Slashdot Mirror
Writing Perl Modules for CPAN
Besides Perl's abilities as a rapid development language, it's widely believed that the CPAN is its most valuable feature. This network of freely distributable code allows competent developers to achieve great heights of productivity, reusing the work of a generous community of programmers.
Of course, just as some will argue that Perl's copious documentation (spread over two thousand pages) is not immediately obvious to beginners, neither is how to use and even to contribute to the CPAN. For every coder who's successfully published a module, how many more would jump at the chance? How many registered CPAN authors would like to improve their skills?
With that audience in mind, Sam Tregar's Writing Perl Modules for CPAN plants itself firmly in the gap between novice and intermediate user. While much of the book presents information present in a multitude of FAQs, manpages, and the bittersweet experiences of those of us who did things the hard way, he's collected much knowledge into a short and readable guide.
What's to Like?Tregar starts by describing the history and usage of the CPAN itself. This includes the three most popular approaches to building modules: through the CPAN shell (including its configuration), by hand, and with ActiveState's PPM tool. Next, he explains module development in forty pages. This is pretty dense stuff for the intended audience and might require several passes by newer coders. Only after re-reviewing the chapter for this summary did I realize how much he covered. The next chapter covers design and style, from naming schemes to appropriate laziness through code reuse. It's more philosophical and more important.
The next two chapters cover bundling and submitting modules to the CPAN, as well as being a good author and maintainer. The general tone is quite similar to the impressive Open Source Development with CVS. While manpages usually describe the mechanics of making a distribution, for example, they rarely explain the reasons why things are done that way. As with previous chapters, several code examples illustrate the concepts under discussion.
After a brief chapter discussing a few very effective CPAN modules, Tregar dives into XS (the interface between Perl and C). In 60 pages, he describes just enough of XS and the Perl API to teach careful programmers how to be effective at extending Perl. This introduction compares favorably to the first few chapters of the new (and excellent) Extending and Embedding Perl. As expected in an overview, he provides links to more information. The writing and example style is clear enough that a decent coder with sufficient C knowledge should be able to write a Perl wrapper to a C library with relative ease.
The last two chapters describe Inline::C, an abstraction layer that makes XS much easier, and CGI::ApplicationC, a state machine framework for Perl CGI applications. It's not quite clear why the last chapter was included (besides Tregar's desire to see more CPAN modules extending CGI::Application), but it serves as an example of using and extending a CPAN module. Perhaps a future version of the book will elaborate further.
What's to ConsiderThe book's code samples are generally good. In the first half, they are all related parts of a larger project. The rest of the book moves away from this approach. Perhaps it would have been worthwhile to continue the theme, though the nature of the material makes it difficult to see exactly how to accomplish this.
Tregar also avoids the use of strictures and warnings in his code examples, claiming that they would make the examples too verbose. I disagree with the given reasoning -- teaching is the best time to enforce good habits, especially when encouraging the students to distribute their code to the world. This is a minor issue, though, as the code is readable and reasonable.
In the past few months, two projects have gained a great deal of momentum in Perl space. These are the CPANPLUS (disclaimer: I am contributing to this project and have contributed to CPAN.pm) and Module::Build. They may become the new standards, replacing CPAN.pm and MakeMaker as early as Perl 5.10. The book omits mention of these. This is understandable, given the time frame -- and the current tools will not be disappearing any time soon. Potential replacements for h2xs are described in a sidebar, though.
The SummaryThis is a readable book. It took only a couple of hours to read (though I'm assuredly not the target audience), and is well packed with good advice. Fresher Perl programmers who aren't yet comfortable enough with packages and interfaces will get the most benefit, but there's plenty of information for intermediate hackers as well.
Table of Contents- CPAN
- Perl Module Basics
- Module Design and Implementation
- CPAN Module Distribution
- Submitting Your Module to CPAN
- Module Maintenance
- Great CPAN Modules
- Programming Perl in C
- Writing C Modules with XS
- Writing C Modules with Inline::C
- CGI Application Modules for CPAN
You can purchase Writing Perl Modules for CPAN from bn.com. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.
