Re: Order of files with 'cp'

On Fri, Nov 18, 2005 at 09:45:27AM -0600, Eric Anderson wrote:
> Brian Candler wrote:
> >>This just adds a -o flag to cp, which preserves order.
> >
> >
> >Hmm, that's another solution that I hadn't thought of.
> >
> >Advantages: simple to implement. (Even simpler if you use the ?: operator).
> >
> >Disadvantages: it's still strange that the default behaviour is to copy the
> >files in an arbitary shuffled order. The manpage will need updating to
> >document the -o flag, and hence will have to explain the strangeness.
> >Commands arguably have too many flags already.
> 
> I didn't think cp (or any tool, like tar) did it 'arbitrarily', but in 
> order of mtime.  Is that not true?

No, it's not true, for cp anyway.

As far as I can tell, cp indirectly calls qsort() on the source items, using
its own mastercmp() function to compare them. The only comparison it does is
whether each item is a file or a directory.

qsort() is not a stable sort, so even if all items compare equal, it has a
habit of shuffling them around.

brian_at_mappit brian$ cat x.c
#include <stdio.h>
#include <stdlib.h>

static int foo(const void *a, const void *b)
{
    return 0;
}

#define NMEM 7

int main(void)
{
    int a[NMEM] = {1,2,3,4,5,6,7};
    int i;
    
    for (i=0; i<NMEM; i++) printf("%d ", a[i]);
    printf("\n");

    qsort(a, NMEM, sizeof(int), foo);
    for (i=0; i<NMEM; i++) printf("%d ", a[i]);
    printf("\n");
    
    return 0;
}
brian_at_mappit brian$ gcc -Wall -o x x.c
brian_at_mappit brian$ ./x
1 2 3 4 5 6 7 
4 2 3 1 5 6 7 
brian_at_mappit brian$