XSB Src (SVN)-old

Micro optimization of lattice and fold aggregation in case of all true
answers. This gives maybe a 5-8% improvement in time for intensive
aggregations of true answers.

Fixed aggregation so lattice and fold now work with undefined answers.
This restores the functionality that Teri introduced, but in which we
found a bug. It also extends it to fold. It detects if no undefs
occur and uses the earlier more efficient version in that case. There
is some overhead for the checking but it is small compared to always
assuming undefineds occur.

It should be noted that producing all possible undef answers can be
combinatorically explosive, so care should be taken when using this
option.

Removed code that hasn't been used in years (old_excess_vars/4)

memory_xsb.c: print warning if double-freeing instead of crashing
biassert.c: clean up after retracting to avoid reusing garbage and ultimately crashing.

Updated insert_psc to make set_env field conditional on whether
the symbol is defined in usermod or not.

fixed psc_insert to initialize a couple of psc fields: data and env_type
(for which T_IMPORT doesn't work, so used T_GLOBAL; hopefully OK.)

Fixed read_canonical and write_canonical to handle a() forms.

Note: read_canonical will *not* recognize a( ), i.e. with a whitespace
between the parens. Since this is read_canonical (and write_canonical
produces the accepted form,) I think this is OK. It's a more
complicated fix if we were to accept the form with the whitespace.

Fixed atom(a()) to fail, and compound(a()) to succeed.
(Had been intentionally programmed to be the opposite, for some reason.
System self-compiles and all tests pass with this change. It is now
consistent with SWI's implementation of the f() form.)

cleanup in dbdrivers package

Minimal documentation for table_index directive. (There may be a better
place in the manual for this? And it needs to be expanded if it stays.)