I don't know. Apparently, a lot of money is being spent on numerical software.
These amateur programmers typically cost a few thousand dollars a week
and each of the packages available on the internet probably required several
man-weeks
to develop. There are probably many more packages developed in private industry
which we know nothing about because they have disclosure rules which prevent
them
from publishing and contributing their work. If these people were to buy the
numerical
packages that they use instead of developing their own in-house, they could
probably
support several competing library developers.
So why do people develop their own packages instead of buying them from library
developers?
I think the real reason has more to do with controlling risk than anything else.
Without a standard, you are obliged to commit to a single vendor
and if that vendor refuses to support the package or accommodate your needs
when you decide to move to another platform, you are stuck.
You will probably need to re-write all your application source code
in order to migrate to another vendor's library.
Also, without a standard, library developers are gambling
that potential customers will understand and accept their particular approach.
I think the economics might change if there was an accepted standard.
> The professionals have a strong incentive to short circuit the competition,
> so we tend to get products that only function together within a tight bundle.
> Of course, we amateurs don't plan anything on a grand scale
> (except things like this mailing list, which is making an effort!),
> so nothing seems to work together.
I'm not a socialist. I don't think it is necessary or wise to legislate
compatibility.
Application programmers are responsible for finding and linking compatible
libraries.
> One thing that is left out from your decomposition are the "things"
> (other than matrices) that these tools take as input and output:
>
> Geometrical regions and grids
> Functions
> Operators
>
> Note that matrices are used in NA to represent many more things than linear
operators
> (like functions on regular grids), and I think this distinction between the
mathematical object
> and the data structure must be more firmly made.
So, when should we expect to see your proposal? Bob Tisdale <edwin2@gte.net>