On 16.05.2022 14:33, Rick McGuire wrote:
There are also other places where this check is made. Search for Error_Execution_super to find it. The entire validateOverrideContext() method and it's calls should be deleted.
Thank you for your hints and pointers, will take care of it. ---rony
On Mon, May 16, 2022 at 8:24 AM Rick McGuire <[email protected]> wrote: You have only fixed part of the problem. There's also a change required to MessageInstruction.cpp and also tests needed for that case. Rick On Mon, May 16, 2022 at 8:15 AM Rick McGuire <[email protected]> wrote: Weren't there any tests for the restriction that needed to be removed? I only need new tests added for the case where this is not restricted. Also, I'd recommend adding some tests using mixins to make sure the correct targets are getting invoked. Rick On Mon, May 16, 2022 at 8:10 AM Rony G. Flatscher <[email protected]> wrote: On 15.05.2022 14:47, Rony G. Flatscher wrote:On 15.05.2022 12:27, Rony G. Flatscher wrote:On 14.05.2022 22:06, Jean Louis Faucher wrote:* So there is a need for having one or more methods that can be used for forcing the invocation of the ooRexx .Object methods.The syntax described in 4.2.7 Changing the Search Order for Methods could be used, if the restriction "Message search overrides can be used only from methods of the target object” was removed. It works with oorexx4, after removing the check if (_target != context->getReceiver()) in RexxExpressionMessage::evaluate. s1 = "hello" s2 = "hello" say s1~"="(s2) -- display 1 say s1~"=":.Object(s2) -- display 0 because not the same objectsIndeed that would really be a general, fine solution alleviating a programmer to come up with weird and cumbersome solutions. Rather than having to create methods SEND.SUPER, SENDWITH.SUPER, CLASS.SUPER and COPY.SUPER to allow programmers to invoke the ooRexx root class methods in .object, this problem with OLEObject, but also all comparable in general would be solved with this. So instead one could code * ole~send(msg) ... will check for existence on the Windows side, and if present invoke it, otherwise lookup super (which is .object) * ole~send:.object(msg) ... will start resolving the method in the superclass bypassing inspecting .oleobject and with the same technique: * ole~sendWith:.object(msg,arrArg) * ole~copy:.object * ole~class:.object This would be much easier and very clear. In ooRexx 5.0 this would be the place to change: Index: interpreter/expression/ExpressionMessage.cpp =================================================================== --- interpreter/expression/ExpressionMessage.cpp (revision 12388) +++ interpreter/expression/ExpressionMessage.cpp (working copy) @@ -161,6 +161,7 @@ // do we have a super class override? if (super != OREF_NULL) { +/* // super class overrides are only allowed if the // sender and the target are the same object (i.e., a message to SELF) if (_target != context->getReceiver()) @@ -167,6 +168,7 @@ { reportException(Error_Execution_super); } +*/ _super = (RexxClass *)super->evaluate(context, stack); // we send the message using the stack, which Doing so will make your example work on ooRexx 5 as well! Also experimented with other scenarious, including ones where "mistakingly" wrong override classes get supplied. arr=.array~of("a", "b") ........................................... rexxtry.rex on WindowsNT say arr~items 2 ........................................... rexxtry.rex on WindowsNT say arr~items:super Oooops ! ... try again. Object method not found. Object "an Array" does not understand message "ITEMS". rc = 97.1 ................................. rexxtry.rex on WindowsNT say arr~items:.collection 2 ........................................... rexxtry.rex on WindowsNT say arr~items:.rexxinfo Oooops ! ... try again. Object method not found. Object "an Array" does not understand message "ITEMS". rc = 97.1 ................................. rexxtry.rex on WindowsNT say arr~copy a b ........................................... rexxtry.rex on WindowsNT say arr~copy:.rexxinfo Oooops ! ... try again. Object method not found. Object "an Array" does not understand message "COPY". rc = 97.1 ................................. rexxtry.rex on WindowsNT say arr~copy:.object a b ........................................... rexxtry.rex on WindowsNT So ooRexx 5 already catches wrong overrides and raises the appropriate conditions (cf. overrides "super", ".rexxinfo" above)! --- Conceptually this change will allow the programmer to not only send a message to the object, but also to tell the object in which superclass to start the search for a matching method if he has a need to do so. In the case of .OLEObject it makes it simple for programmers to tell the OLE object to start its search for a method in the root class .object applying existing knowledge! So no need to come up with awkwardly named methods or another dispatch.super method to somehow get access to the root class methods making the usage/protocol of such classes rather complicated. So such a change would simply allow to apply the message resolution override pattern that the programmer is accustomed to already. --- The question would be whether there are any potentially dangerous side-effects or incompatibilies with existing code that could get introduced by removing this particular check. ---ronyOpened a RFE for this: <https://sourceforge.net/p/oorexx/feature-requests/802/> <https://sourceforge.net/p/oorexx/feature-requests/802/> ---ronyImplemented <http://sourceforge.net/p/oorexx/code-0/12390> <http://sourceforge.net/p/oorexx/code-0/12390>. Added appropriate tests. ---rony
_______________________________________________ Oorexx-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/oorexx-devel
