On Fri, Oct 31, 2008 at 12:14 PM, Mike Jackson <mike.jackson@bluequartz.net> wrote:

I am working through the "testing" part of the build system at this point and I have some questions about how the typical boost developer actually runs the "tests".

The tests are compiled using the build/source tree. Is it probable that boost developers actually install boost then try and compile tests against the installed boost?

Mostly I just work within an svn working copy, but once in a while do actually install boost and then build and run tests.

My usual approach is to develop using the VC++ IDE, and when whatever I'm working on runs for that single compiler, I run a script that invokes bjam for the six compilers I test against before an actual svn commit.

The IDE setup needs to be reasonably straightforward; I want to be able to specifiy a single Boost include directory, not one for each library.

My guess is that usage pattern is common, but there are probably other usage patterns too.

I am working through some dependency issues with compiling the actual tests. With boost modularized there are lots of missing includes. The current way I am going about adding them is probably NOT maintainable. I need to do it like the actual build system does it in that you have a "module.cmake" file that lists the dependencies. I need to work through that system and extend it for the testing cmake files.

Is there a way to (via hard links, perhaps) make it appear both (1) the headers are in a single location in the root directory as is done currently, and (2) also in an include directory within each library's directory in the lib directory hierarchy? How do Subversion and CMake do that sort of thing?

--Beman