E cient and Precise Sharing Domains for Logic Programs

Sharing information between logical variables is cru cial for a lot of analyses of logic programs e g free ness analysis detection of And parallelism and occur check Therefore the development of accurate sharing domains has attracted a lot of research The sharing domain JL of Jacobs Langen which represents sub stitutions by powersets of variables is considered one of the most precise sharing domains However it is too ine cient in practice lots of programs cannot be analyzed in reasonable time Improvements of JL by adding auxiliary information like linearity su er from the same ine ciency too To improve upon this situ ation we systematically derived a new sharing domain JL from JL which represents variables by downward closed powersets of variables We combined JL with the groundness domainPOS Both JL and the new do main JL POS have been implemented with the help of the Prolog analyzer generator GENA In order to study the impact of linearity we also implemented the abstract domains JL LIN and JL POS LIN The new domains are much more e cient as their counter parts JL and JL LIN respectively Even more impor tant they can analyze even largest real world programs in reasonable time Surprisingly the new sharing do mains seem to have the same precision than JL and JL LIN in practice