This package contains a tool to verify that all operations in a given module are non-failing, i.e., their evaluation does not result in a failure, if they are called with appropriate arguments. The tool automatically infers abstract call types specifying supersets of appropriate arguments.
As an example, consider the following operation to compute the last element of a given list:
last :: [a] -> a
last [x] = x
last (_ : xs@(_:_)) = last xs
In the default mode (where types are abstracted into a set of allowed top-level constructors), the inferred call type is
last: {:}
specifying that last
does not fail if the argument is a
non-empty list, i.e., evaluable to some data term rooted by the list
constructor “:
”.
When an operation with such a restricted call type is used, the tool checks whether it is used in a context which ensures its call type. For instance, the operations
head (x:xs) = x
and
tail (x:xs) = xs
have the inferred call types
head: {:}
tail: {:}
When these operations are applied, it should be ensured (in a top-level functional computation) that the actual arguments are non-empty lists, as in the following operation:
readCommand :: IO ()
readCommand = do
putStr "Input a command:"
s <- getLine
let ws = words s
case null ws of True -> readCommand
False -> processCommand (head ws) (tail ws)
The tool automatically verifies that the operation
readCommand
is non-failing.
The ideas of the implementation of this tool are described in:
M. Hanus: Inferring Non-Failure Conditions for Declarative Programs, Proc. of the 17th International Symposium on Functional and Logic Programming (FLOPS 2024), Springer LNCS 14659, pp. 167-187, 2024, DOI: 10.1007/978-981-97-2300-3_10,
If the SMT solver Z3
is installed and the executable z3
can be found in the
path, the tool can also infer and verify arithmetic non-fail conditions.
For instance, it infers for the factorial function defined by
fac :: Int -> Int
fac n | n == 0 = 1
| n > 0 = n * fac (n - 1)
the non-fail condition
fac'nonfail :: Int -> Bool
fac'nonfail v1 = (v1 == 0) || (v1 > 0)
specifying that the argument must be non-negative in order to ensure
that fac
does not fail. With this information, it also
proves that the following I/O operation does not fail (where the
operation readInt
reads a line until it is an integer):
printFac :: IO ()
printFac = do
putStr "Factorial computation for: "
n <- readInt
if n<0 then putStrLn "Negative number, try again" >> printFac
else print (fac n)
If you downloaded the sources of this package, then the tool can be installed via the command
> cypm install
in the main directory of this package. Otherwise, the tool can also be downloaded and installed by the command
> cypm install verify-non-fail
This installs the executable curry-calltypes
in the
bin-directory of CPM.
If one uses KiCS2 to install this tool, one should use version 3.1.0 of April 11, 2024 (or newer) due to a memory leak in an older version of KiCS2.
In order to support a modular analysis of applications consisting of
several modules, the tool caches already computed analysis results of
modules in under the directory ~/.curry_verify_cache
. This
path is defined in Verify.Files.getVerifyCacheDirectory
. To
store the analysis results for different Curry systems and abstract
domains separately, the analysis results for a Curry module which is
stored in a file with path MODULEPATH.curry
are contained
in
~/.curry_verify_cache/CURRYSYSTEMID/DOMAINID/MODULEPATH-*
For instance, the call types of the prelude for KiCS2 Version 3.1.0
w.r.t. the abstract domain Values
are stored in
~/.curry_verify_cache/kics2-3.1.0/Values/opt/kics2/lib/Prelude-CALLTYPES
provided that KiCS2 is installed at /opt/kics2
so that
the prelude is contained in the file
/opt/kics2/lib/Prelude.curry
.
All cache files (for the Curry system used to install this tool) can be deleted by the command
curry-calltypes --delete