On Thu, Dec 16, 2004 at 09:50:36AM -0500, Caleb Epstein wrote:
On Thu, 16 Dec 2004 08:44:20 -0500, Stefan Seefeld <seefeld@sympatico.ca> wrote:
There are a number of flags that have to be made public, such as what type of runtime to use (that could mean, among others, the definition of __REENTRANT or the '-pthread' compiler flag). Also, on some platforms the dependency on dynamic libraries isn't well encapsulated, i.e. if libA.so depends on libB.so, an executable linking against libB.so also needs to be linked (explicitely) against libA.so. There is quite some variation between different platforms, and pkg-config tries to make life a bit easier for developers of portable code.
I understand all this. I was following up to Jon asking what the differnce would be between: "the flags that /should/be/ used to compile against Boost, not the flags the /were/ used to compile Boost itself."
Sorry, I hadn't had a chance to think of anything. You might have compiled Boost with a certain level of optimisation, say by setting GXX="g++ -O3" (for the GCC toolset) but you don't necessarily want to optimise your code at that level (if at all) so --cflags should not emit -O3. Similarly for -ansi, -pedantic, -W and other switches that don't affect the ABI or whether you can link to a library safely. Or is that irrelevant? jon -- "The optimist believes this is the best of all possible worlds, and the pessimist fears this is so..." - H. Allen Smith