Unit tests of internal functions

[Ken Raeburn]

On Dec 28, 2009, at 19:51, Russ Allbery wrote:
> FWIW, I'm moving all of my packages to Automake and Libtool, and while
> that approach is not without problems, I think it's the right  
> direction to
> go.  It means that there's at least some hope that other people have
> already run into whatever weird build problem you have and have  
> written
> recipes to deal with, or even taught Automake and Libtool about it
> directly.  Upstream on both projects is fairly responsive.

We do some odd things in places, so it might take some work, even from  
someone experienced with the tools.  But I'm not going to tell anyone  
not to try.  (It'd be major enough a change that they'd probably need  
to write up a project proposal explaining how it would be done and why  
it would be a win.)  I'd also like to see more parallelism allowed in  
the build, while someone is hacking on it; there's no reason we can't  
build MD5 and SHA1 code at the same time, for example.

Last I looked, though, libtool supported neither library  
initialization/finalization functions nor symbol export lists, and we  
use both (and perhaps eventually symbol versioning).  The export lists  
we could make do without, but the finalization support would be a  
problem for anything dynamically loading and unloading our libraries.   
Some platforms let you do it via source-file declarations to the  
compiler, but some require extra options to be passed to the linker on  
the command line.  Have you seen anything to indicate that they've  
addressed this?

Ken