Author:bmeyer

Date:2014-02-12T17:17:34.000000Z


git-svn-id: https://svn.eiffel.com/eiffel-org/trunk@1260 abb3cda0-5349-4a8f-a601-0c33ac3a8c38
This commit is contained in:
bmeyer
2014-02-12 17:17:34 +00:00
parent 10bfd79ee3
commit ab3d8466ba

View File

@@ -5,7 +5,7 @@ The classes of the Iteration cluster encapsulate control structures representing
=Iterators and Agents=
The recent introduction of the agents mechanism in Eiffel offers an attractive alternative to the <eiffel>Iterator</eiffel> cluster of EiffelBase.
Eiffel's agent mechanism offers an attractive alternative to the <eiffel>Iterator</eiffel> cluster of EiffelBase.
=The Notion of iterator=
@@ -33,16 +33,16 @@ You can perform iterations without any special iteration classes. For example if
then a class <eiffel>SPECIAL_PROMOTION</eiffel> of a text processing system may include in one of its routines a loop of the form
<code>
from
customers.start
until
customers.exhausted
loop
if recent_purchases.has (customers.item>) then
target_list.put (customers.item>)
end
customers.forth
end</code>
from
customers.start
until
customers.exhausted
loop
if recent_purchases.has (customers.item>) then
target_list.put (customers.item>)
end
customers.forth
end</code>
Such schemes are quite common. But it is precisely because they occur frequently that it is useful to rely on library classes to handle them. One of the principal tasks of object-oriented software development is to identify recurring patterns and build reusable classes that encapsulate them, so that future developers will be able to rely on ready-made solutions. <br/>
The classes of the Iteration library address this need. Using them offers two benefits:
@@ -51,55 +51,50 @@ The classes of the Iteration library address this need. Using them offers two be
=Simple Examples=
To get a first grasp of how one can work with the Iteration library, let us look at a typical iteration class and a typical iteration client.
==An example iterator routine==
Here, given with its full implementation, is a typical Iteration library routine: the procedure until_do from [[ref:libraries/base/reference/linear_iterator_chart]] , the class defining iteration mechanisms on linear (sequential) structures.
<code>
until_do
-- Apply action to every item of target,
-- up to but excluding first one satisfying test.
-- (Apply to full list if no item satisfies test.)
require
traversable_exists: target /= Void
do
from
target.start
invariant
''invariant_value''
until
target.exhausted or else test
loop
action
target.forth
end
ensure
achieved: target.exhausted or else test
invariant_satisfied: ''invariant_value''
end</code>
until_do
-- Apply action to every item of target,
-- up to but excluding first one satisfying test.
-- (Apply to full list if no item satisfies test.)
require
traversable_exists: target /= Void
do
from
target.start
invariant
''invariant_value''
until
target.exhausted or else test
loop
action
target.forth
end
ensure
achieved: target.exhausted or else test
invariant_satisfied: ''invariant_value''
end</code>
The precise form of the procedure in the class relies on a call to another procedure, until_continue, and on inherited assertions. Here everything has been unfolded for illustration purposes. <br/>
This procedure will traverse the linear structure identified by target and apply the procedure calledaction on every item up to but excluding the first one satisfying test. <br/>
The class similarly offers <eiffel>do_all</eiffel>, <eiffel>do_while</eiffel>, <eiffel>do_for</eiffel>, <eiffel>do_if</eiffel> and other procedures representing the common control structures. It also includes functions such as <eiffel>exists</eiffel> and <eiffel>forall</eiffel>, corresponding to the usual quantifiers. <br/>
These iteration schemes depend on the procedure <eiffel>action</eiffel>, defining the action to be applied to successive elements, and on the function <eiffel>test</eiffel>, defining the boolean query to be applied to these elements. These features are declared in class [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] (the highest-level deferred class of the Iteration library); here is <eiffel>test</eiffel>:
