Re: [boost] [filesystem] create_directories

From: Beman Dawes (bdawes_at_[hidden])
Date: 2002-12-17 16:59:45

At 04:42 AM 12/17/2002, Vladimir Prus wrote:
 >
 >Beman Dawes wrote:
 >
 > >  >6.   Would it be reasonable to introduce a function
 >"create_directories",
 > >  >      similiar in spirit to "remove_all"? That function would create
 > >  >      intermediate directories, not only the leaf one.
 > >
 > > Yes, that would be both reasonable and useful. The
 > > filesystem/convenience.hpp header we've discussed would be a good 
place.
 >
 >Agreed. BTW, where would that header live? In CVS tree or in sandbox? And
 >what would be the access rules?
 >
 > >
 > > Care to contribute it?
 >
 >See attachment.

Nice!

 > A pretty trivial excersice, of course, except for one
 >point. The create_directories function as written by me does nothing (not 

 >throws) if you call it on existing directory. And explicit "exists" all 
is
 >certainly needed for parent directories, and I'm not sure what to do with
 >the full directory.
 >
 >Allowing it to exist seems in spirit of this function --- after all it
 >automatically creates needed directories and does nothing with already
 >existing.

Yes, I think so too.

How about the following?  Note that moving the "exists" up also prevents 
the problem of trying to create_directories on the root directory.

     void create_directories(const path& ph)
     {
         if (ph.empty() || exists(ph))
             return;

         // First create branch, by calling ourself recursively
         create_directories(ph.branch_path());
         // Now that parent's path exists, create the directory
         create_directory(ph);
     }

 >OTOH, this behaviour is different from 'create_directory', which throws 
if
 >directory does not exist.

No different from remove/remove_all. The convenience functions should be 
just that; convenient. Worrying about consistency produces a less useful 
interface.

 >Lastly, we probably can't change create_directory, at least for one 
reason.
 >On Linux, this operation is atomic, and can be used to make locks. I 
guess,
 >the same is true on Windows. If 'create_directory' checks for 
existence, >you cannot use it for locking any more.
 >
 >So, what's the right interface?

See above.

Thanks,

--Beman