Filtering iterators

[wysota]

Yes, but filtering may depend from the data and the data may change.

In that case your iterator is pretty much useless anyway.

What if you filter your XML tags using the value of an attribute (say "class=foo") and you change that value later on?

Then you are violating the basic rule of not cutting the branch you're currently sitting on. Regular iterators are not mutable by default. Special considerations need to be taken to make an iterator mutable.

Anyway, you're right about filtering containers: I didn't think of that, but it's kinda obvious now that you noticed it. But I think that in case of "infinite iterators", for example when data is generated from external devices, network or pipelined algorithm, containers would be pointless. Not that's my case, just pointing it out

Iterators would be as pointless. The point of iterator is that you can iterate over all elements. If the data is infinite, it is hard to iterate through it. Then you have a generator and not an interator.

[akiross]

I'll be quick

1. Why?

2. I wasn't talking about mutable iterators. Probably there is a misunderstanding here, because caching works the way I say: when you cache something, you must ensure it's validity when you're using it later. If data has changed (and it may change also when you finished to iterate it), you must refresh. That's my point.

3. Generators generates. Iterators iterates. I think they are quite different things. You generate data with generators and process it; you CAN use iterators to process such data. If you want to call "iterators" the things that are associated with a finite container, fine, but you've to find another word for such things that work on infinite data sequences, because generator is wrong in this case. By the way, STL calls these things iterators, or stream iterators. For example, istream_iterator is not bounded by any number, and it's a perfectly legal iterator.

You may be interested in this: http://stackoverflow.com/questions/2...tor-bad-design

~Aki

[wysota]

I'll be quick

Likewise.

1. Why?

Deleting the pointed item invalidates the iterator. So does addition in certain cases.

2. I wasn't talking about mutable iterators. Probably there is a misunderstanding here, because caching works the way I say: when you cache something, you must ensure it's validity when you're using it later. If data has changed (and it may change also when you finished to iterate it), you must refresh. That's my point.

My point is that the iterator/container/whatever can only work on const objects. And you can't modify those.

but you've to find another word for such things that work on infinite data sequences

Yeah, I call them generators If something generates something then it is a generator. If you want to iterate over a generator, that's fine but the iterator iterates, not generates, if you ask me. I know it's just a matter of taxonomy but still.

istream_iterator is not bounded by any number, and it's a perfectly legal iterator.

If it has end() defined, then it is finite. Otherwise I hate to be in a skin of a person trying to process every element of such collection.

[akiross]

Deleting the pointed item invalidates the iterator. So does addition in certain cases.

My point is that the iterator/container/whatever can only work on const objects. And you can't modify those.

Modification doesn't regard only creation and deletion. There is also modification which doesn't invalidate iterators. Anyway, *this was not the point* and you went on another topic. The point was: if you use iterators for filtering, and data is highly mutable (NOT while iterating), then caching is useless and iterators are Ok to use.

Yeah, I call them generators If something generates something then it is a generator. If you want to iterate over a generator, that's fine but the iterator iterates, not generates, if you ask me. I know it's just a matter of taxonomy but still.

Probably you didn't read well what I wrote OR I expressed badly. I said:

I think that in case of "infinite iterators", for example when data is generated from external devices, network or pipelined algorithm, containers would be pointless.

I didn't say that iterators were used to generate data (and I thought that was obvious). A stream is not an iterator, but is an infinite source of data. I was talking about using iterators OVER an infinite collection of data, and as you say: "if you want to iterate over a generator, that's fine". And that was the point of my statement in first place.

If it has end() defined, then it is finite. Otherwise I hate to be in a skin of a person trying to process every element of such collection.

I don't agree. You're assuming that we are talking about STL-style iterators, which uses begin() and end(). Java-style iterators, for example, checks using "hasNext()" and not testing if they reached a defined end. That method could return true to give an infinite iterator.
In addition, I think that the semantic of the "end" method in STL iterators is NOT do define a length of a container, but to define a case in which has no longer sense to go on (which is very very different from saying that the iterator will find an end).
It's easy to build a circular list with >0 elements which will iterate forever, even if the end() is defined.
~Aki

[wysota]

Modification doesn't regard only creation and deletion.

No, but if there exist a situation when modifying a container is unsafe then modifying a container is in general unsafe.

The point was: if you use iterators for filtering, and data is highly mutable (NOT while iterating), then caching is useless and iterators are Ok to use.

I don't see your point. If you modify an element in a way that it changes the result of filtering then from the "filtering container's" point of view you "add" or "remove" an element from the container. I don't see how you intend to cope with that in a general case. At some point you will wake up in "nowhereland" and you won't even notice it (e.g. you modify the current element in such a way that changes the filtering result of the element hence your iterator should immediately become invalid, however you have no way of rechecking the filtering condition for the current element and also calling next() and prev() will not get you to the original element as during "prev()" the condition will be rechecked and the element filtered out). Of course in many cases it will work out just fine but I'm not really fond of software that "usually works".

[akiross]

I don't see your point. If you modify an element in a way that it changes the result of filtering then from the "filtering container's" point of view you "add" or "remove" an element from the container.

Sure. Didn't say no The point was: if you have to change 90% of the container data every time you have to access it, then, probably, caching is not worth the effort. The point was: containers with caching are surely useful, but there exist at least one case when caching is not and iterators alone will do. Got it?

~Aki

[wysota]

If it doesn't cost you anything to cache then I don't see why not cache. Let's end this discussion, it probably doesn't lead to any useful conclusions