- Boost - lists.boost.org

[contract] concept error messages
by Lorenzo Caminiti 09 Oct '12

09 Oct '12

Hello all, Main purposes of concepts are to (a) document the type requirements and (b) generate useful error messages when a type does not met the requirements. I'd like to discuss how the ideal error messages should look like when a given type does not model the specified concepts. One point of discussion is: In case a type doesn't model a concept, should the compiler generate an hard error or the function/class template should simply be removed from the overload/specialization set? For example, consider: template< typename T> requires EqualityComparable<T> bool equal ( T a, T b ) // #1 { return a == b; // #3 } ... struct x { explicit x(char const*); }; x abc("abc"); equal(abc, abc); // #2 Let's assume we want an hard error then I think ideally the compiler should generate two errors: 1) An error at line #1 indicating that the concept EqualityComparable<T> is not modeled. Ideally this error will read: line #1: failed to model concept EqualityComparable<T> 2) An error at line #2 indicating that the call equal(abc, abc) cannot be resolved. Ideally this error will read: line #2: no matching call for function equal 3) Note that no error should be generated from within the equal function implementation (line #3, etc) because the concept is not modeled so the function definition should not be compiled. Now, this is what I was able to actually implement. === Option A) Using enable_if === template< typename T > typename std::enable_if< AlwaysTrue<T>::value && contract::std::EqualityComparable<T>::value, bool>::type equal1 ( T const& a, T const& b ) { return a == b; } ... equal1(abc, abc); // 08.cpp:44 08.cpp:44:5: error: no matching function for call to 'equal1' equal1(abc, abc); ^~~~~~ [...] This error is not ideal because it only satisfies 2) and 3) but not 1). It does not give information about which concept was not modeled -- you don't see EqualityComparable or "concept failed" anywhere in the error. (Clang actually mentions EqualityComparable in a "note" following the error (see below) but that's not a message that can be relied on across different compilers.) Users can use this approach if they don't want an hard error but they only want to remove the function from the overload set on concept failure. === Option B) Using concept_check === template< typename T > typename contract::concept_check< AlwaysTrue<T>::value, contract::std::EqualityComparable<T>::value, bool>::type equal2 ( T const& a, T const& b ) { return a == b; } ... equal2(abc, abc); // 08.cpp:45 In file included from 08.cpp:5: include/contract/concept_check.hpp:61:5: error: static_assert failed "model concept number 2" static_assert(Concept2, "model concept number 2"); ^ ~~~~~~~~ [...] 08.cpp:45:5: error: no matching function for call to 'equal2' equal2(abc, abc); ^~~~~~ This also satisfies 2) and 3) however it only partially satisfies 1). The error mentions the number of the concept "2" but not its name EqualityComparable (that's probably OK) but the error line number refers to the file implementing concept_check and not to the point of declaration of equal2. In other words, the error for 1) does not indicate line #1 (that's probably acceptable but not ideal). Note that the error for 1) is generate using a static_assert so it is portable. === Option C) Using CONCEPT_CHECK macro === CONTRACT_CONCEPT_CHECK( // 08.cpp:26 (template< typename T >), (AlwaysTrue<T>) (contract::std::EqualityComparable<T>), bool, equal3 ) ( T const& a, T const& b ) { return a == b; } ... equal3(abc, abc); // 08.cpp:45 08.cpp:26:1: error: static_assert failed "model concept number 2: contract::std::EqualityComparable<T>" CONTRACT_CONCEPT_CHECK( ^~~~~~~~~~~~~~~~~~~~~~~ [...] 