cl-perec blog series by pinterface

For several years, the best (and only) option for portable access to relational databases from Common Lisp has been the CLSQL library, a free reimplementation of LispWorks' CommonSQL interface with support for many backend databases, a lispy SQL abstraction as well as simple object-relational mapping support.

Recently, several alternatives have emerged: Postmodern provides particularly good support tailored to PostgreSQL only, other libraries abstract away from relational databases entirely, and finally, there are two new stars on the library horizon: CL-RDBMS and CL-PEREC.

Having used CLSQL for several years, I am currently investigating a switch to CL-RDBMS and PEREC for a code base of non-trivial size, aiming to improve scalability, portability across databases, and last not least readability of the code base.

And I am very happy with everything I have found so far:

In the CL-RDBMS/PEREC ecosystem, work is split into two layers.

  • CL-RDBMS forms the lower layer. It abstracts over database access libraries (currently with support for PostgreSQL, Oracle, and SQLite) and defines an optional, Lispy, extensible syntax for SQL for data definition and query statements.
  • The optional PEREC layer is the one which really shines. It is an object-relational mapping (ORM) seamlessly integrating CLOS classes with database tables.


  • Sensible caching model: Configurable caching of slots within a transaction, but not between transactions.
  • Object identity preserved (within a transaction).
  • Protection against accidental cross-transaction access.
  • Lispy SELECT statements using code that looks like ordinary CLOS use, but which usually compiles down to efficient database-side SQL statements.

Overall, an architecture that is ideal for robust, scalable setups with multiple threads and processes without risk of the "oversold concert tickets" situation reported for a popular Lisp web site based on older ORM software a few years ago...

The only fly in the ointment for me was the lack of introductory material. PEREC comes with an extensive test suite and nice examples, but I would have wished for a few more hints to get me started.

Thanks to fellow blogger pinterface, my wish has been granted. He has written a blog series about PEREC with many details about getting set up, and including various customization tips.

Articles in the series:

  1. Getting Started with cl-perec
  2. Persisting Simple Types with cl-perec
  3. Sensible Serializing with cl-perec
  4. Peering Down the Rabbit Hole with cl-perec

PEREC is still evolving, but the tips presented are definitely helpful for PEREC as implemented today. Also check the comments on the blog articles, written by PEREC's author!


Ben Hyde's project dependency graph

Ben Hyde writes about clbuild and has a very nice graph of project dependencies in his post.

His approach is extraordinarily simple and uses a small Lisp program to parse clbuild's dependencies file and write out a .dot file. Of course, using Graphviz is a well-known approach for this kind of dependency diagram. Bill Clementson mentioned Richard Newman's graphs a few years ago.

But Ben's graph is a fresh take on this because of the way it uses clbuild's recorded dependency information, which shows dependencies of projects rather than dependencies of systems. What's the difference between projects and systems? Lisp projects often contain multiple .asd files.

Before thinking about the details of that distinction, let's look at the picture. Here's his graph, showing Hunchentoot, Drakma, cxml, and others:

  • Yes, there's a lot of Babel in there. (It deserves it!)
  • The project dependency graph is not transitive. Note how CXML depends on Babel, which in turn needs Stefil. But CXML itself does not depend on Stefil. How can that be? It's because the system CXML depends only on the system babel. And that's not where the use of Stefil comes from. It's actually just babel-tests that needs Stefil, and that isn't needed when you're compiling CXML.
  • Where there are indirect dependencies, the graph shows them in full. Note that CL+SSL has an arrow to babel here, although it doesn't use it directly. This indirect dependency is due to its use of CFFI.

While probably not suitable for all dependency graphs, I think that this explicit display of indirect dependencies gives very nice results in this case, because it highlights commonly used libraries like Babel more than an ordinary graph would have done.

You can download his code at github. Note that you don't even have to download the projects that you want to compute a graph for, if you replace the call to directory with any other test of your choice.


XSLT 1.0 implemented in Common Lisp

Newly released: Xuriella XSLT, the first implementation of XSLT written entirely in Common Lisp.

Based on Plexippus XPath, Xuriella is an implementation of XSLT 1.0, written by Ivan Shvedunov and me.

Xuriella is quite complete and correct -- we run the official testsuite, with more than 95% of tests currently passing.


One advantage of a pure-Lisp implementation is that extension elements (as well as XPath extensions) can be defined easily.

That's a huge plus because XSLT itself is a very specialized programming language -- it excels at XML/HTML generation and transformation only. Being able to write custom extensions in Lisp helps with any non-XML-ish parts of the task which XSLT itself might not handle conveniently.


If you just want to try applying stylesheets, there are only two functions you need to know about: parse-stylesheet and apply-stylesheet.

For details about these functions (and all others, including those for extensions), refer to the API documentation.


The example uses Hunchentoot and Xuriella with XSLT as a template language in a simple directory listing request.