Re: [Boost-users] filesystem::path noobish question

From: Nick Dimiduk (ndimiduk_at_[hidden])
Date: 2006-10-27 12:45:16

On 10/26/06, Rush Manbert <rush_at_[hidden]> wrote:
>
> Nick Dimiduk wrote:
> > As per the subject line, I have a rather noobish question about using
> the
> > boost::filesystem tools, specifically the path object.  The idea here is
> > that I want to start my program by removing the "tmp" directory in the
> > current process context if it exists to ensure a clean working
> environment
> > for the remainder of the process.  To the code:
> >
> > ...
> > boost::filesystem::path tmp_dir ("tmp", boost::filesystem::native);  //
> get
> > a native handle for "./tmp"
> > boost::filesystem::remove_all (tmp_dir);  // clean out this tree in the
> fs
> > ...
> >
> > This code acts as expected on linux.  On windows, it causes a
> crash.  The
> > debugger shows the tmp_dir.m_path._Bx is a bad pointer, indicating to me
> > that the constructor failed somewhere along the way.  My questions are
> > these:
> >
> > (1) Is there any error checking I can perform to verify that the path()
> > constructor succeeds, such as a null pointer value or a boolean flag
> within
> > the data type?
> > (2) Is there something about my code which is linux/POSIX specific?
> > (3) Am I missing something else entirely?
> >
> > I'm using boost 1.33.0 and have a mess of other functioning code which
> > suggests that my configuration/installation is correct on both
> platforms.
> >
>
> Hi Nick,
>
> Errors cause a throw of boost::filesystem::filesystem_error, so you need
> to catch it.
>
> I think you also might need to check that boost::filesystem::exists
> (tmp_dir) is true before you call remove_all.
>
> Note also that remove_all throws if tmp_dir is empty, so you need to
> handle that case as well.
>
> So I guess your code should look something like this:
>
> using boost::filesystem;        // Just to save me some typing
> try
> {
>      path tmp_dir ("tmp", boost::filesystem::native);
>      if (exists (tmp_dir))
>      {
>          if (!is_empty (tmp_dir))
>          {   // Directory has content
>              remove_all (tmp_dir);
>          }
>          else
>          {   // Directory is empty
>              remove (tmp_dir);
>          }
>      }
> }
> catch (filesystem_error &er)
> {       // Do something useful here
> }
>
> - Rush

Thanks for the quick response; unfortunately I'm still seeing odd behavior.
I attempted to wrap the code in a try-catch block and it's still failing:

try
{
   boost::filesystem::path tmp_dir ("tmp", boost::filesystem::native);
   boost::filesystem::remove_all (tmp_dir);
   std::string tmp_path (tmp_dir.string().c_str());
}
catch (boost::filesystem::filesystem_error & e)
{
   std::cerr << "boost filesystem exception thrown!" << std::endl;
   exit (1);
}
catch (...)
{
   std::cerr << "Exception thrown!" << std::endl;
   exit (1);
}

No exceptions thrown. The tmp_path constructor fails with an access
violation where the pointer tmp_dir.m_path._Bx is a bad pointer.  I would
expect an exception from both the path constructor and the remove_all call,
but they silently fail.

I can change the construction code to look like the following an it does not
affect the situation in any way:

   boost::filesystem::path env_path;
   env_path /= "tmp";

Thanks again,
-Nick