version 2.0.0, November 14, 2017
================================
* integration of floating-point arithmetic reasoning: this is done via
the support of the rounding operator as done in the Gappa tool. FPA
reasoning can be enabled by calling Alt-Ergo as follows:
alt-ergo -use-fpa -prelude fpa-theory-2017-01-04-16h00.why file.why
where fpa-theory-2017-01-04-16h00.why is a prelude distributed with
Alt-Ergo. More details about the integration are given in this
paper: https://hal.inria.fr/hal-01522770, and dditional benchmarks
can be found here:
https://gitlab.com/OCamlPro-Iguernlala/Three-Tier-FPA-Benchs
* a new lightweight reasoning step before SAT solving that enables a
kind of backward reasoning/goal unfolding (can be disabled with
option -no-backward)
* integration of a simple cache mechanism for unit facts in the SAT
(learnt clauses, assumed facts, instances, theory deductions, ...)
to improve BCP (can be disabled with option -no-sat-learning)
* the code of the profiler is now integrated in Alt-Ergo and is
statically linked
* deep code refactoring. In particular, one can easily build an
Alt-Ergo library or define/register a new parser
* add the ability to parse preludes with -prelude ...
-prelude
* add the ability to disable weak pointers in hash-consing module
using option -disable-weaks (useful for more determinism)
* GUI: goals are now shown in positive form
* bugfix related to -timelimit option (use of ITIMER_VIRTUAL instead
of ITIMER_REAL except for the GUI)
* bugfix in UF related to normalization of terms in presence of AC
symbols
* new options -no-decisions and -no-fm to disable decisions in the SAT
and the Fourier-Motzkin algorithm, respectively
* new improvements and heuristics in the SAT, terms, formulas,
congruence closure, profiler, ...
* update licensing: most files are now licensed under the terms of the
Apache license v2. Some recent OCamlPro additions are licensed under
the terms of the OCamlPro Non-Commercial Purpose License v1
version 1.30, November 21, 2016
==============================
* experimental support for models generation (for more details, see
http://www.ocamlpro.com/2016/11/21/release-of-alt-ergo-1-30-with-experimental-support-for-models-generation/).
* simplify formulas of the form "forall x : t. x <> t or F" when t
does not contain x. The only relevant instance in this case is
"(x <> t or F) {x |-> t}". In a similar way, this allows to simplify
formulas of the form "exists x : t. x = t and F" to "F {x |-> t}".
* integration of general simplex in IntervalCalculus based on
ocplib-simplex.
* important parts of Intervals and IntervalCalculus have been
rewritten and improved.
* improve tightening of intervals for non-linear variables, and of the
CP-like loop.
* extend Fourier-Motzkin to deduce more bounds based on relational
dependencies of the form "c * x <= (Sum_i d_i * y_i) + e"
* various improvements in the default SAT solver, Term, Formula,
IntervalCalculus, Matching, Numbers, case-split, ...
* various (soundness and/or completeness) bug fixes in Uf, Intervals,
IntervalCalculus, Theory, Formula, type-checking, Triggers,
fm-simplex and satML, ...
* new option "-inst-after-bj" that forces an instantiation round after
each backtracking/backjumping in the default SAT solver.
* new option "-no-backjumping" that disables backjumping in the
default SAT solver (for debugging).
* new option "-fm-cross-limit" that controls the number of
intermediate inequalities produces by Fourier-Motzking. Default
value is 10,000.
* new option "-no-ac" to be able to deactivate the AC theory
* new option "-no-NLA" that disables non-linear arithmetic reasoning
(*, /, %). Non-linear multiplication remains AC.
version 1.20, February 09, 2016
==============================
* default value of option -age-bound increased from 10 to 50
* bugfix when using option --all-models
* improvements in Matching: a set of instances may cause more than one
conflict. It's better to detect the more precise one
* remove case-split limit for the theory of arrays
* bugfix in the profiler plugin
* improvements in the default SAT solver
version 1.10, October 19, 2015
==============================
* Improvements and bug fixes in the satML plugin
* Improvement of the solver of linear arithmetic in presence of
non-linear arithmetic parts in the equations to be solved
* Many improvements in case-split analysis. In particular, this is now
performed by explicit calls from the SAT solver. Its behavior can be
controlled with the new option "-case-split-policy". Possible values
for this option are: "after-theory-assume" (default),
"before-matching", and "after-matching".
* Alt-Ergo can now be called on zipped files (eg. .mlw.zip or
.why.zip). Decompression is done on the fly with camlzip.
* Soundness bug fixes in arithmetic, interval calculus, union-find,
...
