Agent wrapper generation - Revision history

Manus: Added category

2007-05-30T18:44:52Z

Added category

Colin-adams: Note to find a better example

2007-04-28T06:12:20Z

Note to find a better example

Peter gummer: /* Alternatives syntaxes for dropping arguments */

2007-04-28T01:50:13Z

‎Alternatives syntaxes for dropping arguments

Peter gummer: /* Alternatives syntaxes for dropping arguments */

2007-04-28T01:49:54Z

‎Alternatives syntaxes for dropping arguments

Colin-adams: Added additional patterns

2007-04-27T14:46:25Z

Added additional patterns

Colin-adams at 08:02, 27 April 2007

2007-04-27T08:02:55Z

Colin-adams at 14:10, 26 April 2007

2007-04-26T14:10:28Z

Colin-adams at 13:39, 26 April 2007

2007-04-26T13:39:49Z

New page

{{UnderConstruction}}

Sometimes the situation arises when you have a routine that you want to use as an agent, but the number of parameters required
by the routine is fewer than those that will be passed by the agent call. An example from Eiffel Software's Vision2 library:

<eiffel>
class
EV_POINTER_BUTTON_ACTION_SEQUENCE

inherit
EV_ACTION_SEQUENCE [TUPLE [INTEGER, INTEGER, INTEGER, DOUBLE, DOUBLE, DOUBLE, INTEGER, INTEGER]]
-- EV_ACTION_SEQUENCE [TUPLE [x, y, button: INTEGER; x_tilt, y_tilt, pressure: DOUBLE; screen_x, screen_y: INTEGER]]
</eiffel>

If have have a routine that wants to know when a button is pressed, and which button it was, then it needs the <eiffel>`button'</eiffel> argument,
but none of the others.
Best practice for this situation at the moment (according to Eric Bezault) is to write a wrapper routine that just passes on the <eiffel>`button'</eiffel> argument.

A similar situation is when you have the correct number of arguments, but they are the wrong way round. This happened to me yesterday, so I had to write a wrapper routine.

It would be much nicer if the compiler were able to generate the necessary wrapper code for you. This artcile explores possible syntaxes for these problems.

== Argument rotation ==

A simple solution for specifying a rotation of 2 arguments (or any permutation that does not drop arguments), is to extend the use of the question mark by suffixing it with a decimal specifying the number of the argument to be passed in that location. For example, to index a list of floating point numbers:

<eiffel>
floats_list: DS_ARRAYED_LIST [REAL_64]
-- Floating point numbers

floats_map: DS_HASH_TABLE [INTEGER, REAL_64]
-- Map of floating-point indices within `floats_array'

build_map
-- Build `floats_map' from `floats_list'.
require
floats_list_not_void: floats_list /= Void
do
create floats_map.make (floats_list.count)
floats_list.do_all_with_index (agent {DS_HASH_TABLE} floats_map.put_new (?2, ?1))
end
</eiffel>

← Older revision		Revision as of 18:44, 30 May 2007
Line 1:		Line 1:
		+	[[Category:ECMA]]
	Sometimes the situation arises when you have a routine that you want to use as an agent, but the number of parameters required		Sometimes the situation arises when you have a routine that you want to use as an agent, but the number of parameters required
	by the routine is fewer than those that will be passed by the agent call. An example from Eiffel Software's Vision2 library:		by the routine is fewer than those that will be passed by the agent call. An example from Eiffel Software's Vision2 library:

← Older revision		Revision as of 06:12, 28 April 2007
Line 100:		Line 100:
	end		end
	</eiffel>		</eiffel>
		+	(this needs a better example, as the above can already be done, since it is the first open argument. Find an example where the second open argument is used)

@@ Line 52: / Line 52: @@
 Franck Arnaud does not like _ to indicate dropping remaining arguments. He prefers -? (or maybe it was ?-). I don't like this. If a question mark must be used, then I would prefer ?0, but I much prefer the underscore.
-A thought that occured to me is should it be an error to supply the underscore, if there are no arguments to be passed? Probably it should.
+A thought that occurred to me is should it be an error to supply the underscore, if there are no arguments to be passed? Probably it should.
 An alternative syntax for dropping an argument might be ?-n, meaning drop the nth argument. If this was used, then all arguments must be accounted for at least once. For example, if you only wanted the button number from the action sequence at the top of the page, then the syntax would be <eiffel>agent notify_button_press (?3, ?-1, ?-2, ?-4, ?-5, ?-6, ?-7, ?-8)</eiffel>. This has the advantage that if the number of expected arguments were to change in the future, then you get a compile error, so you can consider what to do about it. This might be important if the button argument were to disappear, for instance!