<code>
test: BOOLEAN
-- Test to be applied to item at current position in
-- target (default: value of item_test on item)
require
traversable_exists: target /= Void
not_off: not target.off
do
Result := item_test (target.item>)
ensure
not_off: not target.off
end</code>
test: BOOLEAN
-- Test to be applied to item at current position in
-- target (default: value of item_test on item)
require
traversable_exists: target /= Void
not_off: not target.off
do
Result := item_test (target.item>)
ensure
not_off: not target.off
end</code>
This indicates that the value of the boolean function <eiffel>test</eiffel> will be obtained by applying <eiffel>item_test</eiffel> to the item at the current position in the target structure. In [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] , function <eiffel>item_test</eiffel> always return ; descendant classes will redefine it so as to describe the desired test. Similarly, <eiffel>action</eiffel> is declared in class [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] as a call to <eiffel>item_action</eiffel>. Descendants will redefine <eiffel>item_action</eiffel>, which as initially declared in [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] is a procedure with a null body. <br/>
Going through <eiffel>item_action</eiffel> and <eiffel>item_test</eiffel> provides an extra degree of flexibility. Normally the action and test performed at each step apply to <eiffel>target</eiffel> <code> . </code><eiffel>item></eiffel>, so that it suffices to redefine the <eiffel>item_features</eiffel>. This is the case with all examples studied in this chapter. In a more general setting, however, you might need to redefine <eiffel>action</eiffel> and <eiffel>test</eiffel> themselves.
@@ -109,50 +104,50 @@ Going through <eiffel>item_action</eiffel> and <eiffel>item_test</eiffel> provid
Here now is an example illustrating the use of these mechanisms. The result will enable us to resize all the paragraphs of a text up to the first one that has been modified - as we might need to do, in a text processing system, to process an interactive user request. Assume a class <eiffel>TEXT</eiffel> that describes lists of paragraphs with certain additional features. The example will also assume a class <eiffel>PARAGRAPH</eiffel> with a procedure <eiffel>resize</eiffel>, and a boolean-valued attribute <eiffel>modified</eiffel> which indicates whether a paragraph has been modified. Class <eiffel>TEXT</eiffel> inherits from [[ref:/libraries/base/reference/linked_list_chart|LINKED_LIST]] and so is a descendant of [[ref:/libraries/base/reference/linear_chart|LINEAR]] :
<code>
class
TEXT
TEXT
inherit
LINKED_LIST [PARAGRAPH]
...
LINKED_LIST [PARAGRAPH]
...
feature
...
...
end</code>
In a class <eiffel>TEXT_PROCESSOR</eiffel>, you can use an iteration procedure to write a very simple procedure <eiffel>resize_ paragraphs</eiffel> that will resize all paragraphs up to but excluding the first one that has been modified:
<code>
class
TEXT_PROCESSOR
TEXT_PROCESSOR
inherit
LINEAR_ITERATOR [PARAGRAPH]
redefine
item_action, item_test
end
LINEAR_ITERATOR [PARAGRAPH]
redefine
item_action, item_test
end
feature
resize_paragraphs (t: TEXT)
-- Resize all the paragraphs of t up to but excluding
-- the first one that has been modified.
do
set (t)
until_do
end
resize_paragraphs (t: TEXT)
-- Resize all the paragraphs of t up to but excluding
-- the first one that has been modified.
do
set (t)
until_do
end
feature {NONE}
item_test (p PARAGRAPH): BOOLEAN
-- Has p been modified?
do
Result := p.modified
end
item_test (p PARAGRAPH): BOOLEAN
-- Has p been modified?
do
Result := p.modified
end
item_action (p: PARAGRAPH)
-- Resize p.
do
p.resize
end
...
item_action (p: PARAGRAPH)
-- Resize p.
do
p.resize
end
...