08.cpp:46:5: error: no matching function for call to 'equal3' equal3(abc, abc); ^~~~~~ [...] These errors satisfy all 1), 2), and 3) however the use of macros makes clang generate lots of "notes" to backtrace the macro expansion (there's probably a way to disable these notes, plus they are not actual errors and can be ignored). Note that the concept name EqualityComparable appears in the static assertion error message so it is guaranteed to be printed portably among compilers. I'm thinking that these are all valid ways to check concepts. Eventually, Boost.Contract macros will use B) (given that the lib already uses macros to declare functions). However, if users want to use Boost.Contract macros to declare a concept but they want to use enable_if or contract::concept_check to check it, they can do so (in case they don't want to use the lib macros to declare their functions). Complete error messages (including "notes"): 08.cpp:44:5: error: no matching function for call to 'equal1' equal1(abc, abc); ^~~~~~ 08.cpp:12:5: note: candidate template ignored: disabled by 'enable_if' [with T = x] AlwaysTrue<T>::value && contract::std::EqualityComparable<T>::value, ^ In file included from 08.cpp:5: include/contract/concept_check.hpp:61:5: error: static_assert failed "model concept number 2" static_assert(Concept2, "model concept number 2"); ^ ~~~~~~~~ 08.cpp:19:20: note: in instantiation of template class 'contract::concept_check<true, false, bool>' requested here typename contract::concept_check< ^ 08.cpp:21:13: note: while substituting deduced template arguments into function template 'equal2' [with T = x] bool>::type equal2 ( T const& a, T const& b ) ^ 08.cpp:45:5: error: no matching function for call to 'equal2' equal2(abc, abc); ^~~~~~ 08.cpp:21:13: note: candidate template ignored: substitution failure [with T = x] bool>::type equal2 ( T const& a, T const& b ) ^ 08.cpp:26:1: error: static_assert failed "model concept number 2: contract::std::EqualityComparable<T>" CONTRACT_CONCEPT_CHECK( ^~~~~~~~~~~~~~~~~~~~~~~ include/contract/concept_check.hpp:40:33: note: expanded from macro 'CONTRACT_CONCEPT_CHECK' BOOST_PP_SEQ_FOR_EACH_I(CONTRACT_CONCEPT_CHECK_ASSERT_, ~, concepts) \ ^ ../../../boost_1_50_0.cyg/boost/preprocessor/seq/for_each_i.hpp:27:69: note: expanded from macro 'BOOST_PP_SEQ_FOR_EACH_I' # define BOOST_PP_SEQ_FOR_EACH_I(macro, data, seq) BOOST_PP_FOR((macro, data, seq (nil), 0), BOOST_PP_SEQ_FOR_EACH_I_P, BOOST_PP_SEQ_FOR_EACH_I_O, BOOST_PP_SEQ_FOR_EACH_I_M) ^ ../../../boost_1_50_0.cyg/boost/preprocessor/repetition/detail/for.hpp:22:78: note: expanded from macro 'BOOST_PP_FOR_1' # define BOOST_PP_FOR_1(s, p, o, m) BOOST_PP_FOR_1_C(BOOST_PP_BOOL(p(2, s)), s, p, o, m) ^ note: (skipping 7 expansions in backtrace; use -fmacro-backtrace-limit=0 to see all) ../../../boost_1_50_0.cyg/boost/preprocessor/seq/for_each_i.hpp:45:80: note: expanded from macro 'BOOST_PP_SEQ_FOR_EACH_I_M_IM' # define BOOST_PP_SEQ_FOR_EACH_I_M_IM(r, im) BOOST_PP_SEQ_FOR_EACH_I_M_I(r, im) ^ ../../../boost_1_50_0.cyg/boost/preprocessor/seq/for_each_i.hpp:50:62: note: expanded from macro 'BOOST_PP_SEQ_FOR_EACH_I_M_I' # define BOOST_PP_SEQ_FOR_EACH_I_M_I(r, macro, data, seq, i) macro(r, data, i, BOOST_PP_SEQ_HEAD(seq)) ^ include/contract/concept_check.hpp:23:5: note: expanded from macro 'CONTRACT_CONCEPT_CHECK_ASSERT_' static_assert(CONTRACT_CONCEPT_CHECK_COND_(i), "model concept number " \ ^ 08.cpp:26:1: note: in instantiation of template class 'contractXconceptXcheckXequal3X30<true, false>' requested here CONTRACT_CONCEPT_CHECK( ^ include/contract/concept_check.hpp:50:14: note: expanded from macro 'CONTRACT_CONCEPT_CHECK' typename CONTRACT_CONCEPT_CHECK_CLASS_(func_name)< \ ^ include/contract/concept_check.hpp:11:23: note: expanded from macro 'CONTRACT_CONCEPT_CHECK_CLASS_' BOOST_PP_SEQ_CAT((contractXconceptXcheckX)(func_name)(X)(__LINE__)) ^ ../../../boost_1_50_0.cyg/boost/preprocessor/seq/cat.hpp:30:7: note: expanded from macro 'BOOST_PP_SEQ_CAT' )(seq) \ ^ note: (skipping 18 expansions in backtrace; use -fmacro-backtrace-limit=0 to see all) ../../../boost_1_50_0.cyg/boost/preprocessor/seq/cat.hpp:35:37: note: expanded from macro 'BOOST_PP_SEQ_CAT_O_I' # define BOOST_PP_SEQ_CAT_O_I(a, b) a ## b ^ <scratch space>:91:1: note: expanded from macro 'contractXconceptXcheckXequal3X' contractXconceptXcheckXequal3X30 ^ ../../../boost_1_50_0.cyg/boost/preprocessor/seq/fold_left.hpp:295:51: note: expanded from macro 'BOOST_PP_SEQ_FOLD_LEFT_F' # define BOOST_PP_SEQ_FOLD_LEFT_F(op, st, ss, sz) st ^ 08.cpp:29:11: note: while substituting deduced template arguments into function template 'equal3' [with T = x] bool, equal3 ^ include/contract/concept_check.hpp:53:5: note: expanded from macro 'CONTRACT_CONCEPT_CHECK' func_name ^ 08.cpp:46:5: error: no matching function for call to 'equal3' equal3(abc, abc); ^~~~~~ 08.cpp:29:11: note: candidate template ignored: substitution failure [with T = x] bool, equal3 ^ include/contract/concept_check.hpp:53:5: note: expanded from macro 'CONTRACT_CONCEPT_CHECK' func_name ^ 5 errors generated. --Lorenzo

