GNATprove performs auto-active verification, that is, verification is done automatically, but usually requires annotations by the user to succeed. In SPARK, annotations are most often given in the form of contracts (pre and postconditions). But some language features, in particular ghost code, allow proof guidance to be much more involved. In a paper we are presenting at NASA Formal Methods symposium 2017, we describe how an imperative red black tree implementation in SPARK was verified using intensive auto-active verification.
A new feature of SPARK2014 allows to use a memcached server to share proof results between runs of the SPARK tools and even between developers on different machines. Check out this post to see the details.
It turns out that the CodePeer engine can be used as a powerful prover for SPARK programs. This feature will be available in the next version of SPARK Pro, make sure you try it out!
We report on the creation of the first lemma of a new lemma library on arrays: a lemma on transitivity of the order in arrays.
One of the most important challenges for SPARK users is to come up with adequate contracts and annotations, allowing GNATprove to verify the expected properties in a modular way. Among the annotations mandated by the SPARK toolset, the hardest to come up with are probably loop invariants. A previous post explains how GNATprove can automatically infer loop invariants for preservation of unmodified record components, and so, even if the record is itself nested inside a record or an array. Recently, this generation was improved to also support the simplest cases of partial array updates. We describe in this post in which cases GNATprove can, or cannot, infer loop invariants for preservation of unmodified array components.
One year ago, we presented on this blog what was provable about fixed-point and floating-point computations (the two forms of real types in SPARK). Since then, we have integrated static analysis in SPARK, and modified completely the way floating-point numbers are seen by SMT provers. Both of these features lead to dramatic changes in provability for code doing fixed-point and floating-point computations.
Tasking was one of the big features introduced in the previous release of SPARK 2014. However, GNATprove only supported tasking-related constructs allowed by the Ravenscar profile. Now it also supports the more relaxed GNAT Extended Ravenscar profile.
Type invariants are used to model properties that should always hold for users of a data type but can be broken inside the data type implementation. Type invariant are part of Ada 2012 but were not supported in SPARK until SPARK Pro 17.
Something that many developers do not realize is the number of run-time checks that occur in innocent looking arithmetic expressions. Of course, everyone knows about overflow checks and range checks (although many people confuse them) and division by zero. After all, these are typical errors that do show up in programs, so programmers are aware that they should keep an eye on these. Or do they?
Formal verification tools like GNATprove rely on the user to provide loop invariants to describe the actions performed inside loops. Though the preservation of variables which are not modified in the loop need not be mentioned in the invariant, it is in general necessary to state explicitly the preservation of unmodified object parts, such as record fields or array elements. These preservation properties form the loop’s frame condition. As it may seem obvious to the user, the frame condition is unfortunately often forgotten when writing a loop invariant, leading to unprovable checks. To alleviate this problem, the GNATprove tool now generates automatically frame conditions for preserved record components. In this post, we describe this new feature on an example.