end</code>
Thanks to the iteration mechanism, the procedure <eiffel>resize_paragraphs</eiffel> simply needs two procedure calls:
@@ -208,48 +203,48 @@ An iterator will always apply to a certain target structure. The target is intro
Both the iterator classes and the traversal classes are generic, with a formal generic parameter G. The actual generic parameters will be matched through the choice of iteration target: for a generic derivation of the form <eiffel>SOME_ITERATOR</eiffel> [<eiffel>ACTUAL_TYPE</eiffel>] the target can only be of type <eiffel>SOME_TRAVERSABLE</eiffel> [<eiffel>ACTUAL_TYPE</eiffel>] for the same <eiffel>ACTUAL_TYPE</eiffel>, where <eiffel>SOME_TRAVERSABLE</eiffel> is the traversal class matching <eiffel>SOME_ITERATOR</eiffel> according to the preceding table ([[ref:/libraries/base/reference/linear_chart|LINEAR]] for [[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] and so on), or one of its proper descendants. <br/>
Each of the proper descendants of [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] redefines the type of target to the matching proper descendant of [[ref:/libraries/base/reference/traversable_chart|TRAVERSABLE]] , to cover more specific variants of the iteration target, For example in [[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] the feature is redefined to be of type [[ref:/libraries/base/reference/linear_chart|LINEAR]] . [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] also introduces the procedure for selecting a target:
<code>
set (s: like target)
-- Make s the new target of iterations.
require
s /= Void
do
target := s
ensure
target = s
target /= Void
end</code>
set (s: like target)
-- Make s the new target of iterations.
require
s /= Void
do
target := s
ensure
target = s
target /= Void
end</code>
Next [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] introduces the routines describing the elementary action and test that will be applied to items of the iteration targets:
<code>
action
-- Action to be applied to item at current position in
-- target.
-- (default: item_action on item at current position.)
-- Note: for iterators to work properly, redefined
-- versions of this feature should not change the
-- traversable structure.
require
traversable_exists: target /= Void
not_off: not target.off
invariant_satisfied: invariant_value
do
item_action (target.item>)
ensure
not_off: not target.off
invariant_satisfied: invariant_value
end
action
-- Action to be applied to item at current position in
-- target.
-- (default: item_action on item at current position.)
-- Note: for iterators to work properly, redefined
-- versions of this feature should not change the
-- traversable structure.
require
traversable_exists: target /= Void
not_off: not target.off
invariant_satisfied: invariant_value
do
item_action (target.item>)
ensure
not_off: not target.off
invariant_satisfied: invariant_value
end
test: BOOLEAN
-- Test to be applied to item at current position in
-- target (default: value of item_test on item)
require
traversable_exists: target /= Void
not_off: not target.off
do
Result := item_test (target.item>)
ensure
not target.off
end</code>
test: BOOLEAN
-- Test to be applied to item at current position in
-- target (default: value of item_test on item)
require
traversable_exists: target /= Void
not_off: not target.off
do
Result := item_test (target.item>)
ensure
not target.off
end</code>
These routines rely on two others, <eiffel>item_action</eiffel> and <eiffel>item_test</eiffel>, which both take an argument of type G, the formal generic parameter. The reason, already noted above, is that in a vast majority of cases the iterated action and test solely depend, at each step of the traversal, on the item (of type G) at the current position. To define an iteration process, then, it suffices to redefine<eiffel> item_action</eiffel> and <eiffel>item_test</eiffel> in a descendant of the appropriate iteration class. Only in complex cases will it be necessary to redefine <eiffel>action</eiffel> and <eiffel>test</eiffel> themselves. <br/>
If you encounter such a case, note the caveat about action changing the target's structure. Understandably enough, an iterator that attempts to change the data structure while traversing it may engage in strange behavior. No such risk exists if you only redefine <eiffel>item_action</eiffel>, which may change the contents of items but not the structure itself. <br/>
@@ -272,38 +267,37 @@ All these features, and most of the other iteration features introduced in prope
[[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] , an effective class, refines the iteration mechanisms for cases in which the target is a linear structure, such as a list in any implementation or a circular chain. <br/>
The class effects all the deferred features inherited from [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] , taking advantage of the linear traversal mechanisms present in the corresponding traversal class, [[ref:/libraries/base/reference/linear_chart|LINEAR]] . Here for example is the effecting of <eiffel>do_if</eiffel>:
<code>
do_if
-- Apply action to every item of target satisfying
-- test.
