On Saturday 24 March 2007 00:01, AlanCoelho wrote:
> I am to believe that scientists prefer to mull over source code rather than
> pseudo code and mathematical descriptions. Anyone that knows just a little
> about software development would know that source code is the last thing
> that one wants to see. How many has deciphered the source code of an FFT
> routine, how many would want to.
That is true, most of the time. However this is the policy for Nature
*Methods* - people who read their articles in detail will want to have access
to some source code.
You may compare this to the publication of a new structure : do you read
all the CIF files associated to the new structures ? Of course not - but the
CIF is part of the demonstration.
If someone claims that his algorithm allows computing FFT with O(n)
complexity, it is fair to ask him for a practical demonstration available to
all readers.
And their policy *does* take into account commercial software, *not*
forcing the code to be opened, as long as a practical demonstration is
available and details are available to referees :"the possibility to reach an
agreement on restricted distribution does not obviate the need to provide,
during confidential peer review, all programming details deemed necessary by
the reviewers to evaluate the method."
If they had mandated all code (full software) associated to articles to be
open-source, they would clearly have been unreasonable (as much as I like
open-source), but this is _not_ the case here. As I see things, a _small_
piece of code should be associated to each algorithm, not a full software
with > 10 000 lines of code.
> The language of maths has evolved over the
> centuries and computer code whether it be Fortran or c++ is simply not
> adequate for describing complex algorithms.
I think next month their editorial will mandate the use of Python to
demonstrate new algorithms.
> There's a hidden agenda here and those pushing it should have the fortitude
> to come clean about their motives; that is that scientists would like
> software for free.
Most scientists want software which (i) is efficient and (ii) they can play
around with (=modify). Free is a side issue for most scientists I know.
The bottom line as far as Nature Methods is concerned is not whether
software should be free, but how a new method can be spread efficiently -
their conclusion says: "the algorithmic components that constitute integral
parts of new methods must be made available and in a format that will
facilitate the method's adoption"
Vincent
--
Vincent Favre-Nicolin
Université Joseph Fourier
http://v.favrenicolin.free.fr
ObjCryst & Fox : http://objcryst.sourceforge.net
