It doesn't generate a symbol for the destructor of the *FileLoader
interfaces unless explicitly defined. All other platforms, including
OS X 10.6, don't show this problem.
Refs #7263

This better encapsulates the file checking so that other places don't
have to worry about the desciptor stuff.
Refs #7263

A \r left in the file after opening in binary mode confused the check.
Refs #7263

They don't seem to work everywhere. My guess is that the types isn't
quite complete by the time we use it so some compiles don't like it.
Refs #7263

Also added runtime checks to the subscribe method. This allows it to static
cast at runtime when checking rather than a more expensive dynamic cast.

Also fix initialization that is not a pointer.
Refs #7263

Looking up paths is far more common so make that faster.

It appears that we can't start the FrameworkManager during static
initialization or the destruction of the FrameworkManager instance gets
instance gets confused.

Refs #7523

Refs #7263

The FileLoaderRegistry can now be asked to search its list of loaders
for the best that matches a given file. It passes along a *Descriptor
object to minimize the number of times the file is opened.

Refs #7263

The checks for the headers at the start of HDF4 & HDF5 files have been
put in a static isHDF member to encapsulate the details about
checking for a HDF file and make it reusable.
It will not retain an open file handle to the file so does not
inherit from the standard FileDescriptor.
Refs #7263

and also defines two types of loader, NonHDF and HDF.
Refs #7263

Refs #7263

It was previously not updating the version map when the algorithm was
removed and then subscribing the same thing again would throw an error.
Refs #7263

The stream is open for as long as the object is alive.
Refs #7263

FileDescriptor uses streams rather than c file handles. Refs #7263

Currently based on string names.