June 2004 - Boost - lists.preview.boost.org

Re:[boost] Re: Re:A question on serialization library
by Robert Ramey 05 Jun '04

05 Jun '04

Virendra Verma wrote: > Library woundn't care but the user would. For example, how would user > delete or modify the object (pointed by pointer in the class) once > loaded since user has no way of knowing how the object was allocated? My > understanding is that the library creates and owns the object (using some > kind of class factory) when loading from input stream unless user keeps a > common allocator and somehow binds it to library's serializer. > > > If a class has a pointer to an object, how does class know that the > > > pointer was created by library or the class itself before > > > serialization? > > > It doesn't. Why would it need to? > > > Robert Ramey The guiding principle behind saving/loading pointers is that the loaded set of C++ structures is exactly equivalent to the original ones that were saved. So issues regarding pointer ownership don't change. That is, whatever ownership policy that was used when the pointer was saved will be the same one that can be used for the set of C++ structures recovered. The only problem would be where the there are pointers that are not included in the set of serialized data. As long as the serialized set is complete and self-contained there should be no problem. That is, there should be no pointers pointing into the serialized data that are themselves not serialized. Robert Ramey

1 0

boost::ref implicit conversion operator overload problems
by Jonathan Graehl 05 Jun '04

05 Jun '04

If I pass a boost::ref to a template function that calls operator() with one argument, the conversion to T& isn't used to find matching functions. The problem seems to also exist for other overloaded operators. What should I do in this case? I thought the idea was for the ref to be transparent to the template code you pass it to. As a stopgap, I wrote template functions "deref" that must be applied at the calling site (these just cast using unwrap_reference). This code works, but if you remove the "deref", doesn't compile under GCC or visual c++: #include <boost/ref.hpp> #include <cassert> template <class T> inline typename boost::unwrap_reference<T>::type & deref(T& t) { return t; } template <class T> inline const typename boost::unwrap_reference<T>::type & deref(const T& t) { return t; } struct Setter { int i; void operator()(int j) { i=j; } Setter() : i(0) {} Setter(int j) : i(j) {} }; template <class F> void apply1(F f) { deref(f)(1); } int main() { Setter s; apply1(s); assert(s.i == 0); apply1(boost::ref(s)); assert(s.i == 1); }

2 1

bug - function.hpp missing include guard - see sourceforge patch
by Nicolas Desjardins 05 Jun '04

05 Jun '04

Dear Boost Developers, I've submitted a patch to your sourceforge patch section - item # 966723. The patch is also attached to this mail. Summary: function.hpp missing include guard patch function.hpp does not have an #ifndef/#endif include guard. This patch adds #ifndef BOOST_FUNCTION_HPP at the top of the file and #endif at the bottom. Thanks, Nicolas Desjardins Software Developer Orthosoft Inc. *** function.hpp.original 2004-06-04 13:54:25.000000000 -0400 --- function.hpp 2004-06-04 13:53:54.000000000 -0400 *************** *** 10,15 **** --- 10,16 ---- // William Kempf, Jesse Jones and Karl Nelson were all very helpful in the // design of this library. + #ifndef BOOST_FUNCTION_HPP #include <boost/preprocessor/iterate.hpp> #include <boost/detail/workaround.hpp> *************** *** 62,64 **** --- 63,67 ---- # include BOOST_PP_ITERATE() # undef BOOST_PP_ITERATION_PARAMS_1 #endif + + #endif // BOOST_FUNCTION_HPP

