- Boost - lists.boost.org

Phoenix review
by Peter Dimov 30 Sep '08

30 Sep '08

I apologize in advance that I could not dedicate the appropriate amount of time for my evaluation of the library. * Design - The placeholders probably belong in namespace phoenix::placeholders instead of arg_names, for consistency with std::placeholders. - It is very odd to include phoenix_bind.hpp and not have the placeholders available. For that I need phoenix_core.hpp. - The decision to capture the left side of @= and = by value is debatable. It is indeed not clear which one is 'better', but we do have Lambda in Boost, and to deviate from it one needs a stronger supporting argument than that. - In general, the subtle differences between Bind, Lambda and Phoenix need an explanation, even if its only purpose is to clearly show that the deviation is deliberate and not an accident. - bind<R> is not supported. - Smart pointers do not seem to be supported well in conjunction with bind, even though they seem to work in other contexts. - What is the equivalent of boost::unlambda in Phoenix? I think I agree with Giovanni Piero Deretta that lambda[] is an intuitive way to put a 'wall' between the inside lambda expression and the outside world. It doesn't seem to; the following fails: px::bind( px::lambda[ arg1 + arg2 ], 0, 1 )(); - It is not clear what introspection facilities are provided. Users sometimes want to derive a signature from a bind expression, or to compute the minimum arity of a lambda expression. They ought to be able to implement these queries themselves via a documented introspection interface that will not break when V3 arrives. * Implementation - Phoenix objects do not take rvalues. This is a very common stumbling block for people wanting to experiment with lambda libraries. - boost::is_placeholder needs to be specialized for the placeholder types. It might be interesting to see Phoenix respecting is_placeholder in the other direction, too, if possible. - The test coverage seems relatively sparse. I know very well how painful it is to write tests (for arities from 0 to 9 inclusive), but Phoenix 2.0 needs and deserves more of them. Take this, for example: test::x x_; bind(&test::x::m, x_)() = 123; bind(&test::x::m, arg1)(x_) = 123; BOOST_TEST(x_.m == 123); It doesn't test whether the first bind worked, and doesn't try to read from a member using bind. * Documentation - The examples assume a number of using directives without this being stated. - I believe that it would be better to leave the std:: names qualified. As it stands, it is not immediately obvious whether cout is std::cout or phoenix::cout (a plausible assumption would be for a phoenix::cout to be defined as phoenix::ref(std::cout) to avoid capture by value). - The val(3) example stresses that it returns a nullary function, but val(3) happily accepts (and discards) arguments - as it should. - The relationship between phoenix::ref and boost::ref is not made clear, or even mentioned. Can one use boost::ref instead? In what contexts? std::ref? (As an aside, there is a slight tendency in the documentation to be a bit too independent of the rest of Boost for my taste. It mentions Lambda in passing, but doesn't go into any details; it completely ignores boost::bind.) - The "Forwarding Problem" paper is so 2002. :-) - The note that the binders are monomorphic should probably be fixed. - It might be instructive to add an introspection example to the final chapter that returns the (minimum) arity of a Phoenix expression. - The documentation is missing a reference section, or at least an index of the identifiers in the phoenix namespace. * Vote This is hard. On the one hand, it is ridiculous to include <boost/spirit/include/phoenix_bind.hpp> for phoenix::bind instead of <boost/phoenix/bind.hpp>, as one would reasonably expect. As Phoenix is already in Boost, the review merely needs to formally acknowledge the facts. And the library certainly meets the criteria for acceptance. On the other hand... I sense from the state of the documentation that Phoenix being under the Spirit umbrella and not an "official top level" library has led to a somewhat isolated development. It is very high quality, above the Boost median, but somehow doesn't yet feel like a top level library. Still being on the other hand... We know that having two libraries with overlapping goals, overlapping syntax, overlapping identifiers and placeholders, and subtle differences is confusing and a continuous source of recurring questions. Now we will have three. But the fair thing to do is vote for acceptance. Which I will.

2 1

[Review] Phoenix
by John 30 Sep '08

30 Sep '08

First, the big question I vote for a second fast track review of the v3 implementation. The reasoning is simple, if a little painful. I have a lot of sympathy for the view that we should be reviewing the version that is actually going to be released (or at least something close to it). Joel is right that some libraries have been admitted with the condition of large changes. In general, given the exceptional quality of his work to date, I also consider him quite trustworthy when he says that is what he plans to do. However, I have problems with how I'm supposed to review the implementation of a library that won't be implemented the same way when it goes into Boost. That said, even the v2 edition is an exceptional library. It has been pointed out elsewhere that there are some current details missing from v2, and Joel has already promised their inclusion in v3. Aside from those artifacts of its history, it is a very nice piece of work. I have used it in a few projects and I find it clear to use, well documented and largely well behaved (except for afore mentioned problems). It works on compilers that live close enough to the standard, and that happens to be what I currently use. It is well documented (I also vote for the return of the section on FP.), and I enjoy the style. There is a lot to learn to use the library well, but that is partially because it is a powerful library in a programming idiom that many of us have only learned we need recently. It is practical, and so provides a gentler step into functional programming than would be the case if it stressed purity. However, it allows clearly functional constructs and has well thought out abstractions underlying the interface The reasoning is painful because it has me voting against immediate inclusion for a library I obviously like and respect. However, I think this review is valuable for what it accomplishes. In my following of the discussion so far, I haven't seen anyone who has issues with the need for the library, with the basics of the interface, or with the abstractions it embodies. I think this is the 85% good stuff (I may be remembering the wrong number here.) that Joel spoke of. That allows us to focus the fast track I am voting for on the implementation and compliance with a few specific changes. v2 will be available, and the small breaking changes in v3 will be the first boost release (since I'd be very surprised if Joel didn't do all requested and more for v3), so there will be a minimal legacy code base issue. As for the naming, I would also like to see the release version become the new Lambda v2. Though I appreciate creative names for some things, it seems to cause problems for software libraries. John Phillips