* Code reviewing/refactoring/rewriting/optimization in many parts of
the solver
version 1.00, January 29, 2015
==============================
* General Improvements:
** theories data structures: semantic values (internal theories
representation of terms) are now hash-consed. This enables the
use of hash-based comparison (instead of structural comparison)
when possible
** theories combination: the dispatcher component, that sends
literals assumed by the SAT solver to different theories
depending on whether these literals are equalities,
disequalities or inequalities, has been re-implemented. The new
code is much simpler and enables new optimizations and
factorizations
** case-split analysis: we made several improvements in the
heuristics of the case-split analysis mechanism over finite
domains
** explanations propagation: we improved explanations propagation
in congruence closure and linear arithmetic algorithms. This
makes the proofs faster thanks to a better back-jumping in the
SAT solver part
** linear integer arithmetic: we re-implemented several parts of
linear arithmetic and introduced important improvements in the
Fourier-Motzkin algorithm to make it run on smaller sub-problems
avoiding some redundant executions. These optimizations allowed
a significant speed up on our internal benchmarks
** data structures: we optimized hash-consing and some functions in
the "formula" and "literal" modules
** SAT solving: we made a lot of improvements to the default
SAT-solver and to the SatML plugin. In particular, the decision
procedure part now receives facts (literals) in bulk from the
solvers, rather than individually, trimming the costs of
intermediate calls to theories reasoners, such as
Fourier-Motzkin
** Matching: we extended the E-matching algorithm to also perform
matching modulo the theory of records. In addition, we
simplified matching heuristics and optimized the E-matching
process to avoid computing the same instances several times
** Memory management: thanks to the ocp-memprof tool
(http://memprof.typerex.org/), we identified some parts of
Alt-Ergo that needed some improvements in order to avoid useless
memory allocations, and thus unburden the OCaml garbage
collector
** the function that retrieves the used axioms and predicates (when
option 'save-used-context' is set) has been improved
* Bug Fixes:
** 6 in the "inequalities" module of linear arithmetic
** 4 in the "formula" module
** 3 in the "ty" module used for types representation and
manipulation
** 2 in the "theories front-end" module that interacts with the
SAT solvers
** 1 in the "congruence closure" algorithm
** 1 in "existential quantifiers elimination" module
** 1 in the "type-checker"
** 1 in the "AC theory" of associative and commutative function
symbols
** 1 in the "union-find" module
* New OCamlPro Plugins:
** profiling plugin: when activated, this plugin records and prints
some information about the current execution of Alt-Ergo every 'n'
seconds: In particular, one can observe a module being activated,
a function being called, the amount of time spent in every
module/function, the current decision/instantiation level, the
number of decisions/instantiations that have been made so far, the
number of case-splits, of boolean/theory conflicts, of assumptions
in the decision procedure, of generated instances per axiom, ...
** fm-simplex plugin: when activated, this plugin is used instead of
the Fourier-Motzkin method to infer bounds for linear integer
arithmetic affine forms (which are used in the case-split analysis
process). This module uses the Simplex algorithm to simulate
particular runs of Fourier-Motzkin. In practice, the new algorithm
scales better on linear integer arithmetic problems containing
lots of inequalities
* New Options:
-version-info: prints some information about this version of
Alt-Ergo (release and compilation dates, release commit ID)
-no-theory: deactivate theory reasoning. In this case, only the
SAT-solver and the matching parts are used
-inequalities-plugin: specify a plugin to use, instead of the
"default" Fourier-Motzkin algorithm, to handle inequalities of
linear arithmetic
-tighten-vars: when this option is set, the Fm-Simplex plugin will
try to infer bounds for integer variables as well. Note that this
option may be very expensive
-profiling-plugin: specify a profiling plugin to use to monitor an
execution of Alt-Ergo
-profiling : makes the profiling module prints its information
every seconds
-no-tcp: deactivate constraints propagation modulo theories
* Removed Capabilities:
** the pruning module used in the frontend is not available anymore
** the SMT and SMT2 front-ends have been removed. We plan to implement a
new front-end for SMT2 in upcoming releases
version 0.99.1, December 30, 2014
=================================
o the "SAT solving" part can now be delegated to an external plugin;
o new experimental SAT solver based on mini-SAT, provided as a
plugin. This solver is, in general, more efficient on ground
problems;
o heuristics simplification in the default SAT solver and in the
matching (instantiation) module;
o re-implementation of internal literals representation;
o improvement of theories combination architecture;
o rewriting some parts of the formulas module;
o bugfixes in records and numbers modules;
o new option "-no-Ematching" to perform matching without equality
reasoning (i.e. without considering "equivalence classes"). This
option is very useful for benchmarks coming from Atelier-B;
o two new experimental options: "-save-used-context" and
"-replay-used-context". When the goal is proved valid, the first option
allows to save the names of useful axioms into a ".used" file. The
second one is used to replay the proof using only the axioms listed
in the corresponding ".used" file. Note that the replay may fail
because of the absence of necessary ground terms generated by
useless axioms (that are not included in .used file) during the
initial run.
version 0.95.2, September 20th, 2013
====================================
o Alt-Ergo is now maintained and distributed by OCamlPro, while
academic research is conducted in partnership with the VALS team
(LRI),
o source code is reorganized into sub-directories,
o quantifiers instantiation heuristics are simplified,
o bug-fixes in matching, nums, records, sums,
o improvement of the GUI when opening big files.
version 0.95.1, March 05th, 2013
=================================
o bug fixes (existantial elimination, Euclidean division)
o minor enhancement (transformation of boolean equalities into equivalences)
o minor enhancement (sort axioms/definitions instances according to their size)
version 0.95, January 11th, 2013
=================================
+ Main changes in the solver:
-----------------------------
o new combination method for Shostak solvers
o improvement of non-linear multiplication distribution over addition
o input language extension: polymorphic declaration are now allowed (logic x: 'a)
o input language extension: the types of terms can now be forced in a formula using the construct : (see man for an example)
o input language modification: a label should be a string. The construct