November 2011 - Boost - lists.preview.boost.org

Boost.Random.uniform_01 gives different number sequence in Boost 1.37 and Boost 1.45
by Martin.Sovik＠Emerson.com 16 Nov '11

16 Nov '11

Hi. It seems that uniform_01 generates different number sequences in version 1.37 and 1.45 Is this intentional, or is it a bug? This short program #include <iostream> #include "boost/random/mersenne_twister.hpp" #include "boost/random/uniform_01.hpp" int main() { boost::mt19937 random_generator(4315); boost::uniform_01< boost::mt19937, double > uniform_generator(random_generator); std::cout<<"Generating uniform distribution using Boost 1.45"<<std::endl; for (int i=0; i<10;++i) { std::cout<<uniform_generator()<<std::endl; } return 0; } Compiled with the two versions, gives the respective output: D:\dev\test_bed>uniform_dist_boost_1_37 Generating uniform distribution using Boost 1.37 0.239136 0.920749 0.889477 0.208172 0.169688 0.636659 0.78322 0.915117 0.25697 0.358218 D:\dev\\test_bed>uniform_dist_boost_1_45 Generating uniform distribution using Boost 1.45 0.306389 0.511291 0.496429 0.156168 0.910315 0.644836 0.331514 0.558632 0.723934 0.998796 As far as I can tell, Boost.mt19937 gives the same number sequence in both cases, but the output from Boost.uniform_01 differs. This topic has been briefly touched in the thread [boost] Random numbers question<154235.php> Anant Rao (2009-07-24 18:47:31) Where it seems like Rao got different numbers even between Boost 1.38 and Boost 1.39 We use Boost 1.37 in the previous release of our software, and Boost 1.45 in the current, and we have received a bug report from one of our clients that our current release has different behaviour. Is there any way to fix this, other than use the 1.37 version of Boost.uniform_01? Thanks, Martin Søvik

2 1

Re: [boost] [1.48.0] Release candidates available
by Tom Kent 16 Nov '11

16 Nov '11

> Release candidate files for 1.48.0 are available at http://boost.cowic.de/rc/ > > md5sum on the originals: > > 7dd95e71c43ddd35997d9f2346e7f567 *boost_1_48_0.7z > d1e9a7a7f532bb031a3c175d86688d95 *boost_1_48_0.tar.bz2 > 313a11e97eb56eb7efd18325354631be *boost_1_48_0.tar.gz > b08fda829eec96b4f1071ce2ea6831f5 *boost_1_48_0.zip > > As always, the release managers would appreciate it if you download > the candidate of your choice and give building it a try. Please report > both success and failure, and anything else that is noteworthy. > > This helps ensure the candidates build OK before we push them out to > SourceForge. > > Thanks, > > --The release managers On a windows 7 - 32 bit VM, I built: msvc-8.0 msvc-8.0 - address-model=64 msvc-9.0 msvc-9.0 - address-model=64 msvc-10.0 msvc-10.0 - address-model=64 All builds were successful. I was able to smoothly link and run with the binaries on windows 7 32 or 64 (depending on the library build). Tom Kent

