Policy on "one size fits all" builds

I have a (small) dilemma.

discussion relates to dmasound driver (but prop relates elsewhere).

The idea of :

"one size fits all" compiles

is in direct conflict with:

"I want to build the smallest version and don't want to look for any
hardware I know I don't have".

for example, I can put wrappers round the cuda usage in dmasound_awacs that
will prevent it being compiled in if someone switches off CONFIG_CUDA (it
has already been done at 2.4.0 by someone).

So.  Should I wrapper or should available resources be probed at run-time?

(I don't mind which way I do it - but I'd like some guidance from those who
ultimately determine what is acceptable)

Iain.

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/