@@ Line 48: / Line 48: @@
 The compiler now has all the information necessary to check the type of the agent, and to generate a wrapper routine (hopefully, a smart optimizing compiler can eliminate the wrapper altogether).
-=== Alternatives syntaxes for dropping arguments ===
+=== Alternative syntaxes for dropping arguments ===
 Franck Arnaud does not like _ to indicate dropping remaining arguments. He prefers -? (or maybe it was ?-). I don't like this. If a question mark must be used, then I would prefer ?0, but I much prefer the underscore.
@@ Line 57: / Line 57: @@
 But I don't approve of changing APIs like that, without changing the name. So I still prefer to see <eiffel>agent notify_button_press (?3)</eiffel> (or <eiffel>agent notify_button_press (_)</eiffel> if all you are interested in knowing about is that the user clicked any mouse button).
 == Killing three birds with one stone ==

← Older revision		Revision as of 14:46, 27 April 2007
Line 57:		Line 57:

	But I don't approve of changing APIs like that, without changing the name. So I still prefer to see <eiffel>agent notify_button_press (?3)</eiffel> (or <eiffel>agent notify_button_press (_)</eiffel> if all you are interested in knowing about is that the user clicked any mouse button).		But I don't approve of changing APIs like that, without changing the name. So I still prefer to see <eiffel>agent notify_button_press (?3)</eiffel> (or <eiffel>agent notify_button_press (_)</eiffel> if all you are interested in knowing about is that the user clicked any mouse button).
		+
		+	== Killing three birds with one stone ==
		+
		+	There are other patterns which currently need wrappers, but the proposed syntax would help with:
		+
		+	=== Repeating an open argument ===
		+	Another requirement that has occasionally come up is to repeatedly use an open argument. For example:
		+	<eiffel>
		+	event_dates: ARRAY [DATE]
		+	-- Dates on which event occurred
		+	.
		+	.
		+	.
		+	do
		+	event_dates.do_all (agent process_interval (?1, ?1))
		+	end
		+	.
		+	.
		+	.
		+	process_interval (a_start, a_end: DATE) is
		+	-- Process event lasting from `a_start' to `a_end' inclusive.
		+	do
		+	.
		+	.
		+	.
		+	end
		+	</eiffel>
		+
		+	=== Using an open argument as a target ===
		+
		+	Sometimes you have a collection of objects and you want to invoke a routine of the object itself:
		+	<eiffel>
		+	fleet: ARRAY [AEROPLANE]
		+	-- All planes in fleet
		+
		+	do
		+	fleet.do_if (agent {AEROPLANE}(?1).refuel, agent {AEROPLANE}(?1).is_low_fuel)
		+	end
		+	</eiffel>

← Older revision		Revision as of 08:02, 27 April 2007
Line 47:		Line 47:

	The compiler now has all the information necessary to check the type of the agent, and to generate a wrapper routine (hopefully, a smart optimizing compiler can eliminate the wrapper altogether).		The compiler now has all the information necessary to check the type of the agent, and to generate a wrapper routine (hopefully, a smart optimizing compiler can eliminate the wrapper altogether).
		+
		+	=== Alternatives syntaxes for dropping arguments ===
		+
		+	Franck Arnaud does not like _ to indicate dropping remaining arguments. He prefers -? (or maybe it was ?-). I don't like this. If a question mark must be used, then I would prefer ?0, but I much prefer the underscore.
		+
		+	A thought that occured to me is should it be an error to supply the underscore, if there are no arguments to be passed? Probably it should.
		+
		+	An alternative syntax for dropping an argument might be ?-n, meaning drop the nth argument. If this was used, then all arguments must be accounted for at least once. For example, if you only wanted the button number from the action sequence at the top of the page, then the syntax would be <eiffel>agent notify_button_press (?3, ?-1, ?-2, ?-4, ?-5, ?-6, ?-7, ?-8)</eiffel>. This has the advantage that if the number of expected arguments were to change in the future, then you get a compile error, so you can consider what to do about it. This might be important if the button argument were to disappear, for instance!
		+
		+	But I don't approve of changing APIs like that, without changing the name. So I still prefer to see <eiffel>agent notify_button_press (?3)</eiffel> (or <eiffel>agent notify_button_press (_)</eiffel> if all you are interested in knowing about is that the user clicked any mouse button).

@@ Line 1: / Line 1: @@
 Sometimes the situation arises when you have a routine that you want to use as an agent, but the number of parameters required
 by the routine is fewer than those that will be passed by the agent call. An example from Eiffel Software's Vision2 library:
@@ Line 41: / Line 39: @@
   end
 </eiffel>