1 0

[program_options] Improving error text in exceptions
by Leo Goodstadt 16 Nov '11

16 Nov '11

On 15 November 2011 09:08, Vladimir Prus <vladimir(a)codesourcery.com wrote: > > Leo Goodstadt wrote: > > I would very much like to make the error messages in program_options > > exceptions more consistent and informative. > > I've looked over the error messages you have proposed, and I think it's > an overall improvement. I would appreciate patch to make this change. > Thanks! Cool. Will do that. The error messages meant for programmers I shall leave alone. This exercise is so tedious that I doubt anyone else is going to do more than tinker with the message text so I am keen to do an exhaustive job. As I mentioned, for command line messages, <option '--foo_bar' is ambiguous> seems much clearer than than <option 'foo_bar' is ambiguous>. However, it would be inappropriate to add "--" to config file and dos style options. The easiest thing would be to add a "prefix" parameter to the set_option_name() functions in errors.hpp These are called inside store() in variables_map.cpp. To add "--" or "/" prefixes as appropriate, is it sufficient to look at basic_option::original_tokens and basic_option::string_key? A) Use no prefix if original_tokens[0] is unprefixed, B) Use the parsed prefix (original_tokens[0]) if basic_option::string_key.length() = 1 C) Prefer "--" for long option names The same logic would apply to 1) exceptions thrown from cmdline::finish_option() in cmdline.cpp 2) options_description::find() (in options_description.cpp) called from store() (in variables_map.cpp) 3) options_description::find_nothrow() (in options_description.cpp) called from cmdline::run(), cmdline::finish_option(), parse_short_option(), and parse_disguised_long_option() (in cmdline.cpp) (Despite its name, find_nothrow() can throw()!) The difficult case is supporting the correct prefix for "required_option" in variables_map::notify() (variables_map.cpp). I propose that where missing options can be specified in both configuration file and on the command line, the command line should take precedence, and "--option_name" displayed in the error message. To support this, "basic_parsed_options" needs an extra prefix field, which would be set appropriately by basic_command_line_parser<charT>::run(). Does this make sense? Thanks, Leo -- Leo Goodstadt University of Oxford, UK

2 1

[1.48.0] Release candidates available
by Beman Dawes 16 Nov '11

16 Nov '11

Release candidate files for 1.48.0 are available at http://boost.cowic.de/rc/ md5sum on the originals: 7dd95e71c43ddd35997d9f2346e7f567 *boost_1_48_0.7z d1e9a7a7f532bb031a3c175d86688d95 *boost_1_48_0.tar.bz2 313a11e97eb56eb7efd18325354631be *boost_1_48_0.tar.gz b08fda829eec96b4f1071ce2ea6831f5 *boost_1_48_0.zip As always, the release managers would appreciate it if you download the candidate of your choice and give building it a try. Please report both success and failure, and anything else that is noteworthy. This helps ensure the candidates build OK before we push them out to SourceForge. Thanks, --The release managers

13 17

Re: [boost] Boost.Local Review - comments by Andrzej
by Andrzej Krzemienski 15 Nov '11

15 Nov '11

> > One suggestion for the future, although I am not sure how useful it would > > turn out in practice is to have something like local expression, or even > > not necessarily a local one. This would be a function whose body consists > > of a single expression. The benefit you get is that the programmer does not > > need to specify a return type for it (you can use BOOST_TYPEOF for that), > > or doesn't even have to type keyword 'return', but this may be too much off > > topic. > Can you clarify with an example? In C++11 lambdas do not require specifying the return type, provided that the body of the function is a single return statement: because it is easy to deduce the type from such single return statement. Similarly, for Boost.Local if a function body consists of a single return statement you could (I guess) provide a more simple syntax. Instead of BOOST_TYPEOF(sum) & BOOST_LOCAL_FUNCTION_PARAMS( double num, const bind factor, bind& sum ) { return sum += factor * num; } BOOST_LOCAL_FUNCTION_NAME(add); I could write: BOOST_LOCAL_EXPRESSION_PARAMS( double num, const bind factor, bind& sum ) BOOST_LOCAL_EXPRESSION_END( add, sum += factor * num ); You save me typing the return type and 'return' keyword. But now that I have written it down I do not think it is particularly useful. > I think in a nutshell the conclusion (which I will detail in the rest > of my comments below) was that Boost.Local is handy but on compilers > with C++11 lambdas you'd just use lambdas. At the same time, > Boost.Local allows you to write local functions that work also on > C++03 compilers and that have the same compile-time and run-time > performances of lambdas on C++11 compilers. So the /main/ audience > would be someone (like myself) that wants to do local function stuff > and expects lambda-like performances on C++11 but also has to deal > with C++03 compilers. There are other benefits like const& binding > which is not supported by C++11 lambdas (so how can I bind by const a > variable that has a very expensive copy constructor using C++11 > lambdas?) and improve scope exit (i.e., local exit) but these other > benefits are probably minor in the general domain (they can be very > important or even essential is a specific and narrow domain like > Contract Programming but not critical in the general case). OK Lorenzo, I am convinced. Can we trade? I will vote 'yes', and you will add a motivation section in the documentation which will stress that the primary motivation of the library is to provide a uniform (in terms of syntax) support for local functions: fast and 'native' for C++11 compilers, and acceptable (using tricks) in older compilers. The library can be used directly or be a foundation for higher level libraries like Boost.Concept. Review Managers, I am not really trading, I vote YES for the Boost.Local and just request the motivation section in the documentation. I think the situation of Boost.Local is similar to that of Boost.Move, and the latter was accepted into Boost. Also, having a library in Boost whose primary purpose is to serve as the foundation for other Boost libraries is not a precedent (I mean Boost.Proto and now Boost.Context). Regards, &rzej

