[xml] porting to supercomputers

[Scott Murman <smurman nas nasa gov>]

Greetings.  This is a quick note to let the developers of libxml2 know 
about some of the uses we're putting the library to in the 
supercomputing arena.  I work in a field called computational fluid 
dynamics (CFD), and we've recently been building tools using XML as the 
persistent storage, largely to leverage tools like libxml2 for parsing, 
validating, etc.  In the next month or so this package will be released 
to other users throughout the country who work similar problems.  Our 
target platforms are basically the top 500 list 
(http://www.top500.org/list/2003/11/).  For the most part porting and 
using libxml2 on these machines is not a problem, as they are all 
unix-based.  there are a couple of issues we've run into in our limited 
testing though that i'd like to get out to those who maintain the code 
(and also to open a dialog so that issues don't come out of leftfield 
in the future).

It's necessary for us to use 64-bit addressing, which usually means 
passing special compile flags to 3rd-party builds.  Right now these 
flags are not passed to libtool within the libxml2 build.  i.e. 
CFLAGS="-64";./configure will not correctly build the shared libraries.

On shared-memory architectures, pthreads and OpenMP do not play well 
together.  this means that thread support for libxml2 must be disabled. 
 this is not a loss for us, but please don't remove the ability to 
compile libxml2 w/o thread support in the future, as we'll be toast.

Keep up the good work,

-SM-