- Boost - lists.boost.org

Mailing list moderation, (how) does it work?
by Lars Viklund 03 Apr '12

03 Apr '12

Hi list. There's a bloke over in the IRC channel that posted a message to the list last wednesday titled '[Endian] library proposal' which seems to be stuck somewhere as it hasn't showed up yet. Is this list still moderated-on-first-post or might there be another root cause for it not appearing? -- Lars Viklund | zao(a)acc.umu.se

4 4

[Endian] library proposal
by David Stone 03 Apr '12

03 Apr '12

https://bitbucket.org/davidstone/endian A while back, I had the somewhat common problem of networking in C++. Libraries like Boost.Bind and Boost.Asio proved immensely helpful, but there wasn't anything in Boost that would take care of the byte ordering. Fortunately, I could use the hton family of functions (htonl, htons, ntohl, ntohs) because all of my networking code was 4 bytes or smaller. Later on, I was writing some code that created hashes (specifically MD5 and SHA-2). In the quest for more performance, I tried to maximize the amount of memory I accessed at once (grabbing 4, 8, or even sizeof (uintmax_t) bytes at a time, instead of 1). This also meant that I had to make sure I manipulated those bytes in the proper order, so I went to my networking code and tried to reuse that. However, I quickly ran into some problems, the largest of which is that there is no htonll or similar for 8-byte integers. So I did what any self- respecting programmer would do: I wrote my own using bit shifting. My work paid off, and I was rewarded with a few % increase in performance in my library. I naturally wondered if I could do better, and I found out that yes, I could do better than dynamic endian determination and manual bit shifting. This led me to make a complete library, with the goal of submitting it to Boost. Imagine my surprise to find out that just a few months ago, Beman Dawes submitted his own version of such a library! However, I feel my version offers a few improvements over parts of Dawes's library. My library does not provide any extra integer types. All mine does is convert the representation. For this reason, I feel it would probably be best for the final Boost.Endian library to combine my library with Dawes's. Here is an overview of the features of my library, with emphasis given on the relative strengths of what I have written (in no particular order) * My library uses compiler intrinsics where they exist. Currently, it uses intrinsics with GCC, Clang, MSVC, and (according to the documentation, I haven't tested) ICC. On GCC, where I have done most of my testing, using intrinsics gave a small (but measurable) performance boost. I believe my timings suggested about a 3.5% increase in speed using the intrinsic version vs. using a manual bit- shifting version with compiler options: -Ofast -march=native -funsafe-loop-optimizations which I believe is close to the strongest set of optimizations gcc supports. * My library uses static determination of endianness where possible (as defined in boost/detail/endian.hpp), and dynamic determination of endianness only when that fails. Static determination gave me about a 1.25% increase in speed over dynamic determination (and has slightly smaller memory requirements, but they are probably on the order of one word or two). * My library uses templates to determine which byte reordering function to call. This genericity was essential to support my SHA-2 solution, for instance, because the SHA-2 hashing algorithm is actually a family of 4 functions, half of which use a 32-bit word size, and half of which use a 64-bit word size. I had already minimized code duplication by use of templates, and a template-based solution works much better with generic code (while not inconveniencing non- template code) than functions like htonl / htons. * My library returns results via return value, rather than by reference. This is more natural and likely leads to greater performance (due to not having to reference / dereference built-in types). It also allows const-correct programming. * My library includes thorough unit testing that doubles as a performance test. Compiling the .cpp file and calling it normally is a simple unit test that fully exercises its capabilities (it calls multiple versions of conversion functions and verifies dynamic byte order determination is working as expected). It also accepts an optional command-line argument as an integer that states how many iterations of the unit tests to run. The program then reports how long each test took. * Like Dawes's library, my library does not (yet) include support for floating point numbers. The reason for this is that floating point endianness is not well documented or as consistent. It's possible for there to be little endian floating point numbers and big endian integers on the same machine. Rather than suggest that the problem is solved by adding in support for floating point numbers, attempting to use them will result in a compile-time error. However, thanks to my library's use of type traits, changing around most of the functions is as simple as changing is_integral with is_numeric. I'm not even sure if traditional endianness can describe floating point numbers as a whole, or if (for example) the mantissa and the exponent are each stored in a particular (hopefully identical) endianness. This could lead to numbers with byte ordering like DCBA HGEF (where the first part represents the mantissa and the second the exponent, or perhaps it's the other way around). Basic byte-swapping functions will not cope with this, and as such, I'll need to do more research. * My library supports big-endian, little-endian, and PDP-endian. * I have tested the library with gcc 4.6.2 on Fedora 16 64-bit and Fedora 16 32-bit. I have tested with clang 3.0 on Fedora 16 64-bit. I have tested with MSVC 9 and 11 (I believe those were the versions I used) on Windows 7 64-bit. My code compiles with no warnings in GCC, even when set to the paranoid warning levels described in this post: http://stackoverflow.com/a/9862800 . It nearly compiles cleanly in Visual Studio at /W4, except that it complains about a conditional expression being constant in a template (it's always true for some instantiations, and always false for others). Even with /Wall, it compiled cleanly (other than library code), with the exception of the above mentioned warning about constant conditional expressions, and a few confusing signed / unsigned mismatch warnings. Unfortunately, I do not have a big-endian or pdp- endian machine to test on. However, my MD5 and SHA-2 tests still validate (which they would fail to do if my big-endian routines were wrong). * I'm a little uncertain about my dynamic byte ordering. I currently determine it once and save the result to a global const bool. I'm not sure if that's the best way (but it did give a few % increase in performance over computing it every time). For these reasons, I hope you consider my library for inclusion in Boost. https://bitbucket.org/davidstone/endian