Perl 6 will have a Perl 5 compatability mode; which means that Perl modules (especially those which are giving high level functionality) written for 5 will load in this compatability mode and still work fine. Those giving lower level functionality will be replaced by Perl 6 modules with a speed increase since the Perl authors will have 10 years more information about what the community actually wants to use Perl for. More importantly the low level Perl interface (Parrot) will be much easier than the current rather complex Yacc interface.
Perl is a wonderful language, but like any it takes time to ease into, and it has its warts (both the aesthetic "ew, I hate dollar signs" warts that newbies hate and the deeper "roll your own OO is painful" blemishes). However, CPAN ends all argument, IMHO. If you aren't using CPAN, you're wasting your (company's?) time.
There are just so many amazing libraries in CPAN, I can't get over how most of what I want to do is already there, and at my fingertips. Want process management? WWW client emulation? Scientific computing functions? It's all in there, and managable via the CPAN.pm module.
Come Perl 6, CPAN (or its successor) will be even more valuable. The back-end virtual machine (Parrot) will have modules which give it access to external capabilities, and any Parrot-based language will be able to use them. You'll be able to call Perl's LWP from Python, or a scheme tree-handler from Perl. The details are all being worked on as we speak, but the basic building blocks are mostly in place already.
You slightly over-simplify, but more importantly, you've left some things out.
Perl 6 will not have a "compatibility mode" per se (actually it will in patterns, but that's not what you were refering to); it will have a full-fledged Perl 5 compiler that will compile down to Parrot byte-code just like Perl 6 will (as will Python, Scheme and some other languages).
By its nature, it will have to know more about Perl 6's internals than the other language front-ends, but not by much. You will be able to do something like "perl5 program " and your program will run exactly as you expect, and modules will be auto-recognized (because Perl 6 does not use the "package" keyword).
There is a catch. XS modules (those that directly interface with Perl's internals in order to talk to external C or C++ libraries) may need more help. In most cases, the interface will still be available through some emulation, but some will simply have to be fixed to work with Parrot-based Perl 5.
The most difficult part of writing a module for CPAN is getting past the bureaucracy: there are at least two different documents giving different instructions on how to submit modules, and the modules@perl.org mailing list is notorious for never replying to any messages sent to it, especially not those asking for help or 'what went wrong?'. It's not like registering a project on Sourceforge or Freshmeat; you have to send mail to register a namespace, then do an ftp upload and then send mail to have your upload 'noticed'. Sounds reasonable enough, except that it is easy to go wrong or fail to jump through a particular hoop, and if that happens you get no feedback at all. I'm sure the CPAN maintainers are busy and hard-working, but the current situation shows there are cases where an automated system (which at least gives error messages when stuff goes wrong) is better than manual administration.
Myself, I have managed to put modules on CPAN, after a few months of wondering why they didn't appear in the main listing. (It turned out I had forgotten to send an updated module list entry; my fault, but still it would have been friendlier for somebody to answer my question about what I did wrong rather than just ignore it.) But uploading newer versions of the same modules has sometimes been troublesome, with the update appearing in my home directory but not the main listing: again, any request for help or guidance on what part of the process I was doing wrong was studiously ignored.
Despite the troublesome upload procedure, CPAN is Perl's biggest asset and other language communities would do well to copy it (as some are). But please don't copy the management system; have some more structured way to submit code where it's clearer what to do at each stage, and there can be warning messages if you do something wrong, rather than just quietly failing to work.
-- Ed Avis ed@membled.com
perl Makefile.PL
make
make install
sequence, Inline automatically calls the compiler and linker for your C/etc code, and creates the right glue code between your Perl and C/etc code. For a simple example, see the C mailbox parser which comes with <shameless plug>grepmail</shameless plug>.
By the way, recent improvements to the Perl implementation mean that my Perl mailbox parser is now less than 5% slower than the C implementation. Just one data point for those of you who say Perl can't be fast. ;)