Re: [boost] Re: [filesystem] create_directories not thread-safe?

From: Beman Dawes (bdawes_at_[hidden])
Date: 2004-03-09 15:01:59

At 12:55 PM 3/9/2004, Peter Dimov wrote:

 >Walter Landry wrote:
 >> Beman Dawes <bdawes_at_[hidden]> wrote:
 >>> At 01:55 AM 3/9/2004, Vladimir Prus wrote:
 >>>  >After all, caller only cares if directory exists, if it was
 >>>  created by >another thread it's ok.
 >>>
 >>> That's an interesting thought. Of course, it might not be another
 >>> thread but rather another process that created the directory, and
 >>> that could be a sign of real trouble.
 >>
 >> Exactly.  Normally people use lock-files, but I personally have a use
 >> for lock-directories.  If boost::filesystem starts swallowing errors,
 >> it makes it more difficult to achieve that.
 >
 >It doesn't have to swallow errors. It can simply signal some errors via
 >exceptions and some errors via return values.
 >
 >> I would say that people who don't care whether a directory is created
 >> should just catch the exception themselves.  Alternately, you could
 >> generalize make_directory to take an extra, optional argument that
 >> specifies whether to throw if the directory already exists.  Of
 >> course, it should always throw if the error is something different
 >> (e.g. if the parent directory doesn't exist).
 >
 >This almost goes without saying.
 >
 >    create_directory(path p)
 >
 >    post: p exists and is a directory
 >
 >If the postcondition cannot be satisfied, create_directory _must_ throw.
 >But if p already exists and is a directory, the postcondition already
 >holds, so we are free to report the "error" via a different mechanism
 >(COM, for example, has a conventional "alternate success code" S_FALSE
 >for such situations).

We have a similar case with remove(), which is reported by remove() 
returning a bool indicating if it actually removed anything.

It would be quite consistent for create_directory() to return a bool 
indicating if it actually had to create the directory.

--Beman