----- Original Message ----- From: "Michael A. Smith" <[EMAIL PROTECTED]> To: "Jakarta Commons Developers List" <[EMAIL PROTECTED]>; "Morgan Delagrange" <[EMAIL PROTECTED]> Sent: Thursday, February 14, 2002 3:49 PM Subject: Re: [Collections] subclassing LRUMap
> On Thu, 14 Feb 2002, Morgan Delagrange wrote: > > I just made a backward-compatible change to LRUMap to facilitate > > subclassing. Now, you can override removeLRU() if you want to manipulate > > (e.g. persist to disk) an Object before it is automatically removed from the > > Map. You can also subclass remove(Object) if you want to manipulate all > > Objects before they are removed, whether or not they are being removed by > > the LRU algorithm. > > > > This seemed like a pretty reasonable thing to do, but I wanted to drop a > > note to the list since it enhances the contract of LRUMap. Let me know if > > you object to the idea or the approach. > > Actually, I just object to the implementation of LRUMap itself. Many of > its operations have O(n) cost when they could be O(1). I haven't taken a > good look at the changes you've made, but it doesn't look like you've > fixed that. :) No, the changes I've made recently were mainly to fix bugs and allow more control in subclasses. If anything, I've made the class slightly less performant for put(Object,Object) operations. > I'll try to take a closer look at what you've changed > when I have a few spare cycles (possibly not until this weekend). Yeah, there are longer-term issues with LRUMap, particularly wrt. its effectiveness of its algorithm as a cache. I'll launch this in a separate thread soon. I'm going to need to write more unit tests, because any changes I make should try to be compatible with previous externalized versions, or they can't be made in a 1.x version of Collections (Collections has already performed an official 1.0 release). (We really should have had a unit test for externalized versions before I started making changes, but there were some outright bugs that needed attention.) > I don't know if you saw these other messages talking about LRUMap > (there's more, I just selected a couple): > > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg02555.html > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg03084.html > Yup I've seen them. James quoted the relevant portions in his Javadocs. > Oh, and since you seem to have an active interest in collections (no one > else seems to), maybe you could review the patches I've submitted that > have not yet been applied? > > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg03055.html > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg03056.html > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg03057.html > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg03060.html I may take a look, but right now I'm just focusing on LRUMap. I plan to implement a cache with it. > I was planning on "fixing" LRUMap once the SequencedHashMap patch was > applied. I also plan on adding a ton of test cases as well (for all Map > implementations). I mentioned that here: > > http://www.mail-archive.com/commons-dev@jakarta.apache.org/msg03054.html > > I've got quite a few done, but its not nearly as complete as I'd like. I've added several unit tests for LRUMap, mainly to illustrate bugs before I committed the fix. - Morgan _________________________________________________________ Do You Yahoo!? Get your free @yahoo.com address at http://mail.yahoo.com -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
