(In reply to comment #0)
> In the SourceLanguageManager, there's supposed to be a "guess_language"
> function with the following implementation:
> 
> 
> Glib::RefPtr<SourceLanguage> SourceLanguageManager::guess_language(const
> Glib::ustring& filename, const Glib::ustring& content_type)
> {
> 
>   Glib::RefPtr<SourceLanguage> retvalue =
> Glib::wrap(gtk_source_language_manager_guess_language(gobj(), filename.c_str(),
> content_type.c_str()));
>   if(retvalue)
>     retvalue->reference(); //The function does not do a ref for us.
>   return retvalue;
> }
> 
> 
> The semantics are supposed to be that you pass in both parameters, but only one
> or the other are non-NULL (it doesn't know what to do if you specify both or
> neither).

Well, that is not true - the only forbidden situation is passing invalid string as both parameters (where by invalid I mean that either is NULL or "" - both are treated as invalid). See:
http://git.gnome.org/browse/gtksourceview/tree/gtksourceview/gtksourcelanguagemanager.c?id=c5afa6d8c8145942eac6448dde1dde6824465a4b#n634

> However, if you actually try to call it with NULL for either one of
> the two values, you fail with:
> 
>  terminate called after throwing an instance of 'std::logic_error'
>    what():  basic_string::_S_construct NULL not valid
> 

Seems that you probably are passing NULL to Glib::ustring constructor which passes it to std::string constructor, which does not like NULL strings. Compile and run this code to see it:

#include <iostream>
#include <string>

int main()
{
  std::string str (0);
  std::cout << str << std::endl;
}

> Whereas if you pass in an empty string instead, you fail differently with:
> 
> (algoviz:12581): GtkSourceView-CRITICAL **:
> gtk_source_language_manager_guess_language: assertion `(filename == NULL ||
> *filename != 0) && (content_type == NULL || *content_type != 0)' failed
> 
> I've confirmed that this is just a mapping error by replacing the
> content_type.c_str() call with "NULL" directly in the source code, and it
> worked fine after that. But I thought I'd point this bug, since the function is
> pretty useless the way it is right now since it's uncallable.

If such assertion popped, then that means that filename and content_type parameters are both invalid. Before you call the guess_language check the passed strings if they are not empty. Something like this:

Glib::ustring filename (get_filename_from_somewhere ());
Glib::ustring content_type (get_content_type_from_somewhere ());

if (!filename.empty () || !content_type.empty ())
{
  lang = lang_manager->guess_language (filename, content_type);
}
else
{
  // error - both strings are empty.
}

I'm saying it happens even if only ONE of them is invalid. For example, the following line:

Glib::RefPtr<gtksourceview::SourceLanguage> source_language = source_language_manager->guess_language("test.cpp", NULL);

AND the following line:

Glib::RefPtr<gtksourceview::SourceLanguage> source_language = source_language_manager->guess_language("test.cpp", "");

Both fail at runtime. The only way to actually call this function with one parameter only is to go

Glib::RefPtr<gtksourceview::SourceLanguage> source_language = source_language_manager->guess_language(file, " ");

at which point it works fine. Maybe this is how it is intended? But that is not made clear anywhere at all, and it seems like a rather silly workaround since the C version will gladly take the first possibility.

(In reply to comment #2)
> I'm saying it happens even if only ONE of them is invalid. For example, the
> following line:
> 
> Glib::RefPtr<gtksourceview::SourceLanguage> source_language =
> source_language_manager->guess_language("test.cpp", NULL);
> 

This is programming error, so that won't work. Instead of NULL (or 0), Glib::ustring() or just "" should be passed.

> AND the following line:
> 
> Glib::RefPtr<gtksourceview::SourceLanguage> source_language =
> source_language_manager->guess_language("test.cpp", "");
> 
> Both fail at runtime. The only way to actually call this function with one
> parameter only is to go
> 
> Glib::RefPtr<gtksourceview::SourceLanguage> source_language =
> source_language_manager->guess_language(file, " ");
> 

Now I see it. This assertion in C function gtk_source_language_manager_guess_language():
g_return_val_if_fail ((filename == NULL || *filename != 0) &&
			      (content_type == NULL || *content_type != 0), NULL);

is probably a culprit. I'll ask gtksourceview maintainers if there should be `||' instead of `&&'. If not - I'll add overloads for both NULL parameters.

(In reply to comment #3)
> (In reply to comment #2)
> > I'm saying it happens even if only ONE of them is invalid. For example, the
> > following line:
> > 
> > Glib::RefPtr<gtksourceview::SourceLanguage> source_language =
> > source_language_manager->guess_language("test.cpp", NULL);

I am surprised that that even compiles. Does it really?

Yes, in general, it's easiest if the C functions treat "" and 0 in the same way, but they often forget to.

(In reply to comment #4)
> (In reply to comment #3)
> > (In reply to comment #2)
> > > I'm saying it happens even if only ONE of them is invalid. For example, the
> > > following line:
> > > 
> > > Glib::RefPtr<gtksourceview::SourceLanguage> source_language =
> > > source_language_manager->guess_language("test.cpp", NULL);
> 
> I am surprised that that even compiles. Does it really?

It just uses Glib::ustring (const char*) constructor.

Now I'm waiting for response in https://bugzilla.gnome.org/show_bug.cgi?id=624287

The assertion was fixed in gtksourceview (see bug 624284) and will be available in gtksourceview-3.0, so gtksourceviewmm-3.0 won't need the fix. As of gtksourceviewmm-2.0 - I'll change guess_language() method definition in a way it will pass NULL to C function when empty string is passed to the method. I'll do it tomorrow evening.

Fixed in git - the change will be available in gtksourceviewmm-2.10.1 version when it is released.

Thanks, guys! I appreciate it.