2 1

Re:[boost] serialization request
by Robert Ramey 05 Jun '04

05 Jun '04

Russell Balest wrote > My request is support for xml  in the serialization > xml archive. I believe it currently chokes if I add a comment in an xml > archive. > The application is for QA to compose test objects in xml and then > read them into applications with boost serialization. > It would be nice to annotate the test object 'templates' with comments > so QA can know what values are valid. > Does someone know where in the code I can implement this easily. > I just want to ignore the comments. Hmmm - my question is: What program is generating the XML archive in the first place? It its being generated by something other than the serialization library itself, its not going to be easy to generate such an archive that is guaranteed to be compatible with the C++ data structures of the reading program. (The organization of the data in the C++ program is what dicates the XML schema). At best it would be a maintenance pain in the neck to keep the reading program based on serialization in sync with some other program that doesn't use serialization to generate the XML archive. On the other hand, if the original XML archive is generated by the serialization library and just modified with comments, your request is very easy to accommodate. Just create a use the following in your serialize functions: Template<class Archive> Void serialize(Archive, my_class & t){ std::string comment; ar & BOOST_NVP("comment", comment); ar & BOOST_NVP("myclass_data", t);; } In which case your XML output will look like: ... <comment></comment> <myclass_data>..contents of t</myclass_data> ... After the archive is created you can edit the space between <comment> and </comment> with anything you want - it would be just thrown away. Generally I wouldn't encourage the practice of editing an archive but in this circumstance I could see it as an acceptable practice. But then this raises the question if the comments are so important, why are they being thrown them away? Shouldn't they be data read by the program? Robert Ramey

3 2

Re: Boost Mathematicians
by Michael D. Crawford 05 Jun '04

05 Jun '04

I have a bachelor's degree in physics, with the intention to go back to school and get a PhD someday. One thing I've had in mind for quite some time is to improve the state of high-energy physics software. Maybe it's improved since I was last in school, but when I did my senior thesis in 1993 the software I worked with was an awful mess. At the time I became convinced that bogus results were being published in the peer-reviewed journals because of undetected software bugs. Mike -- Michael D. Crawford GoingWare Inc. - Expert Software Development and Consulting http://www.goingware.com crawford(a)goingware.com Tilting at Windmills for a Better Tomorrow. "I give you this one rule of conduct. Do what you will, but speak out always. Be shunned, be hated, be ridiculed, be scared, be in doubt, but don't be gagged." -- John J. Chapman, "Make a Bonfire of Your Reputations" http://www.goingware.com/reputation/

4 3

Filesystem patch
by John Maddock 05 Jun '04

05 Jun '04

Beman, I'm committing a patch (below) for my last filesystem patch, basically I've found another reason why GetFileAttributes can fail when the file does not exist. John. $ cvs diff -u johnmaddock(a)cvs.sourceforge.net's password: cvs diff: Diffing . Index: operations_posix_windows.cpp =================================================================== RCS file: /cvsroot/boost/boost/libs/filesystem/src/operations_posix_windows.cpp, v retrieving revision 1.33 diff -u -r1.33 operations_posix_windows.cpp --- operations_posix_windows.cpp 30 Mar 2004 15:19:58 -0000 1.33 +++ operations_posix_windows.cpp 5 Jun 2004 11:47:23 -0000 @@ -321,10 +321,12 @@ # else if(::GetFileAttributesA( ph.string().c_str() ) == 0xFFFFFFFF) { - if((::GetLastError() == ERROR_FILE_NOT_FOUND) || (::GetLastError() == ERROR_PATH_NOT_FOUND)) + UINT err = ::GetLastError(); + if((err == ERROR_FILE_NOT_FOUND) || (err == ERROR_PATH_NOT_FOUND) || (err == ERROR_INVALID_NAME)) return false; // GetFileAttributes failed because the path does not exist // for any other error we assume the file does exist and fall through, // this may not be the best policy though... (JM 20040330) + return true; } return true; # endif

1 0

Recent changes to minimal.hpp break dynamic_bitset and other libraries
by Gennaro Prota 05 Jun '04

05 Jun '04

Sorry for not investigating this myself, but I really don't have time to. It seems that the recent changes to boost/test/minimal.hpp (line 99, invocation of BOOST_READONLY_PROPERTY) cause troubles on IBM AIX. Many library tests seem to fail for that reason: http://boost.sourceforge.net/regression-logs/cs-AIX.html Genny.