5 20

[result_of] BOOST_RESULT_OF_USE_DECLTYPE in a C++03 mode with libc++
by Michel Morin 08 Oct '12

08 Oct '12

This is just a small note related to boost::result_of. When BOOST_RESULT_OF_USE_DECLTYPE is defined, boost::result_of tries to use decltype-based implementation (regardless of the compiler's decltype support). Surprisingly, in a C++03 mode, libc++ defines `decltype` using `__typeof__`: Bug 12488 (WONTFIX): decltype can be used in C++03 with no errors, no warnings http://llvm.org/bugs/show_bug.cgi?id=12488 So, when used with libc++, boost::result_of can be used with BOOST_RESULT_OF_USE_DECLTYPE even in a C++03 mode. Regards, Michel

2 3

[config] libs not supporting std::u16string and std::u32string.
by Vicente J. Botet Escriba 08 Oct '12

08 Oct '12

Hi, compiling Boost.Ratio/io Boost.Chrono/io with compilers supporting |char16_t and |char32_t fails if the library don't provides std::u16string and std::u32string. Could we add a macro |BOOST_NO_CXX11_HDR_STRING||| (c++11 is not completely provided) or the two more specific |||BOOST_NO_CXX11_U16STRING||| and |||BOOST_NO_CXX11_U32STRING||| for these cases? Best, Vicente || ||

4 7

Is Boost Broken?
by Robert Ramey 08 Oct '12

08 Oct '12

At Boost Con 2010 I made a presentation in which I suggested changes in boost to permit it to continue to prosper. This presentation was titled: "Is Boost Broken?" Here's the video link - but it doesn't seem to work any more http://blip.tv/boostcon/robert-ramey-is-boost-broken-4166775 And the slides http://www.rrsd.com/software_development/boost/BoostCon2010/boost_con_2010.… And a thread of the discussion in the aftermath. http://boost.2283326.n4.nabble.com/quot-Is-Boost-Broken-quot-post-script-td… The response was encouraging but sorry to say that note much concrete has come of it ... until now. I've implemented most of these ideas as a website - Boost Library Incubator - www.blincubator.com which is intended to help library authors and provide a place where libraries can be developed prior to the review process. This touches on just about all of Boost workflow and initiatives in one way or another. This includes Modularization, Documentation standards and preperation, Testing and Test Reporting, Deployment. Probably the main underlying ideas here are: a) Decoupling and Modularization of libraries in general and boost in particular. b) Workflow to improve code quality b) leverage on "free" tools, technologies, and facilities to get most bang for the buck. As a "teaser", I've included a new library "safe numerics" which addresses a long running sore in the C language. It's a serious proposal for a library, but my main interest is to have a prototype to develope the next concept. I'm very much tempted to comment on the current state of development of web technologies - the that would take me waaaay of topic and we'll leavel that for another time. Looking forward to feedback from boost enthusiasts. Robert Ramey