do
from
target.start
invariant
invariant_value
until
target.exhausted
loop
if test then
action
end
forth
end
end</code>
do_if
-- Apply action to every item of target satisfying
-- test.
do
from
target.start
invariant
invariant_value
until
target.exhausted
loop
if test then
action
end
forth
end
end</code>
This routine text relies on features <eiffel>start</eiffel>, <eiffel>forth</eiffel> and <eiffel>exhausted</eiffel> which, together with <eiffel>off</eiffel>, have for convenience been carried over to [[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] from their counterparts in [[ref:/libraries/base/reference/linear_chart|LINEAR]] , with feature declarations such as
<code>
off: BOOLEAN
-- Is position of target off?
require
traversable_exists: target /= Void
do
Result := target.off
end</code>
off: BOOLEAN
-- Is position of target off?
require
traversable_exists: target /= Void
do
Result := target.off
end</code>
and similarly for the others. <br/>
In addition to effecting the general iteration features from [[ref:/libraries/base/reference/iterator_chart|ITERATOR]] , class [[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] introduces iteration features that apply to the specific case of linear structures:
* <eiffel>search </eiffel> <code> ( b :BOOLEAN) moves the iteration to the first position satisfying test if b is true, or not satisfying test if b </code> is false. This use of a boolean argument to switch between two opposite semantics is not part of the recommended style, and you will find few if any other examples in the Base libraries. Here, however, it was deemed preferable to the alternative, which would have involved four separate procedures (if together with <eiffel>search</eiffel> we consider <eiffel>continue_search</eiffel> discussed next).
* <eiffel>search (b :BOOLEAN)</eiffel> moves the iteration to the first position satisfying test if <code>b</code> is true, or not satisfying test if <code>b</code> is false. This use of a boolean argument to switch between two opposite semantics is not part of the recommended style, and you will find few if any other examples in the Base libraries. Here, however, it was deemed preferable to the alternative, which would have involved four separate procedures (if together with <eiffel>search</eiffel> we consider <eiffel>continue_search</eiffel> discussed next).
* With a linear structure we can implement an iteration corresponding to the 'for' loop of traditional programming languages, defined by three integers: the starting position, the number of items to be traversed, and the step between consecutive items. This is provided by procedure <eiffel>do_for</eiffel> <code> ( starting , number , step :INTEGER). </code>
* Since with a linear target the iterator can advance the cursor step by step, the basic iteration operations are complemented by variants which pick up from the position reached by the last call: <eiffel>continue_until</eiffel>, <eiffel>until_continue</eiffel>, <eiffel>continue_while</eiffel>, <eiffel>while_continue</eiffel>, <eiffel>continue_search</eiffel>, <eiffel>continue_for</eiffel>.
@@ -316,70 +310,70 @@ Contrary to what one might at first imagine, class [[ref:/libraries/base/referen
The trick is to use repeated inheritance. [[ref:/libraries/base/reference/two_way_chain_iterator_chart|TWO_WAY_CHAIN_ITERATOR]] inherits twice from [[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] ; the first inheritance branch yields the forward iteration features, the second yields those for backward iteration. There is no need for any explicit declaration or redeclaration of iteration features. Here is the entire class text that yields this result:
<code>
class
TWO_WAY_CHAIN_ITERATOR [G]
TWO_WAY_CHAIN_ITERATOR [G]
inherit
LINEAR_ITERATOR [G]
redefine
target
select
start,
forth,
do_all,
until_do,
do_until,
do_if,
do_for,
search,
forall,
exists,
until_continue,
continue_until,
continue_for,
continue_search
end
LINEAR_ITERATOR [G]
redefine
target
select
start,
forth,
do_all,
until_do,
do_until,
do_if,
do_for,
search,
forall,
exists,
until_continue,
continue_until,
continue_for,
continue_search
end
LINEAR_ITERATOR [G]
rename
start as finish,
forth as back,
do_all as do_all_back,
until_do as until_do_back,
do_until as do_until_back,
do_if as do_if_back,
do_for as do_for_back,
search as search_back,
forall as forall_back,
exists as exists_back,
until_continue as until_continue_back,
continue_until as continue_until_back,
continue_for as continue_for_back,
continue_search as continue_search_back
redefine
target
end
LINEAR_ITERATOR [G]
rename
start as finish,
forth as back,
do_all as do_all_back,
until_do as until_do_back,
do_until as do_until_back,
do_if as do_if_back,
do_for as do_for_back,
search as search_back,
forall as forall_back,
exists as exists_back,
until_continue as until_continue_back,
continue_until as continue_until_back,
continue_for as continue_for_back,
continue_search as continue_search_back
redefine
target
end
feature -- Status report
target: BI_LINEAR [G]
-- The structure to which iteration features will
-- apply
target: BI_LINEAR [G]
-- The structure to which iteration features will
-- apply
feature -- Cursor movement
finish
-- Move cursor of target to last position.