3 2

Re: [boost] Boost.Local Review - comments by Andrzej
by Andrzej Krzemienski 15 Nov '11

15 Nov '11

Just one more thought. I was not participating in the design discussion for Boost.Local when it was probably a better time to mention this. So I am just mentioning it for reference. Functions that take parameters and bind existing objects are called closures: http://en.wikipedia.org/wiki/Closure_%28computer_science%29.<http://en.wikipedia.org/wiki/Closure_%28computer_science%29> This could have been an alternative name for BOOST_LOCAL_FUNCTION: BOOST_SCOPE_CLOSURE... Regards, &rzej

2 1

[1.48.0] Status?
by Beman Dawes 15 Nov '11

15 Nov '11

Are there any 1.48.0 showstoppers outstanding? We are scheduled to release right now. There are still a lot of release branch regression test failures. It is very uncomfortable to see failures for popular compilers such as VC++ 10 and GCC: accumulators (most or all compilers) assign flyweight (most or all compilers) graph ublas pool (all Windows compilers) ptr_container (most or all compilers) python spirit/repository (most or all compilers) tr1 (most or all compilers) wave (all Windows compilers) See http://beta.boost.org/development/tests/release/developer/summary.html Do we really want to release with this many failures? Worried, --Beman

12 33

[property_tree] Assign and Lua parser
by Renato Florentino Garcia 15 Nov '11

15 Nov '11

In my use of Boost Property Tree library I have coded some extensions, which are in a GitHub repository (http://github.com/renatoGarcia/boost-extensions) I would like to know if there are any interest in make any of then part of Boost Property Tree itself, probably after a suggest improvement. The more useful in my view is a Boost Assign Property Tree alike. Code: http://github.com/renatoGarcia/boost-extensions/blob/master/property_tree/i… For instance, to build in C++ code a ptree like: key1 value1 key2 { key3 12345 { key4 "value4 with spaces" } key5 value5 "" 10 "" 11 "" 12 } Instead the current: ptree pt; pt.add("key1", "value1"); ptree tmpPt1; tmpPt1.add("key3", 12345); ptree tmpPt2; tmpPt2.add("key4", "value4 with spaces"); tmpPt1.push_back(std::make_pair("", tmpPt2)); tmpPt1.add("key5", "value5"); ptree tmpPt3; tmpPt3.put_value(10); tmpPt1.push_back(std::make_pair("", tmpPt3)); ptree tmpPt4; tmpPt4.put_value(11); tmpPt1.push_back(std::make_pair("", tmpPt4)); ptree tmpPt5; tmpPt5.put_value(12); tmpPt1.push_back(std::make_pair("", tmpPt5)); pt.add_child("key2", tmpPt1); Would be a code like: ptree pt = tree() ("key1", "value1") ("key2", tree() ("key3", 12345) (tree() ("key4", "value4 with spaces")) ("key5", "value5") (10)(11)(12)); Another extension is a Lua parser, like the XML or JSON. But in this case I don't think that is desirable include it in the Boost, because of the needed of Lua libraries (at least not in a mandatory way). Code: http://github.com/renatoGarcia/boost-extensions/blob/master/property_tree/i… Regards, Renato Garcia