1 1

[scope_exit] C++11 freely uses typename (on MSVC10?)
by lcaminiti 03 Apr '12

03 Apr '12

Hi all, Isn't it C++11 supposed to allow to use typename freely even outside type-dependent context? Does MSVC10 support this? For example, is the following supposed to compile on MSVC10? #include <boost/mpl/identity.hpp> int main() { typedef typename boost::mpl::identity<int>::type t; t i = 0; return i; } Note: Using typename freely will allow me to not introduce SCOPE_EXIT_ALL_TPL and still support this_ (not just this). Thanks. --Lorenzo -- View this message in context: http://boost.2283326.n4.nabble.com/boost-scope-exit-C-11-freely-uses-typena… Sent from the Boost - Dev mailing list archive at Nabble.com.

4 5

[git] Mercurial?
by Daryle Walker 02 Apr '12

02 Apr '12

Git has a competitor called Mercurial? If we're moving to a Distributed-VCS, should we go to Mercurial instead of Git? They're kind-of like CVS vs. Subversion, except I think they came up in parallel. (While Subversion was designed as an updated CVS.) I think Git was made up of a bunch of script hacks, while Mercurial was a regimented single program. I don't have a preference, but I want to make sure we consider the rival options. Daryle W.

48 236

[test] Use of namespace qualifier before floating point functions in boost/test/impl/execution_monitor.hpp
by Jeremiah Willcock 02 Apr '12

02 Apr '12

According to http://pubs.opengroup.org/onlinepubs/009604599/basedefs/fenv.h.html, operations such as feclearexcept can be defined as macros, and at least some versions of Glibc do that by default. The use of explicit :: qualifiers before calls to those in Boost.Test fails to compile. Is there a workaround for that (other than __NO_MATH_INLINES)? -- Jeremiah Willcock

