On Jun 8, 2009, at 8:47 PM, David Bergman wrote:
The bulk are generic language-lifting tools, basically giving the programmer a greater vocabulary. And, even if I bashed math libraries a bit, I would classify the current (as of 1.39.0) libraries as (and I did go through each library, and might have counted wrong with a few units here and there):
1. Language Extensions - 60 (including Statechart and BGL, which are applicable much more often than the developer realizes)
2. Common Types & Features (often part of newer languages standard libraries) - 8 (Accumulate, Numeric Conversion, Date, Format, Random, Regex, Serialization, Xpressive)
3. OS Abstraction Layer - 8 (Filesystem, Asio, Interprocess, Iostreams, Pool, System, Thread, Timer)
4. Math - 8 (Interval, various Math libraries, Rational)
5. Other Specialized - 7 (CRC, GIL, MPI, Proto, Python, Spirit, uBLAS, Units)
The first two categories can be characterized as bringing the language of C++ up to (and beyond) that of newer creations, where the third category is often part of such a standard library. The number of libraries belonging to those three categories is 76.
The latter two categories are clearly domain-specific, though. The number of such libraries is 15.
If you disagree with these figures, I welcome a revised table, otherwise we can clearly state that domain- and target-*agnostic* libraries constitute the bulk of Boost, by far.
We could say "OK, no more domain-specific libraries in Boost!!!~!~1`~!!~111" but first we must formally define the domain of libraries that are acceptable in Boost. Good luck with that. :)
First of all, I do not understand why we would have to use profanity in stating that, but alright, I would definitely welcome such a decision, and it is quite easy; just see the table above. I would welcome exclusion of those 15 domain-specific libraries...
Looking even closer at the potentially domain-specific libraries (those 15; I actually moved Spirit up to #2, I see now), we see tthat 12 of them belong to the domain of Scientific Computing. So, if we state that "Boost is a set of language-enhancing libraries along with the slight exception of one domain, scientific computing" we are left with only three libraries. Three oucasts: CRC, Python and GIL. The first one can easily be transitioned into category #2, and then we are left with only two: Python and GIL. And the former one is not even specific to any business domain. Let me put it this way: if something like Boost.Python was suggested today, and by somebody fairly new to the Boost community, it would not stand a chance to be considered ;-) Going back to your challenge of formally defining the "admissible" domains, where you even wished me good luck: there is only *one* specific domain for which Boost includes support and that is scientific computing, and even in such cases, the libraries should be multi-platform as far as possible. This holds up to two tiny exceptions, GIL and Python. And even those clearly exceptional cases are multi-platform. /David