| Commit message (Collapse) | Author | Age |
|---|
| | |
|
| |
|
|
|
| |
We already have separate tests for those, and we want the output to be the same
with and without /autoTriggers.
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now, loops that may possibly
do an infinite number of iterations (that is, loops marked with 'decreases *') and calls to methods
marked with 'decreases *' are allowed only in methods that themselves are marked with 'decreases *'. As
before, ghost loops and ghost methods are not allowed to be marked with 'decreases *'.
Previously, 'decreases *' was allowed on a method only if the method was tail recursive;
this is no longer so. Note, however, that if the method is not tail recursive and engages
in infinite recursion, then it will eventually run out of stack space.
Previously, a 'decreases *' was not inherited in a refining module; this is no longer so.
That is, 'decreases *' is now inherited. To refine a possibly non-terminating method
or loop, the refining version simply provides a decreases clause that does not mention '*'.
Note that if the refined method is not recursive, it still needs to have _some_ decreases
clause in order not to inherit the 'decreases *' from the refined method, but the expression
stated in the decreases clause can be arbitrary (for example, one can write 'decreases true'
or 'decreases 7' or 'decreases x' for some 'x' in scope).
Note, in the new design, a method needs to be declared with 'decreases *' if
it may recurse forever _or_ if it contains a possibly infinite loop.
Note that the absence of 'decreases *' on a loop does not mean the loop will
terminate, but it does mean that the loop will iterate a finite number of times
(the subtle distinction here is that a loop without a 'decreases *' is allowed
to contain a nested loop that has a 'decreases *' -- provided the enclosing
method is also declared with 'decreases *', as previously mentioned).
|
| | |
|
| | |
|
| |
|
|
| |
previously was an alternative syntax
|
| |
|
|
|
|
|
|
|
|
| |
* started rewriting parsing of qualified identifiers in expressions
* annoyingly, had to introduce AST nodes for concrete syntax
* previous syntax for invoking datatype constructors: #List.Cons(h, t)
new syntax: List.Cons(h, t)
or, if only one datatype has a constructor named Cons: Cons(h, t)
* Removed type parameters for datatype constructors from the grammar
* Helped Test/VSI-Benchmarks/b4.dfy along with a couple of assertions (previously, its proving performance was highly varied)
|
| | |
|
| |
|
|
|
|
| |
* Add support for an {:induction} attribute on universal quantifiers over one bound variable. It causes the universally quantified formulas to be proved by induction.
* For a user-defined function F, introduce not just F and F#limited, but also F#2 (which sits "above" F, just as F sits "above" F#limited)
* In base case of SplitExpr, make use of F#2 functions (unless already inside an inlined predicate)
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* Big change: Add type and allocatedness information everywhere in the Boogie translation. This not only fixes some potential soundness problems (see Test/dafny1/TypeAntecedents.dfy), but it also gives more information about the program. On the downside, it also requires discharging more antecedents in order to use some axioms. Another downside is that overall performance has gone down (however, this may be just an indirect consequence of the change, as it was in one investigated case).
* Increase the applicability of function axioms (extending the coarse-grain function/module height mechanism used as an antecedent of function axioms). (Internally, this uses the new canCall mechanism.)
* Extend language with "allocated( Expr )" expressions, which for any type of expression "Expr" says that "Expr" is allocated and has the expected type.
* More details error messages about ill-defined expressions (internally, by using CheckWellformedness instead of "assert IsTotal")
* Add axioms about idempotence of set union and intersection
* The compiler does not support (the experimental feature) coupling invariants, so generate error if the compiler ever gets one
* In the implementation, combine common behavior of MatchCaseStmt and MatchCaseExpr into a superclass MatchCase
* Fixed error in translation of while(*)
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* If no decreases clause is given, the decreases clause defaults to the set of objects denoted by the reads clause, which was the previous Dafny behavior
* Made Dafny check loops for termination by default. Previously, this was done only if the loop had a decreases clause. To indicate that a loop is to be checked only for partial correctness, Dafny now allows "decreases *".
* Allow "reads *" to say that the function may read anything at all (sound, but not very useful)
* Adjusted frame axioms of functions to speak of allocated objects more liberally; and also added antecedents about the heaps being well-formed and the parameters being allocated
* Added some previously omitted well-definedness checks.
* Fixed some bugs in the resolver that caused some type errors not to be reported
* Added some messages to go with some (previously rather opaquely reported) errors
* Fixed some test cases that previously had ordered conjuncts incorrectly to prove termination and reads checks (such checks were previously omitted)
* Beefed up Test/dafny0/SchorrWaite.dfy to use datatypes to specify that no garbage gets marked. The full-functional total-correctness verification of this Schorr-Waite method now takes about 3.2 seconds.
|
| |
|
|
|
|
|
| |
"!(x in S)".
Changed Dafny test files to use the new operator.
Included the file b8.dfy into the VSI-Benchmarks test harness.
|
| |
|
qunyanm
Clément Pit--Claudel
Rustan Leino