need help with tool requirement

[bmcculley at rcn.com]

THANKS!

That is exactly the kind of informed and helpful reply I'd
needed and hoped for - it is very valuable to me!

Hint to others, I would also appreciate greatly any other
similar tools to the Bullseye coverage analyzer, pointers
thereto and discussion thereof.

Thanks again!

-Bruce

---- Original message ----
>Date: Wed, 21 Apr 2004 10:57:50 -0400
>From: Bob Bell <bbell at hp.com>  
>Subject: Re: need help with tool requirement  
>To: bmcculley at rcn.com
>Cc: "Kevin D. Clark" <kdc at rcn.com>, discuss at gnhlug.org
>
>On Wed, Apr 21, 2004 at 01:16:44AM -0400, bmcculley at rcn.com
<bmcculley at rcn.com> wrote:
>> The approach I thought of is based on doing exactly that, in an
>> emulator or virtual machine environment, single stepping
automagically
>> through the debugger until a conditional transfer of control is
>> recognized.  At that point a snapshot would be taken, the
natural
>> branch followed, iterating until complete.  Then revert to
the last
>> snapshot, force the alternative branch, repeat until done.
>> 
>> What's the big deal?  Seems like it could probably all be
wrapped into
>> a PERL one-liner at the top level, even if the one-liner
has to expand
>> into some more complex subfunctions down the tree.
>> 
>> All right, I concede I am not considering potential data
dependencies
>> (e.g. if this leg of the branch is taken some data item(s)
must be
>> this value not that), which could affect array indexing for
example.
>> I'll accept that limitation for the present, because this
will not be
>> used in isolation.
>
>I think that that's a big problem -- the different
dependencies required
>to take a branch.  Taking branch B may only "make sense" if
back in
>routine R variable X was set to Y, and if you take branch B
when X is Z,
>your program may do weird things, and for good reason.
>
>The approach taken here at HP for HP-UX, where there has been
a lot
>emphasis recently on quality and, in turn, code coverage, is
using
>a tool from Bullseye (http://www.bullseye.com/) that records
when the
>true and false cases of every "condition/decision point" are
taken, and
>then adding testing to make sure that some minimum percentage of
>coverage is reached.
>
>It's actually from trying to add testing to get that coverage
that
>I assert that "forcing" a branch is non-trivial with regard to
>dependencies.  And, if intuition serves correctly by
extrapolating some
>concepts from compiler theory, definitively examining those
dependencies
>would take expotential time with the best algorithms we have,
and in
>fact I believe may be NP-complete.
>
>DISCLAIMER: I think I'm right, but I could just be blowing
smoke...
>
>-- 
>Bob Bell
>_______________________________________________
>gnhlug-discuss mailing list
>gnhlug-discuss at mail.gnhlug.org
>http://mail.gnhlug.org/mailman/listinfo/gnhlug-discuss