2 1

Boost.Local Review - comments by Andrzej
by Andrzej Krzemienski 15 Nov '11

15 Nov '11

Hi Lorenzo, I would like to provide some comments on Boost.Local. > Please state clearly whether you think this library should be accepted > as a Boost library. I am unable to give a yes/no answer. This is mostly because I am not clear who is the intended audience of the library. I explain this in detail below. I am impressed with the library documentation. It is very informative with lots of examples, benchmarks, comparisons, rationale, and layered design (overview, and then details). This is how a Boost library's documentation (and every documentation, IMO) should look like. Yet, one thing that I am missing is what practical problems (that are not already solved) this library is trying to solve. I am not saying that it doesn't solve any. I am just saying I cannot figure it out. A typical story for a Boost library is that there is a common problem that many developers face fairly often, and at some point someone decides to fix this problem comprehensively and ultimately. This is the impression that I had, for example, with Boost.Optional. When I read about it I thought "Yes! This is what I was missing for a long time". I cannot see this in Boost.Local. Perhaps I am not the intended audience of the library, but it would help me if it was clear who the audience is. (I provide examples of what I mean below.) Documentation clearly separates three parts: local functions, local scopes and local exits. Also my impression is different on the three parts. This suggests to me that they could be three separate libraries. What do they have in common? The syntax, I guess. And the fact that you declare them inside functions. But is that enough to bundle them together? Consequently, I divide my comments into three parts. LOCAL FUNCTIONS Why would I want to use it? Personally, I have one compiler that does support local classes as template parameters. All my other compilers support lambdas. So I guess local functions are not for me. So who will use them? Someone that does not have too new compiler (because they have superior - in terms of performance - alternatives: lambdas, local classes). Someone that does not have a too old compiler, because we read in the documentation that "The implementation of this library uses preprocessor and template metaprogramming (as supported by Boost.Preprocessor<http://www.boost.org/doc/libs/release/libs/preprocessor/doc/index.html>and Boost.MPL <http://www.boost.org/doc/libs/release/libs/mpl/doc/index.html>), templates with partial specializations and function pointers (similarly to Boost.Function<http://www.boost.org/doc/libs/release/doc/html/function.html>). As a consequence, this library is fairly demanding on compilers' compliance with the C++03 <http://www.open-std.org/JTC1/SC22/WG21/docs/standards>standard." Further someone who does not mind compile-time and run-time performance overhead. And someone,who prefers the new syntax to creating a functor class non-locally. I am not sure how many users that would be? The documentation mentions proposal N2511<http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2511.html>, which says that even though with lambdas and local classes we already have a good tool, we still want to have a normal function syntax for local functions. Boost.LocalFunction does not enable a normal function syntax either, so it is no better an alternative than lambdas or local classes. Except for "const binding" which I find hard to appreciate. And except for compilers that don't support C++11 features, but than again, working only with the new compilers I may not be the right person to review the library. Is this supposed to provide a seamless migration of the code from compilers w/o lambdas to those that do support lambdas? I am not comfortable with the syntax: we first specify parameters, then function body and last, we specify the function name. I am not comfortable that the other Lorenzo's library (that I await impatiently) - Boost.Contract - will provide a different function syntax than Local Functions. This will render the situation where every library will require of me to learn different function declaration syntax. On the other hand, the documentation clearly says why this syntax is necessary, so I am not saying "change the syntax". I am suspicious about the local function overloading. While I can imagine writing a local function; writing a bunch of them inside another function would be likely be some error in my design. LOCAL BLOCKS Again, why would I need them? The example with assertion does not convince me. That is, I face the practical tradeoff: additional enforcement on assertions at the cost of slower build, slower execution time and less readable syntax. I will choose not to use the blocks. The only application that I can think of is a "temporary usage" while refactoring. If I work with the code where functions are 1000-line-long, I would like to divide them into blocks and check which parts reference what variables. In this case, as one of the footnotes in documentation suggests, allowing "return;" inside blocks would be a dangerous thing. But again, if I had a luxury of lambdas, I