September 2008 - Boost - lists.preview.boost.org

[intrusive] making a list safe for concurrent readers
by Michael Marcin 17 Sep '08

17 Sep '08

I have a program that protects an object containing a linked with what is essentially a shared_mutex. I'd like to take advantage of the logical constness of traversing a linked list with const iterators and allow concurrent traversal protected by the external shared_mutex. Unfortunately the standard makes no such guarantees on std::list so I've determined I'll have to write my own linked list with a slightly stronger guarantee that concurrent read access is safe. I remember reading during the review that intrusive could be used to make high quality non-intrusive data structures easily. Can I use intrusive to make a std::list like container that is safe to read concurrently? Thanks, Michael Marcin

3 3

[ATTN Beman] request permission to merge Proto to release
by Eric Niebler 17 Sep '08

17 Sep '08

I'd like to merge Proto from trunk to release. The API is stable, the tests are now passing on most toolsets (though it's hard to see from the test results at the moment because of a Boost.Test bug), and much of the review feedback about the docs has been accommodated. I still plan to make doc improvements before 1.37, but I don't think that should hold up the show. OK to proceed? -- Eric Niebler BoostPro Computing http://www.boostpro.com

6 11

Re: [boost] Answers re BENUM library for automated enum streaming
by Tony D 16 Sep '08

16 Sep '08

From: David Abrahams <dave(a)boostpro.com> > > [Iony wrote]: > > [BENUM does need variadic macros] So, limited portability - > > unless your preprocessor library has some relevant workarounds? > I don't have a preprocessor library. For C98 compatibility > Boost.Preprocessor would probably support > > BENUM(X, (A = 1)(B)(C = 22)(D = B|N)); Sorry - had it in my head that Boost.Preprocessor was one of yours... apologies to Vesa Karvonen and Paul Mensonides. Anyway, good to know some possibility exists, though half the appeal of BENUM was that the notation (sans markup) resembled normal enums, facilitating uptake and retrofitting, so not sure if it's worth complicating the implementation to have the notation above available on a proportion of compilers. Happy to hear any opinions, and will look into it if I have spare time (my introspective pre-processor project takes priority unless there's significant demand). Regards, Tony

1 0

Re: [boost] C++ Builder 2009 and Boost compliance
by Nicola Musatti 16 Sep '08

16 Sep '08

siliconman wrote: [...] > Yeah, the new version of C++Builder includes a custom patched version > of boost that is tweaked so that spirit and serialization will build. I > saw that Nicola has started the process of moving the custom patches > into the main boost trunk, so this should start to get better results > soon. Yep. As Boost.Config is mostly patched it should be possible to run meaningful regression tests now, even though results are not going to look encouraging until more of the patches get committed. Cheers, Nicola -- Nicola.Musatti <at> gmail <dot> com Home: http://nicola.musatti.googlepages.com/home Blog: http://wthwdik.wordpress.com/

1 0

[move] Moving the move headers into boost/detail
by Daniel James 16 Sep '08

16 Sep '08

Hello all, A cut down version of the move library in sandbox was in 'boost/unordered/detail' in 1.36 and is pretty stable. I'd like to move it into 'boost/detail' so that other libraries can use it, and probably add support for more object types as well. The full library should be put up for review eventually, but until then it's a useful implementation detail. Or should I leave it in unordered for now? There's nothing stopping anyone from including it directly. I'd also like to add more tests, if it's in detail, would it be appropriate to add them to the unordered library's tests? Or somewhere else? Also, what should I do for documentation? There's some written (thanks to its Adobe roots, so anyone who noticed that the unordered review mainly consisted of editting the documentation can breath a sigh of relief) so it'd be a shame to not make it available. Maybe I should put it up outside of boost? Obviously I'll make it clear that since the library is in detail, the interface should not be considered stable. Buyer beware. Daniel

1 0

Re: [boost] C++ Builder 2009 and Boost compliance
by Peter Dimov 16 Sep '08

16 Sep '08

siliconman: > In article <00c101c91801$b08eeb00$6507a80a@pdimov2>, > "Peter Dimov" <pdimov(a)pdimov.com> wrote: > >> Do you have an idea why your 5.6.4 and 5.8.2 tests fail (apparently at >> random) with errors like: >> >> Fatal: Unable to open file 'turboc.$ln' > > It seems that the compiler creates these temporary files and if two > instances of the compiler are working at the same time (I'm currently > testing on a quad core machine) in the same directory that those > versions of the compiler isn't able to handle this situation. I'm > actually retooling my build systems, so I was going to pass this task > back to a single core machine even thought that would mean longer turn > around time. > Alternatively, I could have two machines running tests and have > siliconman-legacy with 5.64 and 5.82 on the single core machine and > siliconman with 5.9.3 and 6.10 with the faster machine. A quick temporary fix - until one of your permanent measures are put in place - might be to just disable one of 5.64 and 5.82; that way we'll at least have one set of reliable results instead of two that are unfortunately of little use in their present state.

