PDL::PP has some facilities for taking care of a number of
the following tasks.

=== Potential bugs or porting problems:

-- I assume sizeof(int) == sizeof(void *), and let
   perl make code for passing an int. Better is
   using $Config{ptrsize} and packing a string and
   using 'char *' in xs code. That way, this module
   has a better chance of being portable

-- Perl has a an interface to dlopen, etc. It may be
   more portable to use that interface, rather than
   the one I wrote. 

-- Use PP::Options to handle options rather than using hand coding.

=== Bugs:

=== Missing Features:

--- Support for other implementations of lapack, e.g. atlas. This should
    be fairly easy, as Makefile.PL has support for trying lists of libs
    in succession, and atlas is a drop in replacment.

--- liblevmar prints warnings and errors to sterr and sometimes exits. Instead
    pass  a buffer to the routines to collect messages. And return, don't exit
    on failed mallocs, etc.

==== Efficiency 

---- Found Inline C and Inline Pdlpp; missed them before even though I looked.

---- Strings representing lpp and C code are copied quite a bit in the
     routines. This should be fixed. refs to strings should be allowed
     in the FUNC option. If a contstant is passed to Func, it should store
     a copy in the object and continue to use pointers.
     Note, this is more difficult now because more code manipulation is required
     for support of single and double precision.

---- In generate_C_src() the regexs that replace t with t[i],
     etc. could probably be made more efficient. Currently I use
     while($fdef =~ s/([^\w]|^)t([^\w\[])/$1t[i]$2/g){}

---- The code that calls dlopen and dlsym would be simpler, more transparant
     (less dereferencing), and save a few function calls if it were
     moved from C code into  pp_addxs.  The same goes for any routines that
     are in pdlperlfunc.c buy get called from an xs wrapper. This is only
     a minor performance issue, but it is cleaner. Hmm, tried some of it,
     but some must be called from the pp_def, which is unweildly because of
     nested quotes, etc. So its not clean after all to have some of the
     DFP functions in pdlperlfunc.c and some in xs (pp_def).

#---------------------------------------------
 THINGS LISTED BELOW HAVE BEEN DONE  

Fri Jul  7 23:19:36 MST 2006

-- liblevmar has float support, but I only use the double functions.
   This should be fixed.

-- liblevmar can either allocate workspace or have the calling routine
   do it. I should add support for allocating space on the perl side
   for efficiency. Not hard.

---- The levmar_driver function is cruft from an earlier version of this
     module. It is an extra routine call that could be eliminated by
     moving it into the pp_def.

---- Fix kludged, info(q) and [o] oinfo(q) solution to getting q==9.
     (more on this in levmar.pd)

--- liblevmar has routines for testing the jacobian that are not implemented
    here.