do
target.finish
end
finish
-- Move cursor of target to last position.
do
target.finish
end
back
-- Move cursor of target backward one position.
do
target.back
end
back
-- Move cursor of target backward one position.
do
target.back
end
end
</code>
</code>
This class provides a good example of the economy of expression that the full inheritance mechanism affords through the combination of renaming, redefinition, repeated inheritance rules and selection, without sacrificing clarity and maintainability.
@@ -410,72 +404,72 @@ The next two techniques will remove this limitation.
One way to obtain several iteration schemes is a simple extension to the single descendant technique. You can use repeated inheritance to provide two or more variants. We have in fact already encountered the technique when studying how to derive [[ref:/libraries/base/reference/two_way_chain_iterator_chart|TWO_WAY_CHAIN_ITERATOR]] and [[ref:/libraries/base/reference/cursor_tree_iterator_chart|CURSOR_TREE_ITERATOR]] from [[ref:/libraries/base/reference/linear_iterator_chart|LINEAR_ITERATOR]] . The general pattern, applied here to just two iteration schemes but easily generalized to more, is straightforward:
<code>
class
DUAL_PROCESSOR
DUAL_PROCESSOR
inherit
LINEAR_ITERATOR [SOME_TYPE]
rename
item_action as action1,
item_test as test1,
do_if as do_if1,
redefine
action1, test1
select
action1, test1
end
LINEAR_ITERATOR [SOME_TYPE]
rename
item_action as action1,
item_test as test1,
do_if as do_if1,
redefine
action1, test1
select
action1, test1
end
LINEAR_ITERATOR [SOME_TYPE]
rename
item_action as action2,
item_test as test2,
do_if as do_if2,
redefine
action2, test2
end
LINEAR_ITERATOR [SOME_TYPE]
rename
item_action as action2,
item_test as test2,
do_if as do_if2,
redefine
action2, test2
end
feature
action1
-- Action for the first scheme
do
...
end
action1
-- Action for the first scheme
do
...
end
test1: BOOLEAN
-- Test for the first scheme
do
...
end
test1: BOOLEAN
-- Test for the first scheme
do
...
end
action2
-- Action for the second scheme
do
...
end
action2
-- Action for the second scheme
do
...
end
test2: BOOLEAN
-- Test for the second scheme
do
...
end
test2: BOOLEAN
-- Test for the second scheme
do
...
end
iterate1
-- Execute iteration of first kind.
do
set (...)
do_if1
end
iterate1
-- Execute iteration of first kind.
do
set (...)
do_if1
end
iterate2
-- Execute iteration of second kind.
do
set (...)
do_if2
end
iterate2
-- Execute iteration of second kind.
do
set (...)
do_if2
end
...
...
end
</code>
</code>
The repeated inheritance machinery takes care of the rest.
@@ -485,97 +479,97 @@ To obtain maximum flexibility, classes that need iteration facilities should be
The following example illustrates the technique. Consider a deferred class <eiffel>FIGURE</eiffel> describing the notion of graphical figure, with many effective descendants ( <eiffel>POLYGON</eiffel>, <eiffel>CIRCLE</eiffel> and so on). It is useful to define <eiffel>COMPLEX_FIGURE</eiffel>, describing figures that are recursively composed of sub-figures. This is a remarkable example of multiple inheritance:
<code>
class
COMPLEX_FIGURE
COMPLEX_FIGURE
inherit
FIGURE,
LINKED_LIST [FIGURE]
FIGURE,
LINKED_LIST [FIGURE]
feature
...
