John S. J. Anderson wrote:
>Well, clearly the software publisher should pay for the certification
>process. Or maybe the cert. should cost for payware publishers, but be
>gratis for non-profit products.
As I just mentioned in another part of this thread, a code review
takes about 1 day per 1,000 lines. Assume $100/hour/reviewer-with-
good-development-experience-and-also-overhead and assume programs
are about 5,000 lines long (hmm, could get better statistics from
the CCL archive). Then a review would cost about
5,000 lines * 8 hours * $100
---------- --- == $4,000
1,000 lines hour
(The $100/hr is low, compared to quote I know of in the $200s, but
this is assuming there's enough people using the service to guarantee
a constant money stream.)
Now suppose non-profits are supported gratis. How many might send
in source code for a free code review, to find any remaining bugs?
You would also have problems with some languages. For example, most
people develop in C, Fortran or Perl. A few (like me) use Python
and others use Tcl. Those are pretty standard. But then there are
the groups using Smalltalk, Prolog, Matlab and S-Plus, as well as
locally-developed languages (eg, XPLOR's scripting language). They'll
proably be ignored because it's too expensive to support them.
>The metric(s) would be tricky -- some sort of test suite would
>probably need to be developed. Standard sets of data for each type of
>application.
There are a few. Just saw one mentioned today for quantum chem
programs computing the enthalpy of some compound. But there are few
programs which do exactly the same thing, so you have your BLASTs,
your FASTAs and your Smith-Watermans.
>I do think there's a business model in there somewhere; if anybody
>runs with it, please feel free to cut me in on the IPO. 8^)=
The closest I could see would be something like what epinions or
amazon does and be an accumulation of views about the usefulness of
a program and not an objective review per se.
Andrew
dalke at acm.org
