Unit tests of internal functions

[Nicolas Williams]

On Tue, Jan 05, 2010 at 01:50:37PM -0500, Jeffrey Hutzelman wrote:
> --On Monday, January 04, 2010 02:50:42 PM -0600 Nicolas Williams 
> <Nicolas.Williams at sun.com> wrote:
> >I favor (1).  However, you could do a hybrid of (1) and (3).  Namely,
> >alter the symbol export list according to whether you're making the
> >check target.  One would "make check && make clobber && make install".
> 
> Ahhh!!!! My eyes are bleeding!!!
> 
> With _very_ few exceptions, make targets should be idempotent.
> "make all check install" and "make all install" had better not install 
> different versions of the library.

I agree, and still prefer #1.  I did say "could", not "should".  I'd not
object too strongly to the hybrid approach for the simple reason that
this is not exactly a rare thing (e.g., SQLite3 builds differently for
testing than not; the SQLite3 folks have pretty much the same
how-to-unit-test-private- interfaces problem, and that is their
solution).

Nico
--