[udf-dlm-develop] Support for 32-bit "sensors"

[Ed S. <es...@la...>]

Dear all,

In response to problems encountered by Harald Frey and Steve Geller 
on FUV, I've checked in a new udf.c.  Please try it out at your 
convenience.

In short, some "sensors" are 32 bits, and cannot be converted to
regular ol' FLOAT without losing significant precision.  

The new udf.c does some checks to see if it should allocate a DOUBLE
instead of a FLOAT.  Thus you should get a returned struct containing
mixed FLOATs and DOUBLEs, where appropriate.

I'm not 100% certain that the checks are correct, so I'd like to hear
if you think that the returned struct is wrong.

There's also some code to make sure byte alignments are okay.  It
seems to do the right thing on my Intel platform, and *should* work
on Sparc [and anything else], but I have no way of testing.  

As always, CVS access is documented at:

	http://sourceforge.net/cvs/?group_id=868

Your help in testing this will be most welcome.

Thanks, and regards,
^E