Re: isa
Re: isa
- Subject: Re: isa
- From: Aram Greenman <email@hidden>
- Date: Sun, 2 Jun 2002 20:48:12 -0700
On Sunday, June 2, 2002, at 01:16 AM, Marcel Weiher wrote:
I also suggested changing the object's isa temporarily, with the
disclaimer that it might be a bad idea.
It is a very bad idea, especially the temporary swizzling.
Exactly how? That is my suspicion, but I would like to know
specifically why it is a bad idea.
Unintended and unforseen side-effects.
There are no unforeseen side-effects. There are only defined and
undefined behaviors.
The point is that the problems are difficult to forsee precisely,
because the assumption that the isa pointer is correct is a very basic
assumption that is made throughout the system.
However, if A' inherits from A, saying an A' is an instance of A _is_
correct, insofar as it has all the instance variables of A and therefore
possesses everything that makes an A an A.
The other, maybe bigger, problem is that if you even *want* to do this,
you are ignoring a message that your code is trying to tell you very
loudly: I am organized incorrectly!
Suppose you asked a Java programmer, "I want to declare a variable that
can hold any kind of object. How can I do that?", they would probably
say "That is completely unnecessary. Your code is organized incorrectly."
If you were strictly limited to a Java way of thinking, that would be
correct. With all due respect, it seems that your response here is
limited to an 'normal' Obj-C way of thinking.
When I first started learning Obj-C, it didn't seem that dynamic typing
was really that useful. Of course I quickly saw that I was wrong. I
admit that this isa-manipulation doesn't _seem_ that useful and of
course it isn't strictly necessary. But if it was possible to for an
object to invoke multiple behaviors for the 'same' message, it might
find uses that aren't readily apparent. My intent here was not to try to
get around good programming practice or the correct Obj-C way of doing
things, but to see if something was possible, and _then_ see how it
could be used.
Maybe this is a language-level feature and can't be (easily) implemented
in Obj-C. As I said at the beginning of this thread, that is my
suspicion, but I disagree with your statement that:
The point is that the problems are difficult to forsee precisely
I don't see how this can be true. A quick grep of the source to the
runtime shows isa to appear in only 64 lines. I haven't looked at a lot
of it yet, but since it is known exactly how the isa pointer is used by
the runtime, any use that would be undesirably affected by the
manipulation should be identifiable (for example, maybe the runtime also
writes to the isa pointer after it is initialized, or something).
Aram
_______________________________________________
cocoa-dev mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/cocoa-dev
Do not post admin requests to the list. They will be ignored.
- Follow-Ups:
- Re: isa
- From: Ondra Cada <email@hidden>
- Re: isa
- From: Jeff LaMarche <email@hidden>
- Re: isa
- From: Phillip Morelock <email@hidden>
| References: | |
| | >Re: isa (From: Marcel Weiher <email@hidden>) |