cp and "blah/." or "." as a source

[Zak]

Hi,

What should happen when I copy a directory ending in "/." to another  
directory?  Like:
        cp -a src/. dst

What I see happening (both with BSD cp and GNU cp) is something  
similar to having used '*' instead of '.':

> laptop:~/cptest zak$ rm -rf *
> laptop:~/cptest zak$ mkdir src
> laptop:~/cptest zak$ touch src/a src/.a
> laptop:~/cptest zak$ ~/coreutils/coreutils-6.6/src/cp -a src/. dest
> laptop:~/cptest zak$ ls -la dest
> total 0
> drwxr-xr-x   3 zak  zak  102 Dec  2 19:06 .
> drwxr-xr-x   4 zak  zak  136 Dec  2 19:06 ..
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 .a
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 a

It even has a go if the source is just ".":

> laptop:~/cptest zak$ cd dest
> laptop:~/cptest/dest zak$ mkdir blah
> laptop:~/cptest/dest zak$ ~/coreutils/coreutils-6.6/src/cp -a . blah
> /Users/zak/coreutils/coreutils-6.6/src/cp: cannot copy a directory,  
> `.', into itself, `blah/.'
> laptop:~/cptest/dest zak$ ls -laR
> total 0
> drwxr-xr-x   5 zak  zak  170 Dec  2 19:14 .
> drwxr-xr-x   4 zak  zak  136 Dec  2 19:06 ..
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 .a
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 a
> drwxr-xr-x   5 zak  zak  170 Dec  2 19:14 blah
>
> ./blah:
> total 0
> drwxr-xr-x   5 zak  zak  170 Dec  2 19:14 .
> drwxr-xr-x   5 zak  zak  170 Dec  2 19:14 ..
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 .a
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 a
> drwxr-xr-x   4 zak  zak  136 Dec  2 19:14 blah
>
> ./blah/blah:
> total 0
> drwxr-xr-x   4 zak  zak  136 Dec  2 19:14 .
> drwxr-xr-x   5 zak  zak  170 Dec  2 19:14 ..
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 .a
> -rw-r--r--   1 zak  zak    0 Dec  2 19:06 a

This means it handles dotfiles as well - a much neater way to copy  
them than using ugly wildcards, tar, rsync, find or whatever else.   
The documentation doesn't make it clear (to me anyway) what /should/  
happen here, but I'd guess (wildly) that POSIX says cp should fail  
and do nothing rather than do what it does.  So what I'm really  
looking for here is to have the current behaviour classified one way  
(correct, and explicitly stated as the expected outcome in the docs)  
or the other (broken, cp needs fixing, and the expected behaviour  
should be stated in the docs; shame, because I like it and don't  
really want it fixed) so I can decide whether to ingrain it in my brain.

The behaviour seems to be down to the following argument munge in  
cp.c (around line 639; arg is "src/." and target_directory is "dst"):
> /* Append the last component of `arg' to `target_directory'.  */
>
> ASSIGN_BASENAME_STRDUPA (arg_base, arg);

(This puts "." into arg_base.)

> /* For `cp -R source/.. dest', don't copy into `dest/..'. */
> dst_name = (STREQ (arg_base, "..")
>                         ? xstrdup (target_directory)
>                          : file_name_concat (target_directory,  
> arg_base, NULL));

This would normally turn "cp -a foo/a bar" into "cp -a foo/a bar/ 
a" (and non-existent target bar/a would get created), but in this  
case it turns "cp -a foo/. bar" into "cp -a foo/. bar/.", and there's  
no need to create "bar/.".

So, is the current behaviour correct or broken?

Cheers,

Zak