In the interest of making a targeted, focused discussion about an issue
that is causing some confusion at the moment and portends to cause more,
I'd like to bring up topology readers. As I see it, we should ultimately
support two, or perhaps three canonized readers, one for C, C++, and
Fortran. I argue that the differences between C, C++, and Fortran are
sufficient to warrant different data structures for each case, and even if
there were some master object file that reads the topology and then a
series of converters to put it into different formats, there would be
about as much code to maintain as if we had three separate readers.
My recommendation is that we select, perhaps at the devs meeting, the
readers that will serve in each capacity. The readers should all do
precisely the same things, even though information will get filtered
differently:
- Collect contiguous arrays of charges, Lennard-Jones A and B, bond
stiffnesses, atom names, etc.
- Be compatible for parallel execution (i.e. read the topology file as
ASCII text and then parse the text, so that the text can be transmitted
through memory to other processes without having all processes trying to
read directly from the disk)
- Collect all information deemed relevant by the current standard
- Be able to read old-style prmtops as well as "prmtop7" format, with
automatic detection ???
- Efficient! Clean-coded! Expandable, and robust!
Comments welcome. At present, we'll probably need to decide amongst the
readers for nab, mdgx, and ptraj, and the readers for sander and pmemd,
perhaps cpptraj and any other C++ based code in the tree.
Dave
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Nov 29 2011 - 12:00:02 PST
