On 19/12/10 19:40, Brent Spillner wrote:

I uploaded sample code to the Vault (in the "Miscellaneous/" directory) that provides a concise notation for list comprehensions. The comprehensions are evaluated lazily and can be used to populate a std::list, vector, or deque, appended to the same, inserted after a given iterator, or interpreted as a boost::range, including compatibility with BOOST_FOREACH. Containers, iterator-defined ranges, or nullary function objects can be used as generators, and arbitrary function objects can be used to specify filter conditions.

I've tested this so far with GCC 4.4.4, GCC 4.5.2, and Clang 2.9, using Boost 1.42 and 1.45, and with and without -std=c++0x. Reports of success/failure with other environments are most welcome. The code is still in a prototype stage, so I'm more than happy to take design suggestions and would love to hear about use cases I've overlooked. If there's sufficient interest I will work on getting it to production quality and writing some real documentation. I apologize in advance for wasting everybody's time if something like this already exists in Phoenix; I'm not too familiar with it but I skimmed the docs and didn't see anything. I like the idea :) However, how did you manage the lazy part ? Do you roll your own lazy stuff or do you rely on proto ?

If the former, I encourage you to jump the agte and do the latter :)

On 12/19/10 12:40, Brent Spillner wrote:

I uploaded sample code to the Vault (in the "Miscellaneous/" directory) that provides a concise notation for list comprehensions. The comprehensions are evaluated lazily and can be used to populate a std::list, vector, or deque, appended to the same, inserted after a given iterator, or interpreted as a boost::range, including compatibility with BOOST_FOREACH. Containers, iterator-defined ranges, or nullary function objects can be used as generators, and arbitrary function objects can be used to specify filter conditions.

List comprehensions in haskell can be done with monads: http://book.realworldhaskell.org/read/monads.html Why not 1st define a c++ monad and then define the list comprehension using the moand? If that were done, then boost would have both. Actually, fcpp already has it in monad.h. Fcpp is available at sourceforge: http://sourceforge.net/projects/fcpp/ However, I've not used fcpp and several years ago boost reviewed it but decided not to include it. http://lists.boost.org/Archives/boost/2004/02/61840.php However, I think defining a monad acceptable to boost would be useful since it is used extensively in functional programming. -regards, Larry

Dave Abrahams wrote:

I'm usually interested in comprehensions like (in Python):

   [ x, foo(x) for x in some_other_sequence ]

and your consistent use of range() above obscures whether "for x in some_other_sequence" can be modeled in general.  Can I substitute any sequence for range(...) in the above examples?

Yes, sorry for not making that clear. Continuing the prime number example, you can generate Mersenne primes with std::vector<int> mersenne_primes = _((1 << p) - 1)(7)[p <<= primes, apply(is_prime, (1 << p) - 1)]; This illustrates several points that I neglected before-- use of apply() to invoke a function object with arguments, specification of a finite limit (7) on the number of values to be returned (assigning to a container via <<= would also implicitly limit it to the size of that container), and the fact that I haven't yet allowed any assignment or side-effect (increment/decrement) operators within the comprehension body. My initial instinct was that that wouldn't be very FP-ish, but now I'm thinking it would be handier to allow things like 'm = (1 << p) - 1' or 'count++' within the condition list. I uploaded a new version to the Vault that includes the Mersenne generator example and a few bugfixes.

Anyway, very promising.  Keep up the great work!

Thanks, I appreciate the encouragement! Brent