Re: [boost] [Filesystem] Relational operators

From: Beman Dawes (bdawes_at_[hidden])
Date: 2004-03-21 17:44:19

At 06:01 AM 3/21/2004, Martin wrote:

 >I have used the filesystem library for some time now and I am really 
happy
 >with it.

Glad to hear that!

 >In the CVS I noticed that it recently added relational operators and
 >an "equivalent" function.
 >
 >To my surprice the relational operators are implemented in a non-portable 

 >way.

Umm... The implementation should be portable.

 > It only does a lexicographical compare! In the rationale it says:
 >
 >"Programmers wishing to determine if two paths are "the same" must
 >decide if that means "the same representation" or "resolve to the
 >same actual file or directory", and choose the appropriate function
 >accordingly."
 >
 >In windows the path "file.cpp" is the same as "FILE.CPP" but not in 
posix.

No, that isn't correct. Those paths are always different on all operating 
systems, because only the textual representation is considered. Likewise 
"foo/bar" and "foo/baz/../bar" are different paths, even though they may 
resolve to the same file.

Whether "file.cpp" is equivalent to "FILE.CPP" on Windows depends on 
whether or not the FILE_FLAG_POSIX_SEMANTICS flag was set when the files 
were created, the specific file system, etc.

 >So
 >if I write portable code I need to do add a function like
 >
 >bool eq_path(const path& path1, const path& path2) {
 >#if BOOST_POSIX
 >  return path1.string() == path2.string();
 >#else
 >  return boost::algorithm::iequals(path1.string(), path2.string());
 >#endif
 >}
 >
 >In most of my cases the equivalent function is not an alternative since 
it
 >requires that both paths can be evaluted by the filesystem, i.e. paths 
must
 >exist and the parent of relative paths must be known.

Due to the possibility of links, rules that change according to what kind 
of actual file system is mounted, etc, it isn't possible on modern 
operating systems to tell if two paths which are textually different 
resolve to the same file without actually accessing the file system.

Your code above is fragile; it may work most of the time but then fail 
unexpectedly.

 >
 >---------------------------------------
 >
 >Another minor question: Why does the "leaf()" member function return a
 >string and not a path?

Because only one element is ever returned. Path elements are represented as 
strings.

--Beman