boost 1.56 changed Boost.Filesystem recursive_directory_iterator

boost-1-56-recursive-directory.cpp

/*
 * Workaround Boost.Filesystem V3 behaviour change in boost 1.56.
 * See original at http://thisthread.blogspot.com/2011/07/using-recursivedirectoryiterator.html
 */

void plainListTree(boost::filesystem::path path) // 1.
{
    dump(path, 0);

    boost::filesystem::recursive_directory_iterator it = createRIterator(path);
    boost::filesystem::recursive_directory_iterator end;

    while(it != end) // 2.
    {
        const std::string curPath(it->path().string());

        dump(*it, it.level()); // 3.

        if(boost::filesystem::is_directory(*it) && boost::filesystem::is_symlink(*it)) // 4.
            it.no_push();

            // advance it
        try
        {
            ++it; // 5.
        }
        catch (boost::filesystem::filesystem_error & e)
        {
            std::cout << "warn: " << curPath << ": " << e.what() << std::endl;

            /*
             * boost 1.56 changed the behaviour of the recursive_directory_iterator in two ways:
             * 1. The iterator is advanced before the error is raised (exception thrown),
             *    so the no_push() workaround isn't required. If that workaround is used,
             *    the next path is incorrectly skipped.
             * 2. Because of (1), accessing it->path().string() in error handling
             *    prints the _next_ path, not the erroneous one.
             * Handle the older behaviour below.
             */
#if BOOST_VERSION < 105600
            try
            {
                it.no_push(); // 6.
                ++it;
            }
            catch (boost::filesystem::filesystem_error & e)
            {
                std::cout << "fatal error: " << curPath << ": " << e.what() << std::endl;
                throw;
            }
#endif // BOOST_VERSION < 105600
        }
    }
}

Background

The original code was in http://thisthread.blogspot.com.au/2011/07/using-recursivedirectoryiterator.html

The pattern worked fine with boost 1.41 (Boost.Filesystem V2) on CentOS 6. However, I experienced a behaviour change when using boost 1.64 (Boost.Filesystem V3) on macOS 10.12.

After a lot of debugging, it appears that boost 1.56 changed the behaviour of the recursive_directory_iterator in two ways, one of which directly affects the pattern: (Boost.Filesystem V3 was made the default in boost 1.46, and this behaviour change wasn't made until 1.56):

1. The iterator is advanced before the error is raised (exception thrown), so the no_push() workaround isn't required. If that workaround is used, the next path is incorrectly skipped.
2. Because of (1), accessing it->path().string() in error handling prints the next path, not the erroneous one.

Fix

1. Cache it->path().string() for use in error messages;
2. Only do the it.no_push();++it if boost < 1.56.

I have put a link to this gist in the blog. Thank you, Luke!