[RFC] remove/copy enhancements

[Jordan Breeding]

On Thursday, September 15, 2005, at 12:54PM, Bryan O'Sullivan <bos at serpentine.com> wrote:

>On Thu, 2005-09-15 at 12:30 -0500, Jordan Breeding wrote:
>
>> The problem I see is that `hg cp -p <dir> <new_name>` fails,
>
>Works for me:
>
>~/hg/hg/hg-work $ ls newdir
>ls: newdir: No such file or directory
>~/hg/hg/hg-work $ hg cp -p doc newdir
>copying doc/Makefile to newdir/doc/Makefile
>copying doc/README to newdir/doc/README
>copying doc/hg.1.txt to newdir/doc/hg.1.txt
>copying doc/hgmerge.1.txt to newdir/doc/hgmerge.1.txt
>copying doc/hgrc.5.txt to newdir/doc/hgrc.5.txt
>
>>  `mkdir <new_name>; hg cp -p <dir> <new_name>` works but I end up with
>> <new_name>/<dir>.
>
>If that's not the behaviour you want, then don't use -p.
>
>> 50 jbb at cloves /tmp/jbb_test > hg cp -p three ten
>> abort: with --parents, destination must be a directory
>
>You appear to be getting completely different errors from me.  This
>works for me, for example.  Can you instrument the code a bit, and look
>into what's happening?
>
>	<b

I will just make a quick list of what I have found:

1) if I have a directory structure with only one file in it

   three/four/seven

then I get the error "with --parents, destination must be a directory", which seems to come from the check in mercurial/commands.py:733 through 737

2) I understand how `hg cp` is working now an I had originally though it would operate much more like the `cp` command in UNIX/Linux/OS X/etc.

for two example directory structures:

a)
  three/four/seven

  one/two/three

  (seven, and three are files)

b)
  three/four/seven
  three/four/eight
  three/nine

  one/two/three

  (seven, eight, nine, and three are files)

real cp would do something like:

`cp three ten`: warn that three is skipped because it is a directory
`cp -R three ten`: end up with:
  ten/four/seven
`cp -R three one ten`: error, ten doesn't exist

with hg you get something like the following (with the letter to the left being the starting tree):

(a) `hg cp three ten`: end up with:
  ten (a copy of three/four/seven)
(a) `hg cp -p three ten`: error message "destination must be directory"
(a) `mkdir ten; hg cp -p three ten`: end up with:
  ten/three/four/seven
(a) `hg cp three one ten`: end up with:
  ten/seven
  ten/three
(a) `hg cp -p three one ten`: end up with:
  ten/three/four/seven
  ten/one/two/three

(b) `hp cp three ten`: end up with:
  ten/seven
  ten/eight
  ten/nine
(b) `hg cp -p three ten`: end up with:
  ten/three/four/seven
  ten/three/four/eight
  ten/three/nine

So, it seems that if there are multiple files involved hg acts one way, if it is a single file, it behaves another.  Regular `cp` will only differ in behavior based on the way it was called, and not how many files it ends up needing to handle, regular `cp` also will not ever create a directory as a container, it will either make an exact copy of a tree/file with just the base renamed (dir one to dir two, file one to file two), or it will complain that you have specified multiple files/directories/both on the command line and that the specified destination directory doesn't exist.

I think it would be nice if `hg <command>` behaved like its `<command>` counterpart as much as possible.  Would any of this be made simpler if hg started tracking very basic information for directories (name, version, contents as of that version)?

Jordan