Hi Andy, I saw the source code in seed_rng.hpp, I found you use an uninitialized buffer[] if fopen fails. It often rely on the stack contents when previous calling function runs, and that is not always random. I think UUID does not need to be random, it just need to be uniquely. So the seed should contains process id, time, MAC address and so on. 2008/10/9 Andy Tompkins <atompkins@fastmail.fm>:

Hi Johannes,

On Wed, 8 Oct 2008 15:03:54 +0200, "Johannes Brunen" <JBrunen@DataSolid.de> said:

...

Hello Andy,

I just did a small test with the latest uuid library from the vault.

IMHO, I found a small error in file uuid.hpp (line 303): function hash_value must be inlined as it is not a template function.

Absolutely correct. Thanks. I will fix this.

...

Additional is seems that the documentation is out of sync with the implementation. The section 'design notes' is still refering to a former implementation with the create(...) functions.

Again, absolutely correct. Thanks. I will fix this too.

...

Otherwise, I could compile and link the code on msvc .net 2008 with my setup.

thanks for your library,

best

Johannes

Thank you, Andy Tompkins _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

Hi Mathias, On Thu, 09 Oct 2008 12:54:52 +0200, "Mathias Gaunard" <mathias.gaunard@ens-lyon.org> said:

Atry wrote:

...

Hi Andy, I saw the source code in seed_rng.hpp, I found you use an uninitialized buffer[] if fopen fails.

As a side note, would it be possible to use hooks or something so that reading that uninitialized memory doesn't generate errors with memory checking tools such as valgrind?

I assume it would be possible, and I'm quite open to the idea. Do you have any suggestions on how to do this? Regards, Andy Tompkins

Hi Atry, On Thu, 9 Oct 2008 13:15:23 +0800, "Atry" <pop.atry@gmail.com> said:

Hi Andy, I saw the source code in seed_rng.hpp, I found you use an uninitialized buffer[] if fopen fails. It often rely on the stack contents when previous calling function runs, and that is not always random.

I think UUID does not need to be random, it just need to be uniquely. So the seed should contains process id, time, MAC address and so on.

There are different types of uuids based on different algorithms. In this case uuid_generator is generating a random-based uuids, and so a random number generator is used and needs to be seeded well to produce the best possible random data. Another type of uuid is a time-based uuid, which does incorporate data like the MAC address. I have not implemented this algorithm. Still I could use data such as process id, and MAC address as random data itself, or to seed a random number generator, but I don't know of portable ways to get this data. < snip > Regards, Andy Tompkins

Hello Andy, after some investigations, I think that I have found the origin of the problem: In seed_rng.hpp the following code is used: template <typename UniformRandomNumberGenerator> inline void seed(UniformRandomNumberGenerator& rng) { seed_rng seed_gen; boost::generator_iterator<seed_rng> begin(&seed_gen); boost::generator_iterator<seed_rng> end; rng.seed(begin, end); } Problematic is the construction of the end iterator. Looking into the code (boost/generator_iterator.hpp) shows, that the default constructor of generator_iterator does not initialize its member variables. generator_iterator() {} Howerver, the mersenne_twister seed code does compare the two given iterators as shown below (boost/random/). template<class It> void seed(It& first, It last) { int j; for(j = 0; j < n && first != last; ++j, ++first) x[j] = *first; i = n; if(first == last && j < n) throw std::invalid_argument("mersenne_twister::seed"); } As the end iterator is not initialized the way is open for all throwing the invalid_argument exception. Locally, I did solve the problem by patching the default constructor of class generator_iterator. If this is a valid solution of the problem, it would be fine if anyone applies the correction to the boost code. Greetings, Johannes

Hi Johannes, On Mon, 13 Oct 2008 13:12:02 +0200, "Johannes Brunen" <JBrunen@DataSolid.de> said:

Hello Andy,

after some investigations, I think that I have found the origin of the problem:

In seed_rng.hpp the following code is used:

template <typename UniformRandomNumberGenerator> inline void seed(UniformRandomNumberGenerator& rng) { seed_rng seed_gen; boost::generator_iterator<seed_rng> begin(&seed_gen); boost::generator_iterator<seed_rng> end; rng.seed(begin, end); }

Problematic is the construction of the end iterator. Looking into the code (boost/generator_iterator.hpp) shows, that the default constructor of generator_iterator does not initialize its member variables.

generator_iterator() {}

Howerver, the mersenne_twister seed code does compare the two given iterators as shown below (boost/random/).

template<class It> void seed(It& first, It last) { int j; for(j = 0; j < n && first != last; ++j, ++first) x[j] = *first; i = n; if(first == last && j < n) throw std::invalid_argument("mersenne_twister::seed"); }

As the end iterator is not initialized the way is open for all throwing the invalid_argument exception.

Locally, I did solve the problem by patching the default constructor of class generator_iterator.

If this is a valid solution of the problem, it would be fine if anyone applies the correction to the boost code.

Greetings, Johannes

I have put a new version in the boost vault. http://www.boostpro.com/vault/index.php?action=downloadfile&filename=uuid_v13.zip&directory=& It addresses this problem by making a copy of boost::generator_iterator with your fix. Regards, Andy.