would choose lambdas for that purpose, because they allow specifying the default binding ( [&], [=] ). It looks like local blocks are only a syntax sugar over local functions. Am I right? LOCAL EXITS While exception specifications for locals blocks seam strange, I find allowing exception specifications for local exits an error. Local exits should be subject to the same restrictions as destructors (they are implemented as destructor calls anyways): you must not throw from them. And I expect the library documentation to say it explicitly. I find this "warning" missing from Boost.ScopeExit. So I suggest to disallow (if possible) specifying exceptions in local exit, and adding a section in documentation that warns against potential throws from an exit. I do not see anything against having two libraries for the same purpose. This is pretty much the same what we already have in Boost in case of state machines. But since we have two, it is useful to ask why I would choose local exit over scope exit. My answer is: 1. Local Exit allows me to specify an empty parameter list (I cannot do it with ScopeExit). - this is a practical problem I faced 2. I can bind to "this". I do not find const binding a practical advantage. Typically for short functions it is hardly a real problem that I confuse = with ==. And exits would typically be one-liners in my code. On the other hand there are reasons why I would prefer ScopeExit to LocalExit: it spells shorter. In LocalExit I have to type word 'bind' for every single parameter even though it is obvious I need to bind them all (because scope guards do not accept normal parameters). The same argument goes for blocks, obviously. I also like ScopeExit's potential to be implemented with lambdas, which might result in performance benefits, and allow the 'default binding'. On the other hand LocalExit cannot be implemented with a lambda, if not for any other reason, then due to the const binding. It is really nice that i can bind to variable of the name 'bind'; although I cannot use type 'bind' in local functions. GENERAL THOUGHTS I am really impressed with the usage of macros. It looks like thanks to Lorenzo we are reaching a new level of preprocessor programming. Who knows how far we can go. I do not think they are overused, given the difficult task. On the other hand, error messages that I encountered give me no clue on what I have typed wrong. Decsriptions in the documentation, and the rationale clearly indicates the very detailed knowledge of many C++ aspects. One suggestion for the future, although I am not sure how useful it would turn out in practice is to have something like local expression, or even not necessarily a local one. This would be a function whose body consists of a single expression. The benefit you get is that the programmer does not need to specify a return type for it (you can use BOOST_TYPEOF for that), or doesn't even have to type keyword 'return', but this may be too much off topic. Term 'C99 <http://www.open-std.org/jtc1/sc22/wg14/www/projects#9899> and later compilers' may be a bit of an overuse for C++ compilers. I suggest 'compilers supporting variadic macros'. Also, in reply to particular review questions: > What is your evaluation of the design? It is clear. Some macros are difficult to read, e.g. Function arguments first, function name second, but I guess this is a necessity given the domain. > What is your evaluation of the implementation? The impressive macro magic is beyond my cognitive capabilities > What is your evaluation of the documentation? It is very good. It meets Boost's and my personal criteria for a good documentation. > What is your evaluation of the potential usefulness of the library? > Are you knowledgeable about the problem domain? This is the part where I have problems. I am not sure about the scope and intended audience of the library. > Did you try to use the library? With what compiler? Did you have any problems? I tried to use the library (basic examples) with VC++8 and VC++10 and Boost 1.42. I found that it does not compile if I use compiler option /ZI (Debug Information Format: Program Database for Edit & Continue) on either compiler. > How much effort did you put into your evaluation? A glance? A quick reading? In-depth study? I spent one evening on reading the documentation and playing with the basic examples. For the last remark, I believe that the criterion for accepting or rejecting a library is that if it addresses a real-life problems and if it is well designed. Boost.Local is well designed, but the intended audience and primary motivation is not clearly stated, so I find it hard to assess if the first criterion is met. All in all, Boost.Local is an impressive piece of work. Regards, &rzej

2 1

Determining interest: C++11 parser generator library
by Gene Bushuyev 15 Nov '11

15 Nov '11

Sorry if it turns out to be a duplicate, it looks my original post was lost in cyberspace, so I'm re-posting this request. I'm trying to determine if there is a sufficient interest for including AXE C++11 recursive descent parser generator library in Boost. The zipped sources and documentation are here: http://www.gbresearch.com/axe/axe.zip - Gene Bushuyev

4 7