Re: Conditional tests based on ${MACHINE} and ${MACHINE_ARCH}

[Antti Kantee <pooka%netbsd.org@localhost>]

On Sat Jan 01 2011 at 06:30:07 -0800, Paul Goyette wrote:
> There are some tests which don't apply to every architecture, i.e., the 
> tests in lib/libc/ieeefp/ and lib/libc/gen/ldexp (which probably should 
> be moved to ieeefp).  The old "regress" structure dealt with these by 
> simply not building the tests; since no tests were installed in the old 
> structrue, there was no need to be concerned with the sets-lists.
> 
> However, moving these tests to atf means that things do get installed, 
> and we now have to deal with conditionalizing the sets lists if the 
> tests themselves are conditionally built.  A test that gets built "for 
> everything except vax" would need to get added to 18 out of 19 files [1]
> in distrib/sets/lists/tests/md.*  (I currently have added libc/gen/ldexp 
> which is probably breaking the build for vax people.)
> 
> Havard suggest an alternative - have the test (or framework) be able to 
> make these checks at run-time.  Maybe something could be added to the 
> Atffile ?  Or at the very least, atf could create config_variables that 
> the tests could check?

There's atf_arch and atf_machine (and these are the opposite of MACHINE
and MACHINE_ARCH for some gnuy reason).  Do the aieeefp tests compile
on vax or do you need some intra-tp ifdef games instead?

I agree that having to wrestle with md test setlists is a hopeless battle
and should be avoided.  (if only setlists could be avoided completely ;)

-- 
älä karot toivorikkauttas, kyl rätei ja lumpui piisaa