Re: Can someone please comment on this short program

[Matthias Kaeppler <matthias finitestate org>]

Alexander Larsson wrote:
> I'm not sure where the confusion is. I can't talk for the c++ bindings,
> but for gnome-vfs itself the rule is:
>
>   * file: URIs have no inherent encoding
>
> None, nada, zilch. They are the filename as stored on disk, escaped into
> a uri. As such, they are essentially escaped byte strings, the only
> limit being that zero bytes and '/' bytes are not allowed. This is a
> direct result of the fact that unix filenames have no encoding, and is
> nothing we can anything do about in gnome-vfs.

Yes, I actually think the problem lies in the C++ bindings, because as 
opposed to gnome_vfs_uri_new(), the construction of a Gnome::Vfs::Uri in 
gnome-vfsmm requires the string which is passed to Uri::create() to be 
encoded in UTF-8 (that's because create() takes a Glib::ustring, and if 
you pass an std::string in place of a Glib::ustring--such as the one 
returned from FileInfo::get_name()--a conversion attempt will be made).
If the string passed to create() is not valid UTF-8 however, an 
exception will be raised.

On the other hand however, if the string used to initialize the 
Gnome::Vfs::Uri is valid UTF-8 (e.g. after a successful conversion using 
the glib functions), it is possible that after construction, 
Gnome::Vfs::Uri::uri_exists() will always return false /if/ the system 
uses a different encoding for the filename in question.

> You just have to be very careful about your string encodings. I'd
> recommend using hex escapes when specifying a non-ascii string in C
> source code.

it's not only about strings which are hardwired into the source code; 
the problem applies to all other strings as well.

Regards,
Matthias