<< §5.3 Implicit team activation | ↑ Table of Contents ↑ | §5.5 Unanticipated team activation >> |
§5.4 Guard predicates
The effect of callins can further be controlled using so called guard predicates. Guards appear at four different levels:
- callin method binding
- role method
- role class
- team class
Guards can be specified as regular guards or base
guards,
which affects the exact point in the control flow, where the guard will be evaluated.
(a) General syntax for guards
A guard is declared using the keyword when
followed by a
boolean expression in parentheses:
when (predicateExpression)
Depending on the kind of guard different objects are in scope using
special identifiers like this
, base
.
Any predicate expression that evaluates to true
enables
the callin binding(s) to which it applies.
Evaluation to false
disables the callin binding(s).
(b) No side effects
A guard predicate should have no side effects. A compiler should optionally check this condition, but inter-procedural analysis actually depends on the availability of appropriate means to mark any method as side-effect free.
(c) Exceptions
A guard predicate should not throw any exceptions.
Yet, any exception thrown within a guard predicate cause the guard to evaluate to false
rather than propagating the exception, meaning that the evaluation of a guard predicate will never
interrupt the current base behaviour.
A compiler should flag any checked exception that is thrown within a guard.
Such diagnosis should by default be treated as an error, with the option of configuring
its severity to warning or ignore.
§5.4.1 Regular guards
↑ §5.4
This group of guards evaluates within the context of a given role. These guards are evaluated after a callin target is lifted and before a callin bound role method is invoked.
(a) Method binding guards
A guard may be attached to a callin method binding as in:
void roleMethod(int ir) <- after void baseMethod(int ib) when (ir > MyTeam.this.threshold);
Such a guard only affects the callin binding to which it is attached,
i.e., this specific callin binding is only effective,
if the predicate evaluates to true
.
The following values are within the scope of the predicate expression,
and thus can be used to express the condition:
- The role instance denoted by
this
.
Features of the role instance can also be accessed relative tothis
with or without explicit qualifyingthis
. - The team instance denoted by a qualified this reference as in
MyTeam.this
. - If the callin binding includes signatures (as in the example above):
Parameters of the role method.
If parameter mappings are involved, they will be evaluated before evaluating the guard.
(b) Method guards
A method guard is similar to a method binding guard, but it applies
to all callin method bindings of this method.
A method guard is declared between the method signature and the method body:
void roleMethod(int ir) when (ir > MyTeam.this.threshold) { body statements }
(c) Role level guards
When a guard is specified at the role level, i.e., directly before the class body of a role class, it applies to all callin method bindings of the role class:
protected class MyRole when (value > MyTeam.this.threshold) { int value; other class body declarations }
The following values are within the scope of the predicate expression:
- The role instance denoted by
this
(explicit or implicit, see above). Thus, in the examplevalue
will be interpreted as a field of the enclosing role. - The team instance denoted by a qualified this reference as in
MyTeam.this
(d) Team level guards
A guard specified in the header of a team class may disable the callin
bindings of all contained role classes. The syntax corresponds to the syntax
of role level guards.
The only value directly available within team level guard is the
team instance (denoted by this
) and its features.
Of course all guards can also access any visible static feature of a visible class.
Even if a guard has no direct effect, because, e.g., a role class has no callin bindings (maybe not even a role-base binding), predicates at such abstract levels are useful, because all predicates are inherited by all sub classes (explicit and implicit).
§5.4.2 Base guards
↑ §5.4
The intention behind base guards is to prevent lifting of a callin-target
if a guard evaluates to false
and thus refuses to invoke the
callin bound role method. Using base guards it is easier to prevent any
side-effects caused by a callin binding, because lifting could cause side-effects
at two levels:
- Creating a role on-demand already is a side-effect (observable e.g.
by the reflective function
hasRole (§6.1)
) - Role creation triggers execution of a role constructor (see custom lifting constructor (§2.3.1.(c))) which could produce arbitrary side-effects.
Both kinds of side-effects can be avoided using a base guard which prevents unnecessary lifting.
Any guard (5.4.1 (b)-(e)) can be turned into a base guard by adding
the modifier base
as in:
protected class MyRole playedBy MyBase base when (base.value > MyTeam.this.threshold) { class body declarations }
However, different scoping rules apply for the identifiers that can be used in a base guard:
(a) Base object reference
In all base guard predicates the special identifier base
can be used to denote the base object that is about to be lifted.
(b) Method binding guards
A base method binding guard may access parameters as passed to the
base method. Parameter mappings are not considered.
Additionally, for after
callin bindings, the identifier result
may be used to refer to the result of the base method (if any).
Note:
In order to achieve the same effect of accessing the base method's result, a regular binding guard (not a base guard) must use a suitable parameter mapping (see §4.4.(c)).
(c) Method guards
In contrast to regular method guards, a base guard attached to a role method cannot access any method parameters. See the next item (d) for values that are actually in scope.
(d) Role level guards
Role level base guards may use these values:
- The base instance using the special identifier
base
. - The team instance using a qualified this references (
MyTeam.this
).
(e) Team level guards
Team level base guards have the same scope as role level base guards (d).
However, the type of the role instance is not known here, i.e., here base
has the static type java.lang.Object
.
(f) Unbound roles
In contrast to regular guards, base guards cannot be attached to
unbound role classes nor to their methods.
Only team level base guards are independent of role binding.

§5.4.3 Multiple guards
↑ §5.4
Due to the different ranges of applicability different guards may affect the same method binding.
In that case all applicable guards are conjoined using a logical and
.
Any guard is interpreted as the conjunction of these predicates (if present):
- The direct predicate expression of the guard.
- The next outer guard along the chain method binding -> method -> role level -> team level
- The guard at the same level that is inherited from the implicit super role.
- The guard at the same level that is inherited from the explicit super role.
Example code (Guard Predicates):
1 | public team class ATM { |
2 | private Bank myBank; |
3 | public class ForeignAccount playedBy Account |
4 | base when (!ATM.this.myBank.equals(base.getBank())) |
5 | { |
6 | callin void debitWithFee(int amount) { |
7 | base.debitWithFee(fee+amount); |
8 | } |
9 | void debitWithFee(int i) <- replace void debit(int amount) |
10 | base when (amount < 1000); |
11 | } |
12 | } |
<< §5.3 Implicit team activation | ↑ Table of Contents ↑ | §5.5 Unanticipated team activation >> |
Effects:
The team in this example causes that an additional fee has to be payed while debiting less than 1000 Euros from a "foreign" account.Account
objects only getForeignAccount
roles, if they belong to a different bank than the surroundingATM
team.It accesses the bank of the base via the
base
identifier.debitWithFee
to calls where the base method argumentamount
is lower than 1000.Account.debit
causes a replace callin todebitWithFee
only if both predicates evaluate to true.