2 1

scoped_ptr<T, D>
by Robert Ramey 05 Jun '04

05 Jun '04

Why isn't there a version of boost::scoped_ptr<T> that takes a deleter? That is, why isn't there scoped_ptr<T, D> ? It would be useful to me. Robert Ramey

3 3

FSM idea injetion
by Alexander Nasonov 05 Jun '04

05 Jun '04

Although my idea is supposed to solve multimethod support for dynamic_any it might be useful in FSM library. What I'm trying to build in last two weeks is some sort of typelists for overloaded function set. Each function is identified by overload_id<N> class with unique N. For example: struct read_number_transitions { // Signatures has the form NewState (OldState, Char) Positive operator()(overload_id<0>, StartState, Plus) const; Negative operator()(overload_id<1>, StartState, Minus) const; Positive operator()(overload_id<2>, StartState, Digit) const; Positive operator()(overload_id<3>, Positive, Digit) const; Negative operator()(overload_id<4>, Negative, Digit) const; static std::size_t const max_overload_id = 4; }; This approach has advantage over typelists in some cases because things are right in their places. In the example, every transition, on one hand, is represented by a call operator (this is natural, because transition is a function that takes old state and a char and returns a new state); on the other hand, argument types and return type (a signature) can be retrieved from the call operator. The retrieval is a bit problematic but I'm working on this. The problem is a lack of typeof in C++. If it were available you could write: template<int N, class C, class R, class T> mpl::identity<R(T)> tester(R (C::*)(overload_id<N>, T) const); typedef typeof(tester<0>(&read_number_transitions::operator()))::type sig0; Without typeof you know signature R(T) only inside tester<0>. This means in particular, that, in order to accumulate all signatures you have to call tester<N> recursively: call tester<0> => Positive(StartState, Plus) call tester<1> => Negative(StartState, Minus) call tester<2> => Positive (StartState, Digit) call tester<3> => Positive (Positive, Digit) call tester<4> => Negative (Negative, Digit) For other compile-time algorithms we need something similar. Even worse, often we need chain of algorithms, for example: find a first occurrence of StartState as a first argument, get a return type from that signature, find that type among first arguments and so on. We have to go deeper and deeper. In my stress tests I use 1000 functions. Can you imagine a call stack as deep as 1000 multiplied by a length of algorithms chain? Fortunately, sometimes it's possible to make it plain. 1. Predicates have boolean result, therefore, we can use sizeof/yes_type/no_type to calculate their result: template<class Predicate, int N, class C, class R, class T> typename yes_type_if< Predicate::template apply<R(T)>::type // mpl::apply style >::type tester(R (C::*)(overload_id<N>, T) const); // Example bool const result = sizeof( tester<is_same<Positive(Positive,Digit), _1>, 3>(&read_number_transitions) ) == sizeof(yes_type); I already implemented several variants of *contains* algorithm. You can take a look at g++ 3.4.0 performance here: http://cpp-experiment.sourceforge.net/overloads_contains.png This is a performance plot (in seconds) of various strategies for *contains* algorithm which is applied to a function set of length 1000. Worst case takes 80+ seconds to compile for *one-at-once linear* basic strategy while optimized strategy completes its job in 12 seconds. It's really fast. I doubt that typelist of length 1000 can ever exist without compiler panic ;-) Similar work can be done for find/find_if // returns bool contains<read_number_transitions, Negative(Negative,Digit)>::value; // returns index of Negative(Negative,Digit) find<read_number_transitions, Negative(Negative,Digit)>::value; // returns -1 because int isn't even a signature find_if<read_number_transitions,is_same<int, _1> >::value; 2. Pair comparisons. It's not yet done. The sizeof trick can be modified slightly: template<class Predicate, int N, int M, class C, class R1, class R2, class T1, class T2> typename yes_type_if< Predicate::template apply<R1(T1), R2(T2)>::type >::type tester( R1 (C::*)(overload_id<N>, T1) const, R2 (C::*)(overload_id<M>, T2) const ); Suppose, you have this cool type: typedef is_base_and_derived< function_arg<1, _1>, // first argument of signature _1 function_arg<1, _2> // first argument of signature _2 > cool; Then you can check whether a first argument of overload N is a base of a first argument of overload M: bool const result = sizeof( tester<cool,N,M>(&read_number_transitions, &read_number_transitions) ) == 1; -- Alexander Nasonov

2 3

default version of STLport
by Gennadiy Rozental 05 Jun '04

05 Jun '04

Why don't we switch to 4.6.2 as default version of STLport? Which version is used in regression tests? Gennadiy.

1 0