3 3

[type_traits] common_type and SFINAE
by Lorenzo Caminiti 08 Oct '12

08 Oct '12

Hello all, common_type<T, U>::type is defined in a way that gives a hard compiler error if T and U do not have a common type (this is also how common_type is implemented in C++11). An alternative design (I'm not sure if implementable) would have been to not define the type member in case T and U have no common type. This way I could use SFINAE to detect if common_type<T, U>::type does not exist and therefore if T and U have not common type -- so to implement N3351's concept Common<T, U>. Is there a way to implement this alternative design? Thanks. --Lorenzo

8 22

[graph] kick out or replace transitive_reduction()?
by Johannes Lorenz 08 Oct '12

08 Oct '12

Hi, shall the transitive_reduction function (transitive_reduction.hpp) of boost_graph ever be used? I ask because * I can find no docs about it * It is unhandy to use, since the nodes are permuted (in opposite to transitive_closure(), which offers a version with *constant* permutation) * I've found bugs in the header, which were stated to be bugs by the original author (asked him via email) If this function should not be used, I'd write a very simple one, using the formula from [1]. Note that this will only work for DAGs, but that's better than nothing. What do you think? Regards, Johannes [1] http://en.wikipedia.org/wiki/Transitive_reduction#Graph_algorithms_for_tran…

1 0

[1.52.0] Beta release candidates available
by Marshall Clow 08 Oct '12

08 Oct '12

Release candidate files for 1.52.0 beta 1 are available at http://boost.cowic.de/rc/ 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. The files (and associated md5s) are: 61a852c923c7e27d84fd2e6336d787f2 *boost_1_52_0_beta1_rc1.tar.gz 956fb476326ee634225c683612493fdd *boost_1_52_0_beta1_rc1.tar.bz2 703663eca3122152031c5555e26845c0 *boost_1_52_0_beta1_rc1.zip b277b6f36ead9df9a0754c868956f30c *boost_1_52_0_beta1_rc1.7z Thanks, --The release managers

5 6

[TTI] library now status?
by osyo manga 07 Oct '12

07 Oct '12

Hi. What is the state of Boost.TTI ? I seem to have not changed for more than one year in boost-svn. Thanks.

2 1

[exception] boost::throw_exception vs deriving from boost::exception
by John Salmon 07 Oct '12

07 Oct '12

I'd like to use Boost.Exception's error_info capabilities in an existing code base. It seems to me that I have two choices: 1 - find all the 'throw my_except(...)' in the code base and replace them with boost::throw_exception(my_except(...)). 2 - find all the struct my_except{...} definitions in the current code base and replace them with: struct my_except: virtual boost::exception {...} I find #2 more appealing for the following practical reasons: - there are fewer locations to change. Exceptions are defined less often and in fewer places than they are used. - there is less to remember. Developers who write new throw expressions need to remember nothing. Developers who define new exceptions must remember to derive them from boost::exception, but this is much less common. Are there reasons I should prefer boost::throw_exception instead? I can only think of one: boost::throw_exception also adds N2179 support via enable_current_exception. This isn't a current concern, but our compilers (gcc and clang) already support N2179 directly, so we can use std::current_exception rather than boost::current_exception if the need arises. Are there other reasons to prefer boost::throw_exception? Thanks John Salmon

3 2

[Phoenix] lambda_tests.test fails on gcc 4.3-4.8 with BOOST_RESULT_OF_USE_DECLTYPE
by Michel Morin 07 Oct '12

07 Oct '12

First, thanks Eric for fixing incompatibility between Phoenix and decltype-based result_of. This is really a great fix! I ran Phoenix test-suite on * clang 3.1, 3.2 (trunk) * gcc 4.4-4.7, 4.8 (experimental) with * C++03, C++11 + BOOST_RESULT_OF_USE_DECLTYPE, C++11 + BOOST_RESULT_OF_USE_TR1 The only failure is lambda_tests.test (i.e. scope/lambda_tests.cpp) on gcc 4.4-4.7 with C++11 + BOOST_RESULT_OF_USE_DECLTYPE. Is this caused by deficiency of N3276 decltype support? P.S. What was the main reason for the incompatibility between Phoenix and decltype-based result_of? And how did you fix it? Regards, Michel

2 4