Imported From: https://issues.scala-lang.org/browse/SI-4835?orig=1
Reporter: @kmizu
Affected Versions: 2.9.0
Other Milestones: 2.10.0-M1
Attachments:

• StreamIterator.patch (created on Jul 25, 2011 4:04:25 AM UTC, 1109 bytes)
• StreamIterator.patch.2 (created on Jul 26, 2011 3:50:23 PM UTC, 915 bytes)
• StreamIteratorProblem.scala (created on Jul 25, 2011 4:04:25 AM UTC, 241 bytes)

@dcsobral said:
This patch breaks the lazyness of stream for the initial element. I think that isn't a big problem, though effects of that can be subtle. Assigning null to self shouldn't be necessary -- if you don't use it in a method, it shouldn't become part of the object.

I'm surprised, however, that "val stream" inside the definition of "var these" doesn't suffer from the same problem -- even though it is in a small scope, it ought to be saved so that it is preserved when the by-name LazyCell gets invoked. It might even be a bug here.

@kmizu said:
Hello. Thanks for commenting to my patch. However, I couldn't understand that my patch breaks the lazyness of stream for the initial element. If you don't mind, I would like to see concrete example of it.

I think that isn't a big problem, though effects of that can be subtle

I think that it isn't big problem, too. However, I think that Iterator#next is not to expected consume O(n) memory generally.

Assigning null to self shouldn't be necessary – if you don't use it in a method, it shouldn't become part of the object.

Because primary constructor's arguments self continues to exist until ths object is garbage collected, assigning null to self is needed now if StreamIterator's API shouldn't be changed.

Thanks again.

Kota Mizushima

@dcsobral said:
Indeed, the first element isn't lazy at all, as it is passed by value. So that shouldn't be a problem.

As for self, it * shouldn't* continue to exist. It only did so because it was used in a by-name parameter.

@kmizu said:

Indeed, the first element isn't lazy at all, as it is passed by value. So that shouldn't be a problem.

I see that because argument self: Stream itself is passed by value, then the first element wasn't lazy
without my patch. Is is right?

As for self, it * shouldn't* continue to exist. It only did so because it was used in a by-name parameter.

OK. I created another version of patch and will attach it as StreamIterator.patch.2. By applying this patch,
no longer, unneeded field self is erased.

huynhjl said:
Note that this used to run in constant memory in 2.8.1 and most likely in 2.8.0 as well. Something happened in the 2.9.0 release.

@kmizu said:
Several days ago, I sent pull request fixing this isssue on git://github.com/scala/scala.git and the pull request was merged into master branch:

commit 39a837c8357e78c949613db881bd060d176f3279
Author: Kota Mizushima mizukota@gmail.com
Date: Mon Jan 23 18:53:05 2012 +0900

Add test case for SI-4835 (https://issues.scala-lang.org/browse/SI-4835)

This test case only confirm that StreamIterator's lazyiness is not broken.
Test case about memory consumption could be created.  However, such a
test cause a greatly increased time of test.

I'm not a commiter. However, because the fix was merged into master branch, I've thought that it should be noted.

Jed Wesley-Smith (jedws) said:
this is tagged as a fix in the 2.9.2 release notes but is unresolved here. What is the source of truth?

@kmizu said:
As far as I see https://github.com/scala/scala/blob/v2.9.2/src/library/scala/collection/immutable/Stream.scala, my fix is included in 2.9.2. However, because I'm just another contributor, I don't know whether or not I have permission to close an Issue myself.