Re: memory management with glibmm & giomm

[Jonathon Jongsma <jonathon quotidian org>]

nico wrote:
> Hi,
>
> I tried to benchmark my program...
> Results can be found on these charts:
> http://yojik.shtooka.net/profile.pdf
>
> The first graph represent the memory usage after each execution of the
> scan_directory() function.
> (program ends with a memory usage of 25 MB)
> The second graph represents the depth of scanned directories (max depth
> is only 7, so at this point the program store no more than 7
> Gio::FileEnumerator objects).
> The third chart represent the number of scanned files.
>
> With these charts, it quiet clear that the memory usage depend of the
> number of scanned files and doesn't depend of the depth.
> So I guess, my program doesn't correctly "unref" objects. Something goes
> wrong.
>
>
> I paste here the source code of the program :
>
> > #include <iostream>
> > #include <glibmm.h>
> > #include <giomm.h>
> >
> > int depth;
> > int files;
> >
> > bool scan_directory(std::string path) {
> >     depth++;
> >     try {
> >         Glib::RefPtr<Gio::File> directory = 
> > Gio::File::create_for_path(path);
> >         Glib::RefPtr<Gio::FileEnumerator> enumerator = 
> > directory->enumerate_children();
> >         Glib::RefPtr<Gio::FileInfo> file_info;
> >         while (file_info = enumerator->next_file()) {
> >             files++;
> >             if (file_info->get_file_type() == Gio::FILE_TYPE_DIRECTORY)
> >                 scan_directory(path + file_info->get_name() + 
> > G_DIR_SEPARATOR_S);
> >             else {
> >                 //do some staff, for example:
> >                 //std::cout << path + file_info->get_name() << "\t" << 
> > file_info->get_content_type() << std::endl;
> >             }
> >         }
> >         enumerator->close();
> >         depth--;
> >         g_mem_profile();
> >         std::cout << "depth: " << depth << std::endl;
> >         std::cout << "files: " << files << std::endl;
> >         std::cout << "###" << std::endl;
> >         return true;
> >     }
> >
> >     catch(const Glib::Exception& ex) {
> >         depth--;
> >         std::cerr << "Error: " << ex.what() << std::endl;         
> > return false;
> >     }
> > }
> >
> > int main(int argc, char** argv) {
> >     depth = 0;
> >     files = 0;
> >
> >     g_mem_set_vtable(glib_mem_profiler_table);
> >
> >     setlocale(LC_ALL, "");
> >     Glib::init();
> >     Gio::init();
> >
> >     scan_directory((argc == 2) ? argv[1] : "");
> >
> >     return 0;
> > }
>
> > it depends a bit on the structure of your filesystem. you are doing
> > depth-first descent through a directory structure. in some cases, this
> > would not use many resources; in others, you'd expect it to use a great
> > deal.
> >
> > every single directory encountered will be enumerated, then each
> > directory within it will be recursively entered. by the time its done,
> > you've got enumerated data for just about every file below the starting
> > point. none of your data structures are cleaned up until control returns
> > to the top level.

If you think this is a bug, it would be helpful if you could open a bug 
in bugzilla so that we can remember to investigate:
http://bugzilla.gnome.org/enter_bug.cgi?product=glibmm

Thanks.

--
jonner