2 1

request for interest: stable vector
by JOAQUIN M. LOPEZ MUÑOZ 29 Sep '08

29 Sep '08

I recently wrote up the implementation of a *stable* vector, a container mimicking the interface of std::vector except that it provides iterator and reference stability at the expense of losing memory contiguity: http://bannalia.blogspot.com/2008/09/introducing-stablevector.html If there is interest in the community this could be grown into a full-fledged proposal for Boost (either by me or by any other interested colleague). Joaquín M López Muñoz Telefónica, Investigación y Desarrollo

9 22

[interprocess] atomic_write32, etc. -> public?
by k-oli＠gmx.de 29 Sep '08

29 Sep '08

Hello, interprocess provides in its detail namespace atomic ops like atomic_write32, atomic_cas32 etc. Is it possible to move this functionality to a public namespace? I'm using interprocess::detail::atomic_write32 for instance in an work-stealing queue and I'd like to use 'official' functions instead from a detail namespace. regards, Oliver

2 1

[Phoenix] review reminder
by Hartmut Kaiser 29 Sep '08

29 Sep '08

All, this is just a gentle reminder for all of you willing and planning to write a review for Joel de Guzman's and Dan Marsden's library Phoenix V2. The review ends tomorrow, September 30th. If you are for some reason not able to write the review before tomorrow night, please get in contact with me directly, so we can work something out. Regards Hartmut Review Manager

1 0

[threadpool] work-stealing support
by k-oli＠gmx.de 29 Sep '08

29 Sep '08

Hello, I've the work-stealing strategy ready - I'd like to know if it is desired to remove lazy and adaptive strategy from the lib (as Vicente Botet and Johan Torp mentioned)? So only fixed an work-stealing will be supported?! thx, Oliver

2 3

[smart_ptr] shared_ptr<T> T destructor required even if a deleter function is provided
by vicente.botet 29 Sep '08

29 Sep '08

Hello, I have compiled the example Preventing delete px.get()and every thing is OK until we use the reset function because shared_ptr<X> request that the destructor of X must be public even if a deleter function is provided. I'm wondering if this is a BUG on the smart pointer library Next follows the trace: ../../../boost/checked_delete.hpp: In function `void boost::checked_delete(T*) [ with T = X]': ../../../boost/detail/shared_count.hpp:91: instantiated from `boost::detail::s hared_count::shared_count(Y*) [with Y = X]' ../../../boost/shared_ptr.hpp:173: instantiated from `boost::shared_ptr< <temp late-parameter-1-1> >::shared_ptr(Y*) [with Y = X, T = X]' ../../../boost/shared_ptr.hpp:394: instantiated from `void boost::shared_ptr< <template-parameter-1-1> >::reset(Y*) [with Y = X, T = X]' test_shared_ptr.cpp:37: instantiated from here test_shared_ptr.cpp:13: error: `X::~X()' is private ../../../boost/checked_delete.hpp:34: error: within this context And here it is the source code #include <boost/shared_ptr.hpp> using namespace boost; class X { private: ~X(){}; class deleter; friend class deleter; class deleter { public: void operator()(X * p) { delete p; } }; public: static shared_ptr<X> create() { shared_ptr<X> px(new X, X::deleter()); return px; } }; int main () { shared_ptr<X> ptr = X::create(); ptr.reset(X::create().get()); return 0; } ______________________ Vicente Juan Botet Escribá

2 4

[threadpool] relation with TR2 proposal
by joaquin＠tid.es 29 Sep '08

29 Sep '08

Hello, I learnt yesterday that there is a TR2 proposal for a thread_pool, sponsored by our Boost colleague Anthony Williams: http://www2.open-std.org/JTC1/sc22/wg21/docs/papers/2007/n2276.html What is the relationship with this proposal of the threadpool library that is being discussed these days here at the Boost list? Thank you, Joaquín M López Muñoz Telefónica, Investigación y Desarrollo

6 24

[lightweight_mutex] Patch to support new thread api
by Frank Mori Hess 29 Sep '08

29 Sep '08

Here is a patch which makes boost::detail::lightweight_mutex conform to the Lockable concept from the new Boost.Thread api. It shouldn't break backwards compatibility for code already using it. Note I haven't tested it much, especially the windows part (which I haven't even tried to compile), but the changes are pretty straightforward.

1 0

[serialization] merge to releasse ready
by Robert Ramey 28 Sep '08

28 Sep '08

I want to merge the current trunk version of the serialization library into the release-ready version. I've checked the wiki page and it seems that I've met all the criteria. I'm just double checking. There are some changes that I'm not included right now. Basically these are the original versions of certain modules located in the boost directory. E.G boost/static_warning.hpp. The serialization library doesn't use them any more and each one has been modified to include a #error -... in case someone has used them. I haven't recieved any reports that this has broken anything so I'm presuming that no other library - and perhaps no user application is including them. The question of whether these should be removed is separate from the above. I just thought I'd post this comment before removing them. Robert Ramey

3 4