JDK-4513622: keeping a substring of a field prevents GC for object

http://bugs.java.com/view_bug.do?bug_id=4513622

by luu • 11 years ago

14 10 11 years ago

bugs.java.com

eranation• 11 years ago

A friend of mine was actually asked this in an interview (how does substring works). The initial idea of taking an advantage of the immutability of an object was nice, but it turns out to be a big GC issue if your original string is extremely large.

dfkf• 11 years ago

It is one of those cases where you can actually leak memory with a GC. But at least Java gives (gave?) you a choice: to create a new string a use the current one, in .NET substring always returns a new string.

jeffdavis• 11 years ago

This is why I find rust intriguing. It tracks the lifetimes using the type system, so it would be obvious to the programmer that the large strings would need to live as long as the list holding the small substrings.

I'm not a rust expert, but I believe that it's basically impossible to write something like the code in the bug report without getting a compile time error in rust. You'd either need to use a version of substring that makes a copy, or explicitly keep the large strings around.

pcwalton• 11 years ago

Yeah, in Rust you'd have to go out of your way to use Rc to reference count the large strings.

ephemeral007• 11 years ago

fixed in 6 (and backported in 5, 4) see http://bugs.java.com/view_bug.do?bug_id=6924259

ddispaltro• 11 years ago

This is really old, they changed the way it worked to be a copy from jdk5 to jdk6 for exactly this reason.

pimlottc• 11 years ago

This needs a 2001 tag, particular since it is no longer the case as others have pointed out.

masklinn• 11 years ago

It is probably the case in other languages though (I believe — though I'm open to being wrong — it is in Go, string slicing is implemented through array slicing, so a substring will share the parent string's storage and keep it alive).

hiringcat2• 11 years ago

Even better - installing java on the mac install the ask toolbar that actually adds garbage into my search results, let alone garbage collection. Howzabout then?

strictfp• 11 years ago

The view technique was actually great. You could write performant parsing and matching code without having to keep track of positions in the string. The current situation is much worse. You get crazy amounts of allocation and GC for code that used to produce almost none.

JDK-4513622: keeping a substring of a field prevents GC for object

Related Stories