Re: [Boost-build] [BBv1][Testing] Running the same test multiple times with different sets of parameters?

From: David Abrahams (dave_at_[hidden])
Date: 2005-12-15 11:09:06

Douglas Gregor <doug.gregor_at_[hidden]> writes:

> On Dec 12, 2005, at 12:01 PM, Rene Rivera wrote:
>>> Does anyone know how I can achieve this functionality in BBv1's  
>>> testing
>>> environment?
>>
>> The easiest way is close to (2). If you create a "lib" with most of  
>> the
>> sources of the test and link that into each "exe" you'll save the
>> compilation time at least.
>
> Okay. I have some vague recollection that some platforms don't link  
> executables properly when main() is stored in a library, but it's too  
> vague a recollection to worry about. But, I'll try it out. Worst  
> case, I pull main out into stub .cpp file.
>
>> Other than that it would require some internal hacking/kludging to/ 
>> with
>> testing.jam. I'd need to know what the MPI_RUN code is doing to really
>> answer entertain this route.
>
> Definitely not worth it, then. However, I'll describe what's  
> happening in case we want to think about it for BBv2.
> I'm building programs using MPI (the Message Passing Interface).  
> These programs cannot typically be run standalone. For instance, if I  
> build all_gather_test.exe, I can't just run it by itself. Instead, I  
> need to run it through an auxiliary program called "mpirun", like this:
>
> 	mpirun -np 7 all_gather_test.exe

This looks very familiar.  The Python tests are being run through an
auxilliary program, called, strangely enough, "Python." ;-)

> The "-np 7" tells mpirun to launch 7 separate processes, each of  
> which will run a unique copy of all_gather_test.exe. Then the MPI  
> library that all_gather_test.exe will allow them to communicate.
>
> For proper testing of all_gather_test.exe, we can't just fix the  
> number of processors and run it once. We really need to run on  
> several different configurations (one node, powers-of-two, powers-of- 
> three, prime numbers, and one less than powers-of-two are typical).  
> So, for instance, we'd like to run all of these as separate tests:
>
> 	mpirun -np 1 all_gather_test.exe
> 	mpirun -np 2 all_gather_test.exe
> 	mpirun -np 7 all_gather_test.exe
> 	mpirun -np 8 all_gather_test.exe
> 	mpirun -np 9 all_gather_test.exe
> 	mpirun -np 17 all_gather_test.exe
>
> Thanks for your help!

You might take a look at the way Python testing is broken down.  You
can build separate Python extensions and then test them any number of
different ways using separate test targets.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com