Re: test suite generation for Fortran compiler testing

nmm1@cus.cam.ac.uk (Nick Maclaren)
16 Dec 2004 00:36:43 -0500

          From comp.compilers

Related articles
test suite generation for Fortran compiler testing gauravga@noida.hcltech.com (Gaurav Gautam, Noida) (2004-11-20)
Re: test suite generation for Fortran compiler testing arargh411@arargh.com (2004-11-26)
Re: test suite generation for Fortran compiler testing gah@ugcs.caltech.edu (glen herrmannsfeldt) (2004-11-26)
Re: test suite generation for Fortran compiler testing beliavsky@aol.com (2004-11-26)
Re: test suite generation for Fortran compiler testing nmm1@cus.cam.ac.uk (2004-11-29)
Re: test suite generation for Fortran compiler testing henry@spsystems.net (2004-12-11)
Re: test suite generation for Fortran compiler testing henry@spsystems.net (2004-12-11)
Re: test suite generation for Fortran compiler testing nmm1@cus.cam.ac.uk (2004-12-16)
| List of all articles for this month |

From: nmm1@cus.cam.ac.uk (Nick Maclaren)
Newsgroups: comp.compilers
Date: 16 Dec 2004 00:36:43 -0500
Organization: University of Cambridge, England
References: 04-11-086 04-11-093 04-11-125 04-12-046
Keywords: Fortran, testing
Posted-Date: 16 Dec 2004 00:36:43 EST

henry@spsystems.net (Henry Spencer) writes:
|> Nick Maclaren <nmm1@cus.cam.ac.uk> wrote:
|>
|> >That myth about needing to exercise every code segment has been
|> >around for at least 35 years but, like all good myths, tramples
|> >ruthlessly on the actual facts...
|>
|> It's not a myth; it's merely not the whole story. Path coverage
|> testing and statement coverage testing are two different things. Path
|> coverage testing is *better*, but it's also rather harder to do.
|> Statement coverage testing isn't as good, but it's a lot better than
|> nothing and it finds a lot of bugs.


The myth is that it is adequate - see the previous postings for the
context. The harm that it does is that naive programmers put serious
effort into achieving completeness of statement coverage and believe
that, having done so, the program is well tested.


Whereas, in fact, there is no point in wasting any effort in even
ATTEMPTING complete statement coverage. The initial target should be
a thorough set of tests with empirical coverage rules, and the minimal
completeness target wirth considering is the path one. There are
harder completeness targets beyond the path one, too.


Regards,
Nick Maclaren.


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.