Difference between revisions of "Agents in SCOOP"

m (→‎Agent classes: Changed separateness status of TUPLE to be non-separate.)
m (Fix typo)
 
(7 intermediate revisions by one other user not shown)
Line 1: Line 1:
 
The page reflects current decisions taken for the current SCOOP implementation, in particular without support for processor tags. So it may change in the future.
 
The page reflects current decisions taken for the current SCOOP implementation, in particular without support for processor tags. So it may change in the future.
  
−
== Status of a target ==
+
== Status of agent targets ==
−
: Because a agent object is created on a processor associated with a target, this processor needs to be controlled.
+
: Because any agent object is created on a processor associated with the agent's target, that target has to be controlled.
−
'''Agent target validity rule.''' A target of an agent is controlled.
+
'''Agent target validity rule.''' The target of an agent must be controlled.
  
 
== Types ==
 
== Types ==
−
'''Agent argument types validity rule.''' Formal arguments of a feature used in an agent expression with a separate target are separate.
+
'''Agent argument types validity rule.''' If an agent has a separate target, every non-expanded argument of the associated feature must be separate.
−
: Because an agent is created on a processor of a target, without processor tags it's impossible to create an agent on an open target of separate type and then pass the corresponding target object at the agent call. Also there is no way to make an open target controlled.
+
: Because an agent is created on a processor of a target, without processor tags it is impossible to create an agent on an open target of separate type and then pass the corresponding target object at the agent call. Also there is no way to make an open target controlled (see previous rule).
−
'''Agent open target type validity rule.''' An open target of an agent is not separate.
+
'''Agent open target type validity rule.''' If the target of an agent is open, it may not be separate.
  
 
'''Example.''' <e>
 
'''Example.''' <e>
Line 21: Line 21:
 
</e>
 
</e>
  
−
: The first parameter is not separate because both a target and an agent object run on the same processor.
+
: The first parameter is not separate because the target and the agent object have the same handler.
 
: The separateness status of the type <e>TUPLE</e> used for arguments does not matter, what matters is the types of arguments. In many cases when <e>TUPLE</e> is used, it is used as a value type, i.e. it could even be an expanded type, so that a separate mark is not required. But then we need some conversion rules to attach different types of <e>TUPLE</e>. In order to avoid unnecessary processor dependency, it's going to be created on the target processor (as a routine object itself), therefore it is not <e>separate</e>.
 
: The separateness status of the type <e>TUPLE</e> used for arguments does not matter, what matters is the types of arguments. In many cases when <e>TUPLE</e> is used, it is used as a value type, i.e. it could even be an expanded type, so that a separate mark is not required. But then we need some conversion rules to attach different types of <e>TUPLE</e>. In order to avoid unnecessary processor dependency, it's going to be created on the target processor (as a routine object itself), therefore it is not <e>separate</e>.
  

Latest revision as of 13:08, 7 May 2013

The page reflects current decisions taken for the current SCOOP implementation, in particular without support for processor tags. So it may change in the future.

Status of agent targets

Because any agent object is created on a processor associated with the agent's target, that target has to be controlled.

Agent target validity rule. The target of an agent must be controlled.

Types

Agent argument types validity rule. If an agent has a separate target, every non-expanded argument of the associated feature must be separate.

Because an agent is created on a processor of a target, without processor tags it is impossible to create an agent on an open target of separate type and then pass the corresponding target object at the agent call. Also there is no way to make an open target controlled (see previous rule).

Agent open target type validity rule. If the target of an agent is open, it may not be separate.

Example. 
f (p: separate P; q: separate Q): R
t: separate T
...
r := agent t.f (c, ?)
x := r.item ([a])
The type t of agent t.f (c, ?) is 
separate FUNCTION [ANY, TUPLE [separate Q], separate R]
The first parameter is not separate because the target and the agent object have the same handler.
The separateness status of the type TUPLE used for arguments does not matter, what matters is the types of arguments. In many cases when TUPLE is used, it is used as a value type, i.e. it could even be an expanded type, so that a separate mark is not required. But then we need some conversion rules to attach different types of TUPLE. In order to avoid unnecessary processor dependency, it's going to be created on the target processor (as a routine object itself), therefore it is not separate.

Agent classes

The first parameter in the class ROUTINE and descendants is not used, so in the future it will be dropped altogether.

Example. The agent type from the example above becomes
separate FUNCTION [TUPLE [separate Q], separate R]

Correctness and wait conditions

The feature {ROUTINE}.precondition should be split into 2 parts: correctness_condition and wait_condition. Then the routine call looks like
call (a: OPEN_ARGS)
          require
              correctness_condition (a)
          do
              from
              until
                  not wait_condition (a)
              loop
                  execute_body (a)
              end
          end

The same change applies to the feature item.

Retrieved from "https://dev.eiffel.com/index.php?title=Agents_in_SCOOP&oldid=14754"