although you should feel
free to continue the conversation if you have more to add.

> Here is my version:
> $ ls --version
> ls (GNU coreutils) 8.26
> Packaged by Cygwin (8.26-2)

[hmm - a good reminder that my TODO list includes packaging an updated
coreutils for cygwin...]

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

signature.asc
Description: OpenPGP digital signature