Larry Evans a écrit :
On 05/12/2007 04:23 PM, remi.chateauneu@gmx.de wrote: [snip]
I do not know enough of this great library 'Fusion' to make a serious analysis, and BOOST_FUSION_ADAPT_STRUCT seems very close to what I did,
FWIW, I found the following two links useful for comparison:
For crd, see lines 29-42 of:
http://crd.cvs.sourceforge.net/crd/crd/README?revision=1.4&view=markup
For fusion, see lines 39-43 of:
Yes, crd's representation is heavier, although preprocessor macros could easily lighten it (but this is not the point I guess). Crd is a very modest library compared to Fusion and simply tries to focus on the needs I presented. Although it is very probably possible to do that with Fusion, crd is modelled after the need of some business applications to have an all-purpose definition of classes on which many commonly used algorithms would be 'plugged' onto. For example, this motivated the choice to use std::type_info * as unique key for all manipulation of members lists. This defines, for example, the column names of a table, with a std::map< const std::type_info *, std::string >. There are, as well, two specific features I wish to implement: - Loading classes definition from a compiled object files with debug information ( Please see this: http://www.garret.ru/~knizhnik/cppreflection/docs/reflect.html or what is returned by readelf), or from shareable libraries. This would target existing applications which cannot be modified for organisational or budget reasons. - Given a dynamic class definition (list of members), creating an object with these members, but with a blob as underlying storage. The goal is to be able to manipulate class-like objets, with the same interface as a real class. ( ... maybe std::map< const std::type_info *, boost::any > , with type_info referring a class'member ? ) I agree that the concept is not quite clear at the moment, but again this is related to communicating data with software components which cannot be re-compiled at each modification of a user's class. Thanks for your attention.
although the whole library seems more oriented towards compile-time operations ?
Well, according to:
the library is a "fusion" of compile time metaprogramming with runtime programming.
from the page:
libs/fusion/doc/html/fusion/preface.html
it should do both. Of course I realize "seems more oriented towards compile-time operations" is a "judgment call". However, it seems that the "basic operations" in both crd and fusion is accessing the I-th member in a tuple (or struct) and that the specific needs you mention below can be implemented using either the fusion "basic operations" or the crd "basic operations". If that's true (i.e. the "basic operations" are the "basis of comparison") then fusion is just as runtime oriented as crd.
My specific needs are: - Iterations at run-time on class members, not only at compile-time, because some operations might be difficult to code (Trivial example: Add the latest ')' in a SQL query string). This implies, for example, that manipulation on a list of class members are done with plain STL algorithms. - The possibility to add dynamic attributes for class members, indexed for example by typeid (Column names in a SQL table, or specific format for displaying a given class member), which may change from one object to another of the same class. - The possibility to consider as a plain data member, any method (getter/setter ...) of a class, even if there is no underlying member, or if they are private, and also base classes.
[snip]
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost