Consensus on error handling with DLMs

[Randall Skelton <rhskelto(at)atm.ox.ac.uk>]

Hi all,

I've more or less finished writing an IDL interface to Postgres and I'm
now in the debugging stage. I thought I'd take a poll to see what people
think of appropriate error returns.  In this library I have a variety of
function returns... integers, floats, doubles, strings, complex structures
and so forth.  For integer returns, I usually default to giving the user a
message with the handle IDL_MSG_INFO in IDL_Message and returning -1 on
failure.  Is there a good protocol for signifying an error in strings,
structures and arrays?  Some of my default string returns are themselves
null strings (indicating that no data or message was found) so it wouldn't
be wise to simply return a null string on error.  I am also very reluctant
to return a float -1.0000 as testing for this can lead to problems with
IEEE number definitions in C.  For the moment, I am using the
IDL_MSG_LONGJMP to signal an error in all routines that don't return an
IDL integer.  It stops the interpreter immediately (which isn't
necessarily bad) as it signifies a major fault.  Comments?

Thanks,
Randall