1 0

Re: [boost] Answers re BENUM library for automated enum streaming
by Tony D 16 Sep '08

16 Sep '08

From: David Abrahams <dave(a)boostpro.com> > [I/Tony wrote]: > > BENUM(X, A = 1, B, C = 22, D = B|N); > > This interface requires variadic (C99) preprocessor support, does it not? Quite so David. So, limited portability - unless your preprocessor library has some relevant workarounds? Cheers, Tony

3 3

Re: [boost] Answers re BENUM library for automated enum streaming
by Tony D 16 Sep '08

16 Sep '08

> From Manuel Fiorelli: > You could try to look at [boost preprocessor lib docs] to avoid variadic macros. > I am not sure if it is applicable at your library, though. As best I understand it, they're not applicable :-(. Perhaps David will confirm. Also appreciate if someone can fill me in on boost's (historical?) attitude to facilities with this dependency. FWIW, http://en.wikipedia.org/wiki/Variadic_macro lists compilers with __VA_ARGS__ support (GCC 3.0+, VC++2005+, C++Builder 2006+). > From: Marcus Lindblom <macke(a)yar.nu> > However, is it possible to define a separate mapping if one would like? > > For instance, I like my strings to be lower case, and without prefixes, > whereas my enums need a bit more scaffolding: > > enum Context { CXT_APP, CXT_PHYS, CXT_VIS } > enum Aspect { ASP_APP, ASP_PHYS} > CXT_APP = "app", CXT_PHYS = "phys", etc. > ASP_APP = "app", ASP_PHYS = "phys", etc. Hi Marcus, There are several points of customisation that do or might allow it. For example, derive an "Info_Container" (into which identifier/value pairs are inserted via registration callback) with your callback performing simple automated substitutions like removal of text up to and including an underscore and conversion to lower case. At worst - not recommended as it requires manual maintainence - you can resort to ignoring the automated registration and hardcode one-by-one insertions using arbitrary identifiers, or even create a completely customer Info_Container class hardcoding support for a specific enum. The main BENUM macro also allows specification of extra functions to include in the generated class (a set of such precanned "features" are provided in benum_features.h). I haven't tried it, but I suspect a static method could be introduced to perform a transformation, but it would need to be called before first use. > > - smart enums requires user-maintained "meta-data" - from the smart enum docs: > > bool init = (::EnumIO<Foo>::the() (Bar, "bar") (Baz, "baz")), true; > > EnumIO only does that, and it hides the usage of EnumIO<> from the > header file, thus it is very lean and efficient. > How much extra template code is instantiated in each source file > including BENUM-use? Not so much: - only a thin wrapper is instantiated for each type, performing type casting as it interacts with a single instance of the core storage/lookup code that works with int enumeration values - there's a preprocessor #ifdef to control whether one of the implementation files is sucked into the related header for convenience vs out-of-line for decoupling physical dependency / faster compiles Regards, Tony

1 0

Re: [boost] C++ Builder 2009 and Boost compliance
by Peter Dimov 16 Sep '08

16 Sep '08

siliconman: > In article <48CE15DC.6010301(a)tid.es>, joaquin(a)tid.es wrote: > >> It'd be nice if we could have a regression test runner for this >> product. > > I'll be adding it to my test runs soon. > -- > -David David, Do you have an idea why your 5.6.4 and 5.8.2 tests fail (apparently at random) with errors like: Fatal: Unable to open file 'turboc.$ln' Error: Could not delete C:\Boost\results\boost\bin.v2\libs\smart_ptr\test\sp_accept_owner_test.test\borland-5.6.4\debug\sp_accept_owner_test.tds (project already open in IDE?) ====== BEGIN OUTPUT ====== '"C:\Boost\results\boost\bin.v2\libs\smart_ptr\test\sp_atomic_test.test\borland-5.6.4\debug\sp_atomic_test.exe"' is not recognized as an internal or external command, operable program or batch file. EXIT STATUS: 9009 ====== END OUTPUT ====== I'd like to monitor these test results, but their unreliability makes it impossible.

1 0

Symbol not found for OS X
by Michel Valstar 16 Sep '08

16 Sep '08

Hello, I'm trying to build a project that was developed for Ubuntu on a Mac OS X 10.5. Building goes well for a long while, but then I get the following error: boost::python::converter::arg_to_python<unsigned char>::arg_to_python(unsigned char const&)in trackerbasewrapper.os "_PyObject_IsTrue", referenced from: boost::python::api::object_operators<boost::python::api::object>::operator _object* (boost::python::api::object::*)() const() constin trackerbasewrapper.os "_PyExc_KeyError", referenced from: _PyExc_KeyError$non_lazy_ptr in trackerbasewrapper.os "_PyObject_IsInstance", referenced from: boost::python::enum_<FeatureUsage>::convertible_from_python(_object*) in trackerbasewrapper.os "_PyExc_UserWarning", referenced from: _PyExc_UserWarning$non_lazy_ptr in trackerbasewrapper.os "_PyBool_FromLong", referenced from: boost::python::to_python_value<bool const&>::operator()(bool const&) constin trackerbasewrapper.os ld: symbol(s) not found collect2: ld returned 1 exit status scons: *** [trackerbase/libtrackerbasewrapper.dylib] Error 1 scons: building terminated because of errors. Is this a symbol that still needs to be added for the Max OS X codebase? Thanks, Michel -- Dr. Michel Valstar, Research Associate http://www.doc.ic.ac.uk/~mvalstar Department of Computing Imperial College London 180 Queens' Gate London SW7 2AZ, U.K. tel.: +44 20 7594 8234 fax: +44 20 7581 8024

1 0