1 1

[asio] ip::tcp::socket thread safety
by Gaetano Mendola 02 Apr '12

02 Apr '12

Hi all, reading the manual I see that ip::tcp::socket thread safety on shared object is not safe. In my application I'm using: ip::tcp::socket::async_connect boost::asio::async_read( ip::tcp::socket , ...) the application queue an async_read or an async_connect inside the receiving handler (in case of error it retry the connection). >From an external thread (to stop the reception) I perform a: ip::tcp::socket::shutdown ip::tcp::socket::close do I have realy to synchronize the async_connect/async_read with shutdown/close ? Using a mutex to protect those calls seems an overkill, may be shutdown and close can be called in concurrency with async_connect/async_read and the documentation is just conservative? Regards Gaetano Mendola

2 1

[quickbook] coloring code syntax
by lcaminiti 02 Apr '12

02 Apr '12

Hello all, Is there a way to tell QuickBook code fragments to color words that are not C++ keyword for example in blue or in other colors? For example, in `int i = 0;` "int" is colored in blue. I'd like that in `requires Copyable<T>` "requires" is colored in blue (or some other color that I can specify). Thanks a lot! --Lorenzo -- View this message in context: http://boost.2283326.n4.nabble.com/boost-quickbook-coloring-code-syntax-tp4… Sent from the Boost - Dev mailing list archive at Nabble.com.

3 2

[result_of] Test fails on clang trunk and gcc-4.7 due to changes in FDIS
by Michel Morin 01 Apr '12

01 Apr '12

In the code below, const int f(); decltype(f()) i = 0; `i` has type `int` (not `const int`) in C++11. This is because `f()` is a prvalue of a non-class type (i.e. `const int`) and so its cv qualifiers are ignored. C++0x drafts (before FDIS) had a special rule for function calls in decltype: if e is a function call or an invocation of an overloaded operator (parentheses around e are ignored), decltype(e) is the return type of the statically chosen function; but this was removed in FDIS. Clang-3.1 (which will be clang-3.1) and gcc-4.7 seem to have implemented this change, and this change breaks the test case (libs/utility/test/result_of_test.cpp) on those compilers in a C++11 mode. The following are the errors messages: [1] result_of_test.cpp:210:3: error: static assertion failed: (is_same<result_of<const result_of_member_function_template(double)>::type, const double>::value) [2] result_of_test.cpp:211:3: error: static assertion failed: (is_same<result_of<volatile result_of_member_function_template(double)>::type, volatile double>::value) [3] result_of_test.cpp:212:3: error: static assertion failed: (is_same<result_of<const volatile result_of_member_function_template(double)>::type, const volatile double>::value) [4] result_of_test.cpp:239:3: error: static assertion failed: (is_same<result_of<const volatile no_result_type_or_result_of(void)>::type, const unsigned short>::value) [5] result_of_test.cpp:244:3: error: static assertion failed: (is_same<result_of<const volatile no_result_type_or_result_of_template<void>(void)>::type, const unsigned short>::value) Regards, Michel

3 10

Status of proposed boost.overload?
by niXman 01 Apr '12

01 Apr '12

Hello list! boost.overload: http://svn.boost.org/svn/boost/sandbox/overload/ The library will be accepted into the boost libraries ? -- Regards, niXman

5 11

[container][memory] Request to move allocator_arg_t and uses_allocator to a memory related file.
by Vicente J. Botet Escriba 31 Mar '12

31 Mar '12

Hi, I want to add Allocators to the packaged_task and promise class of Boost.Thread template <class Allocator> promise(allocator_arg_t, const Allocator& a); template <class F, class Allocator> explicit packaged_task(allocator_arg_t, const Allocator& a, F&& f); I see that Boost.Container has already defined in namespace boost::container and in file boost/container/scoped_allocator.hpp the following struct allocator_arg_t { }; constexpr allocator_arg_t allocator_arg = allocator_arg_t(); template <class T, class Alloc> struct uses_allocator; I would like to shared these to all the Boost libraries. I don't think it is a good idea to make Boost.Thread depend on Boost.Container just for this. Could these declarations be included in a specific file? These declaration are included in <memory> in the C++11. What would be the best library to contain these declarations? What about a new boost/memory.hpp file? What about moving them to the boost or boost::memory namespace? Best, Vicente

2 5