...
end
</code>
</code>
In the feature clause we want to provide the appropriate effectings for the deferred features of class <eiffel>FIGURE</eiffel>: <eiffel>display</eiffel>, <eiffel>hide</eiffel>, <eiffel>translate</eiffel> and all other basic figure operations. <br/>
We can use loops for that purpose, for example
<code>
display
-- Recursively display all components of the complex
-- figure
do
from
start
until
exhausted
loop
item.display
forth
end
end
</code>
display
-- Recursively display all components of the complex
-- figure
do
from
start
until
exhausted
loop
item.display
forth
end
end
</code>
Although acceptable and even elegant, this scheme will cause significant duplication: all the <eiffel>FIGURE</eiffel> features - not just <eiffel>display</eiffel> but also <eiffel>hide</eiffel>, <eiffel>rotate</eiffel>, <eiffel>move</eiffel> and others - will have the same structure, with a loop. We can use iterators to avoid this duplication. The repeated inheritance technique would work, but given the large number of <eiffel>FIGURE</eiffel> features the amount of repeated inheritance that would be needed seems unwieldy. It is also not very desirable to have to change the inheritance structure of the system just to add a new feature to <eiffel>FIGURE</eiffel>. The more dynamic approach using iterator objects seems preferable. <br/>
To implement this approach, define a class for iterating on complex figures:
<code>
class
COMPLEX_FIGURE_ITERATOR
COMPLEX_FIGURE_ITERATOR
inherit
LINEAR_ITERATOR
redefine
target
end
LINEAR_ITERATOR
redefine
target
end
create
set
set
feature
target: COMPLEX_FIGURE
target: COMPLEX_FIGURE
end
</code>
</code>
Then for each operation to be iterated define a small class. For example:
<code>
class
FIGURE_DISPLAYER
FIGURE_DISPLAYER
inherit
COMPLEX_FIGURE_ITERATOR
redefine
item_action
end
COMPLEX_FIGURE_ITERATOR
redefine
item_action
end
create
set
set
feature
item_action (f: FIGURE)
-- Action to be applied to each figure: display
-- it.
do
f.display
end
item_action (f: FIGURE)
-- Action to be applied to each figure: display
-- it.
do
f.display
end
end
</code>
</code>
Similarly, you may define <eiffel>FIGURE_HIDER</eiffel>, <eiffel>FIGURE_MOVER</eiffel> and others. Then the features of <eiffel>COMPLEX_FIGURE</eiffel> are written almost trivially, without any explicit loops; for example:
<code>
display
-- Recursively display all components of the complex
-- figure
local
disp: FIGURE_DISPLAYER
do
create disp.set (Current)
disp.do_all
end
</code>
display
-- Recursively display all components of the complex
-- figure
local
disp: FIGURE_DISPLAYER
do
create disp.set (Current)
disp.do_all
end
</code>
and similarly for all the others. <br/>
Note the use of <eiffel>set</eiffel> as creation procedure, which is more convenient than requiring the clients first to create an iterator object and then to call <eiffel>set</eiffel>. This is also safer, since with <eiffel>set</eiffel> as a creation procedure the client cannot forget to initialize the target. (If a class <eiffel>C</eiffel> has a creation clause, the creation instruction <code> create </code> <eiffel>C</eiffel>.)
Note the use of <eiffel>set</eiffel> as creation procedure, which is more convenient than requiring the clients first to create an iterator object and then to call <eiffel>set</eiffel>. This is also safer, since with <eiffel>set</eiffel> as a creation procedure the client cannot forget to initialize the target. (If a class <eiffel>C</eiffel> has a creation clause, the creation instruction <code>create </code> <eiffel>C</eiffel>.)