Update: fix the errors in Makefile. If you have any problem with compiling SpecLearn (DOrder), send an email to rowangithub@yahoo.com (without having to identify your name and institution) and I will try to make it right for you.

SpecLearn is a tool that automatically infers useful invariants for (higher-order) programs. SpecLearn finds invariants from program samples and represents invariants into refinement types.

SpecLearn is not dedicated to higher-order functional programs. It can infer and verify useful properties for array programs (with side-effects).

A recent extension of SpecLearn is able to prove the safety properties of over 200 loop programs and a suite of recursive programs (collected from relate tools), using numeric invariants.

Another extension of SpecLearn, which we call DOrder, can automatically infer expressive shape specifications beyond the scope of competing tools.

SpecLearn is still under very active development. Our next effort is to enable SpecLearn to verify (imperative) heap-manipulating programs.

A very old version of the virtual machine of SpecLearn is available at this link.

We will update the VM to reflect our progress on data structure verification soon.

1. The loop (numeric) program benchmarks are included in ./tests/folprograms/ directory.

    To try an example, run ./msolve.py -no_hoflag ./tests/folprograms/misc/popl07.ml
   

2. The recursive (numeric) program benchmarks are included in ./tests/recursive/ directory and ./tests/dml/. The latter directory contains benchmarks for learning numeric properties over data structure programs.

    To try an example, run ./msolve.py -no_hoflag ./tests/dml/bdd.ml
   

3. The higher-order program benchmarks are included in ./tests/mochi/ ./tests/lists/ and ./tests/popl13/ directories.

    To try an example, run ./msolve.py -hoflag ./tests/mochi/amax.ml
   

4. The data structure program benchmarks, for DOrder implementation, are included in ./tests/reachability/ directory.

    To try an example, run ./msolve.py -reachability -no_hoflag ./tests/reachability/reverselist.ml or
    					   ./msolve.py -no_hoflag -reachability ./tests/reachability/avl2.ml or
    				 	   ./msolve.py -no_hoflag -reachability ./tests/reachability/redblackset.ml
   

At this stage, we are improving the algorithms for checking array programs. As a result, the evaluation on benchmarks under ./tests/array/ is not stable. However, this part is not included in the paper, and the rest of the directories are also just for our own research purpose.

SpecLearn requires OCaml 3.12 and Z3 4.3.

Currently, the source code of SpecLearn can only be complied and run in Mac OS (with a proper Z3 distribution, SpecLearn can run in Linux).

The tool has some troubles to be compatible with OCaml 4.0+. We will improve our code soon.

In order to make sure that you have Z3 ready for the tool, we require users to manually

	Go into external/z3/ocaml, and run ./build-lib.sh /usr/local/lib/ocaml/
	please adjust the path to Ocaml library and install CamlIDL if necessary.

If you encounter any error, please follow ReadMe provided under external/z3/ocaml to install Z3. You are also required to put libz3.dylib (you need to compile Z3 to get this file) under external/z3/lib.

Then, to compile SpecLearn (DOrder), from the top directory:

    Run make libs && make

Note that our makefile assumes CamlIDL is installed under /usr/local/lib/ocaml. Please make adjustment if necessary.

Example programs are included in tests/.

1. Be sure that the files in external/Z3/lib are in your library path! One way to do this is to run, from the top directory,

    export DYLD_LIBRARY_PATH="external/z3/lib/:$DYLD_LIBRARY_PATH"
   

2. To check high-order functional programs, run

    ./msolve.py -hoflag [ML source file]
   

3. If you do not want the support for higher-order functions (for performance improvements), run

    ./msolve.py -no_hoflag [ML source file]
   

4. By default, the tool does not check order and reach properties, to infer shape specifications on top of order and reach properties for data structure programs, run

    ./msolve.py -no_hoflag -reachability [ML source file] 
   

5. If you are interested in the performance of our tool in learning numeric invariants over both pure numeric programs and data structure programs, run (with -no_hoflag)

    ./msolve.py -no_hoflag [ML source file]
   

You can observe the inferred function type from the terminal (standard output).