src/HOL/Tools/SMT/smt_translate.ML
Wed, 20 Jul 2011 15:09:53 +0200 boehmes removed debugging facilities accidentally left in the committed code
Wed, 20 Jul 2011 12:23:20 +0200 boehmes generalized lambda-lifting such that it is less specifically tailored for SMT (it does not anymore dependent on any SMT-specific code)
Wed, 20 Jul 2011 09:23:12 +0200 boehmes moved lambda-lifting on terms into a separate structure (for better re-use in tools other than SMT)
Thu, 14 Jul 2011 17:29:30 +0200 blanchet allow lambda-lifting without triggers
Sun, 26 Jun 2011 19:10:02 +0200 boehmes generalized introduction of explicit application constant: consider more functions as possible witness/instance of quantifiers than before (a constant of type T1 -> T2 -> T3 should be considered to have a rank less or equal to 1 if variables of type T2 -> T3 occur bound in a problem);
Wed, 22 Jun 2011 15:07:03 +0200 boehmes export lambda-lifting code as there is potential use for it within Sledgehammer
Tue, 14 Jun 2011 13:50:54 +0200 boehmes slightly more general treatment of mutually recursive datatypes;
Sun, 05 Jun 2011 15:00:17 +0200 boehmes made introduction of explicit application stable under removal of vacuous facts (which only lower the rank of constants but do not participate in the proof)
Fri, 08 Apr 2011 19:04:08 +0200 boehmes fixed eta-expansion: use correct order to apply new bound variables
Fri, 08 Apr 2011 19:04:08 +0200 boehmes unfold and eta-contract let expressions before lambda-lifting to avoid bad terms
Mon, 21 Feb 2011 09:14:48 +0100 boehmes wrap occurrences of quantifiers in first-order terms (i.e., outside first-order formulas) in If-expressions
Mon, 03 Jan 2011 16:22:08 +0100 boehmes re-implemented support for datatypes (including records and typedefs);
Mon, 20 Dec 2010 22:02:57 +0100 boehmes avoid ML structure aliases (especially single-letter abbreviations)
Sun, 19 Dec 2010 18:54:29 +0100 boehmes removed odd decoration of built-in symbols as Vars (instead provide built-in desctructor functions along with their inverse functions);
Fri, 17 Dec 2010 18:15:56 +0100 wenzelm merged
Fri, 17 Dec 2010 14:36:33 +0100 boehmes fixed lambda-lifting: shift indices of bound variables correctly (after locking the required bound variables) and apply bound variables to the new function symbol in the right order;
Fri, 17 Dec 2010 13:12:58 +0100 wenzelm tuned;
Thu, 16 Dec 2010 13:54:17 +0100 boehmes merged
Thu, 16 Dec 2010 13:34:28 +0100 boehmes fix lambda-lifting: take level of bound variables into account and also apply bound variables from outer scope
Thu, 16 Dec 2010 12:33:06 +0100 boehmes fixed introduction of explicit application function: bound variables always need explicit application if they are applied to some term
Thu, 16 Dec 2010 12:07:36 +0100 boehmes fixed eta-expansion: introduce a couple of abstractions at once
Thu, 16 Dec 2010 12:19:00 +0000 paulson merged
Thu, 16 Dec 2010 12:05:00 +0000 paulson made sml/nj happy
Wed, 15 Dec 2010 16:32:45 +0100 boehmes fixed checking and translation of weights (previously, weights occurring in terms were rejected, and weight numbers were unintended translated into Vars)
Wed, 15 Dec 2010 16:29:56 +0100 boehmes the functions term_of and prop_of are also needed in earlier stages, not only for Z3 proof reconstruction, so they really belong in SMT_Utils
Wed, 15 Dec 2010 14:29:04 +0100 blanchet workaround for bug in weight handling -- sometimes numerals got replaced by Vars and this confused the weight extractor
Wed, 15 Dec 2010 10:12:44 +0100 boehmes re-implemented eta-expansion, lambda-lifting, and explicit application on terms (exploiting the control over the term structure);
Wed, 15 Dec 2010 08:39:24 +0100 boehmes tuned
Tue, 07 Dec 2010 14:53:12 +0100 boehmes centralized handling of built-in types and constants;
Mon, 06 Dec 2010 15:38:02 +0100 boehmes tuned
less more (0) -30 tip