From: Jens Maurer (Jens.Maurer_at_[hidden])
Date: 2000-12-13 13:29:23
David Abrahams wrote:
> From: "Jens Maurer" <Jens.Maurer_at_[hidden]>
> > Lois' suggestion seems to entail that there are header files mixed
> > with subdirectories (containing implementation .cpp) in several levels
> > of the directory structure. I would consider this sub-optimal.
>
> Why?
The boost/ subdirectory would look like this:
array
array.hpp
array_traits
array_traits.hpp
compose
compose.hpp
concept_check
concept_check.hpp
[...]
This has twice the amount of directory entries than the
current scheme under libs/. We currently have 21 subdirectories
under libs/, and I expect that number to grow. I would
prefer the number of subdirectory entries in libs/ (or boost/) to
grow with n and not with 2*n.
I do think that splitting large headers into smaller parts and
also providing an "all-in-all" header is a good idea. (I need to
do that with my random number library.)
Probably we should leave the idea that the distributed boost directory
structure can be used as-is and start having a minimal install
procedure. As a first step until we get a decent build system,
we could have a script which copies (or symlinks) all *.hpp files
from each library's include/ subdirectory to the boost/ include
directory. Probably python is the appropriate tool for that.
Note that we already have several libraries whose object files
need to be linked in (i.e. are not headers-only), so an installation
step is necessary anyway, in general.
If we would like to keep the current contents of the boost/
directory, Beman could run the copy script before creating the
release .zip file. People using the CVS version are considered
more knowledgeable and thus can call the script themselves.
Jens Maurer