April 2012 - Boost - lists.boost.org

Re: [boost] C++11 Metaprogramming
by Andrzej Krzemienski 06 Apr '12

06 Apr '12

> I am on the C++Now 2012 schedule giving a talk on metaprogramming in > C++11, which is really just supposed to be an overview of the state of > the art. I am just at the beginnings of my research for this > presentation, having learned a few things and done a few experiments, > and it seemed to me foolish not to ask the Boost community for its > insights. I'm sure y'all have come up with many neat tricks and > techniques. If you'd care to share them here, that would be much > appreciated. For the end users of meta-programming library (at least those interested in type transformations), I believe that alias templates offer a significant improvement to the interface: http://akrzemi1.wordpress.com/2012/03/19/meta-functions-in-c11/ Regards, &rzej

4 5

Belgium C++ Users Group
by Francisco Almeida 05 Apr '12

05 Apr '12

Hello to all, I just subscribed to this mailing list to ask the Boost developer community for anyone actively working on new libraries/library refactoring if they would be interested in presenting to us a lecture about the major changes Boost will undergo, now that the new C++ standard has been published and that support for it is quickly being implemented by the most popular compilers. Of course, assuming any generous Boost developer who happens to live/work in Benelux would be willing to visit us and share his/her experiences/ambitions with us. Best regards, Francisco Almeida

1 0

[type_traits] is_scalar - fit into register?
by Oliver Kowalke 05 Apr '12

05 Apr '12

Hi, is it a correct assumtion that all types which evaluate 'is_scalar<T>::value' to 'true' will fit into a processor register? Oliver

2 1

[math] Explicit message for configure checks
by Steven Watanabe 05 Apr '12

05 Apr '12

AMDG The attached patch just makes the output a little nicer. Okay to commit? In Christ, Steven Watanabe

2 1

boost::indeterminate and std::vector::emplace_back()
by Arno Schödl 04 Apr '12

04 Apr '12

Hi, there is a problem when using boost::indeterminate in conjunction with std::vector::emplace_back: std::vector<boost::tribool> v; v.emplace_back(boost::indeterminate); In VS2010, this code causes error C2383: '_Valty' : default-arguments are not allowed on this symbol The error occurs becasue emplace_back's _Valty template parameter is deduced as the type of boost::indeterminate which is a function with a default argument. What is the recommended way to explicitly create an "indeterminate" tribool value in cases like this? Regards, Arno -- Dr. Arno Schödl | aschoedl(a)think-cell.com Technical Director think-cell Software GmbH | Chausseestr. 8/E | 10115 Berlin | Germany http://www.think-cell.com | phone +49 30 666473-10 | US phone +1 800 891 8091 Amtsgericht Berlin-Charlottenburg, HRB 85229 | European Union VAT Id DE813474306 Directors: Dr. Markus Hannebauer, Dr. Arno Schoedl

2 1

[signals] [vc11] Compile Error (Ticket #1499)
by Joshua Boyce 04 Apr '12

04 Apr '12

The problem causing ticket #1499 is happening again under VC11. https://svn.boost.org/trac/boost/ticket/1499 Build scripts and logs: http://dl.dropbox.com/u/74751/cpp/boost/logs/msvc32-20120319-1118.bat http://dl.dropbox.com/u/74751/cpp/boost/logs/msvc64-20120319-1118.bat http://dl.dropbox.com/u/74751/cpp/boost/logs/msvc32-20120319-1118.txt http://dl.dropbox.com/u/74751/cpp/boost/logs/msvc64-20120319-1118.txt Bumping to check for _MSC_VER <= 1700 does the trick (tested locally). Patch: http://dl.dropbox.com/u/74751/cpp/boost/patches/signals-vc11-fix-20120319-1… Once this is applied the trunk compiles cleanly (well, no errors, there are still lots of warnings) using the VC11 beta. Thanks.

2 3

[propertymap] boost::function as map
by Frank Birbacher 03 Apr '12

03 Apr '12

-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi! I've started working on supporting functors as property maps. I've just seen the commit r77506, with msg "Added function_property_map based on code from Philipp Moeller." So I guess I can stop working on that feature? Maybe regarding a boost::function as a property map all by itself (without make_function_property_map) is a good idea? My work is in ^/boost/sandbox-branches/birbacher/propertymap-functormap, BTW. Frank -----BEGIN PGP SIGNATURE----- Version: GnuPG/MacGPG2 v2.0.17 (Darwin) Comment: GPGTools - http://gpgtools.org Comment: keyserver x-hkp://pool.sks-keyservers.net iEYEARECAAYFAk97a0oACgkQhAOUmAZhnmo/qgCcDv19D3TfjW0nLb0hiykSwrzL yfQAnjvTSJv3MZMgBCFS1mGJQKFfDBYO =HCjT -----END PGP SIGNATURE-----

2 1

[preprocessor] PUSH_BACK/_FRONT on empty array broken in 1.49
by Jan Boehme 03 Apr '12

03 Apr '12

Hi! BOOST_PP_ARRAY_PUSH_BACK & BOOST_PP_ARRAY_PUSH_FRONT on empty array leads to invalid code if Boost 1.49 is used. In 1.48 and older it works as expected. An incident was already filed and a sample attached: https://svn.boost.org/trac/boost/ticket/6760 Thanks for solving. Let me know if I can be of further assist. Jan

2 2

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