package require Tcl 8.5 package require ral namespace import ::ral::* package require ralutil namespace import ::ralutil::*if 0 { Like in Relations as Stacks, we will be using a utility in ralutil to automatically add a consecutive integer identifier. }
::ralutil::sysIdsInitif 0 { In this design we will use a class to encapsulate relvars. There are a large number of relvar and relation operations that could be encapsulated as methods, but to keep the example small, I have only put in methods that are used by the example.Also, we use a bit of a trick here. Since relvars have their own namespace-like names, we can use the name of the object as the name of the relvar. Hence self refers to both the object command and the relvar name. }
oo::class create ooRelvar {
constructor {attrs ids} {
relvar create [self] [list Relation $attrs $ids]
}
destructor {
relvar unset [self]
}
method format {args} {
relformat [relvar set [self]] {*}$args
}
method Insert {tuple} {
relvar insert [self] $tuple
}
method Delete {args} {
relvar deleteone [self] {*}$args
}
}if 0 { As discussed in Relations as Stacks, the semantics of stacks and queues are closely related in relational terms to the manner in which the last or first element is selected. Expanding on that theme here, we introduce a class that imposes an order on a relvar. We will do that by adding an identifier consisting of a single incrementing integer and use the helpers in ralutil to set the identifier values. }
oo::class create RelOrdered {
superclass ooRelvar
constructor {args} {
# Add an attribute that will be a system generated identifier
lappend args __Id__ string
next $args __Id__
::ralutil::sysIdsGenSystemId [self] __Id__
}
destructor {
next
}
method Insert {tupleValue} {
# We have to make sure that the identifying attribute is
# mentioned, and its value will be set by a relvar trace.
lappend tupleValue __Id__ {}
next $tupleValue
}
method Delete {rel} {
next __Id__ [relation extract $rel __Id__]
}
method Item {rel} {
pipe {
relation eliminate $rel __Id__ |
relation tuple |
tuple get
}
}
# The difference between stack semantics and queue semantics
# is in the sorting order of the tag command. Sorting "descending"
# selects the last item placed in the relvar. Sorting "ascending"
# selects the first item placed in the relvar.
method Pick {order} {
pipe {
relvar set [self] |
relation tag ~ -$order __Id__ __Order__ |
relation choose ~ __Order__ 0 |
relation eliminate ~ __Order__
}
}
}if 0 { So now we can create a stack class based on relvars. The stack items are actually a dictionary of attribute name / attribute value pairs. This could be made simpler to just accept a string, but we get a little more structure for very little additional effort. }
oo::class create RelStack {
superclass RelOrdered
constructor {args} {
next {*}$args
}
destructor {
next
}
method push {args} {
my Insert $args
return
}
method pop {} {
set rel [my Pick descending]
my Delete $rel
my Item $rel
}
method peek {} {
my Item [my Pick descending]
}
}if 0 { The queue class is similar. The method names are chosen to be the conventional ones. }
oo::class create RelQueue {
superclass RelOrdered
constructor {args} {
next {*}$args
}
destructor {
next
}
method put {args} {
my Insert $args
return
}
method get {} {
set rel [my Pick ascending]
my Delete $rel
my Item $rel
}
method peek {} {
my Item [my Pick ascending]
}
}if 0 { So now for simple test runs. }RelStack create myStack Item string Name string myStack push Item "Item 1" Name A myStack push Item "Item 2" Name B puts [myStack format] puts "peek: [myStack peek]" puts "pop: [myStack pop]"if 0 {
+------+------+======+
|Item |Name |__Id__|
|string|string|string|
+------+------+======+
|Item 1|A |1 |
|Item 2|B |2 |
+------+------+======+
peek: Item {Item 2} Name B
pop: Item {Item 2} Name B}myStack push Item "Item 3" Name C myStack push Item "Item 4" Name D puts "pop: [myStack pop]" puts [myStack format] puts "pop: [myStack pop]"if 0 {
pop: Item {Item 4} Name D
+------+------+======+
|Item |Name |__Id__|
|string|string|string|
+------+------+======+
|Item 1|A |1 |
|Item 3|C |3 |
+------+------+======+
pop: Item {Item 3} Name C}myStack destroy RelQueue create myQueue Item string myQueue put Item "Item 1" myQueue put Item "Item 2" puts [myQueue format] puts "[myQueue get]" puts "[myQueue get]" myQueue destroyif 0 {
+------+======+
|Item |__Id__|
|string|string|
+------+======+
|Item 1|1 |
|Item 2|2 |
+------+======+
Item {Item 1}
Item {Item 2}So, it is possible to use relvars as the underlying data structure for an object oriented formulation of the interfaces. Now I'm trying to decide if I think it is a good idea or not. It will take a good bit more thought and experimentation. There does seem to be a fundamental tension between the usual object oriented attempts to encapsulate data and the relational approach to make all the data manipulation operations polymorphic with respect to the actual relation type. I'm also not sure where referential integrity constraints come into the approach. They are vitally important in relational terms, but don't seem to have an object oriented home.See also TclRAL, Relations as Stacks }
