Comments on: Persistent Trees in git, Clojure and CouchDB

By: Daniel Waterworth

Daniel Waterworth — Thu, 24 Feb 2011 07:09:53 +0000

Hi,

I wondered if you’d be interested in aodbm. It’s an append only database in the style of dbm. Here’s it’s url, https://sourceforge.net/projects/aodbm/ .

By: D'gou

D'gou — Mon, 29 Mar 2010 02:26:57 +0000

It is interesting to note that Mercurial, another DVCS, uses the same goodness of “append only” that CouchDB uses. I don’t know the timeline for CouchDB, but my impression is that Mercurial was doing “append only” before then. But “append only” itself is a very old technique.

By: Frando

Frando — Tue, 09 Feb 2010 14:43:50 +0000

These are all directed acylclic graphs (http://en.wikipedia.org/wiki/Directed_acyclic_graph).

By: Ashwin Jayaprakash

Ashwin Jayaprakash — Wed, 20 Jan 2010 21:32:35 +0000

I believe Apache Cassandra also implements some form of LSM storage.

By: Yuval Kogman

Yuval Kogman — Thu, 24 Dec 2009 14:58:21 +0000

http://www.cs.cmu.edu/~rwh/theses/okasaki.pdf became a book called Purely Functional Data Structures: http://books.google.com/books?id=SxPzSTcTalAC&dq=purely+functional+data+structures&printsec=frontcover&source=bn&hl=en&ei=XoEzS-DLBOqJ4Ab1yejZAQ&sa=X&oi=book_result&ct=result&resnum=4&ved=0CBoQ6AEwAw

Secondly, http://www.amazon.com/Algorithms-Functional-Programming-Approach-International/dp/0201596040

By: Jorge Vargas

Jorge Vargas — Mon, 21 Dec 2009 15:49:34 +0000

Really interesting post. Thank you. I have been puzzling with that patter for some time. Now I have a description and a name!

By: Nathan Youngman

Nathan Youngman — Tue, 15 Dec 2009 03:11:58 +0000

I’ve been following Clojure for a few months and find it’s data structures very intriguing. Thanks for this article… I do hope it’s an idea whose time has come.

By: Robert Virding

Robert Virding — Mon, 14 Dec 2009 15:01:00 +0000

Immutable data and trees are old, much older @FuncFan than Chris Okasaki, although he has written a very good description of them. They are a fundamental part of functional and logic programming.

By: Stephen Colebourne

Stephen Colebourne — Mon, 14 Dec 2009 12:52:52 +0000

I’d also suggest that RESTful architectures have a link to this pattern. The RESTful approach conceptually doesn’t have an update – PUT is equivalent to a replace, which maps on to the append-only approach very nicely.

By: Manuel Woelker

Manuel Woelker — Mon, 14 Dec 2009 08:17:38 +0000

@FuncFan: Thanks for putting a name to the face! I added a small update to the post.

@Dan Fabulich: Actually the git tree objects themselves are immutable. What rebase does is create a new series of trees, that are changed version of the originals; but the original trees remain untouched. Only the HEAD reference now points to the new history.

Rebasing being a bit of work is a Good Thing IMHO. For one, it discourages excessive use of the feature. But it also makes it impossible to secretly “spirit away” certain parts of the history. Every downstream repository is going to know that history has been mucked with, and can verify independently what changes were made with relative ease.

Note that rebasing does have its legitimate uses. One for example are local branches that might need a bit of pruning or polishing. Another use case are situations where certain files have to be removed for legal reasons (e.g. license incompatibilities). C.f. https://bugs.eclipse.org/bugs/show_bug.cgi?id=270852

But again, the reason that the rebase is so much work is exactly the fact that the trees themselves (and thus their SHA-1 identifiers) are immutable.