Rethinking some core OI2 packaging and distribution assumptions

lachoy on 2005-02-07T12:45:09

I've been working a lot recently on making OpenInteract2 more standard -- 'standard' meaning you can install without having extra directories around and that the applications are just CPAN packages. This is a long-winded discussion of the issues and the process that's led me to a very different view of the subject.

One of OpenInteract's strengths has always been that you can package and distribution applications very easily -- a command from the command-line tool 'oi2_manage' will create a new package for you, another will turn it into a zip file, and other will install that zip file to a website.

This is nifty, but the problem is that we have to create our own distribution infrastructure because CPAN isn't built to handle such files. And since there has been no effort to ever do so it's a pretty safe bet there never will be.

For a long time I didn't believe these issues were a big deal. (Actually, I thought about them very little. Too busy implementing. Lame.) The motivations for creating distributable packages in their current form came from a different set of needs. But in thinking more about these and about how usability affects adoption I've changed my opinion. This wasn't done overnight. At first I just wanted a way for people to install OI2 without a 'source directory'. A little background -- with the current CPAN version and and earlier you need to do something like this to install OI2 from CPAN, create a website and create a package for development:

 # standard Perl module stuff...
 tar -zxvf OpenInteract-1.99_05.tar.gz
 cd OpenInteract-1.99_05
 perl Makefile.PL
 make
 make test
 make install
 # now we get into the customizations; 'source_dir' is the default here:
 oi2_manage create_website --website_dir=/path/to/site
 
 # if you're creating a website later, you need:
 oi2_manage create_website --website_dir=/path/to/site --source_dir=/path/to/oi2-source
 
 # if you're developing a new package:
 oi2_manage create_package --package=myapp --source_dir=/path/to/oi2-source

The 'source_dir' held two directories, 'sample' and 'pkg'. The 'pkg' directory held a set of zip files for the core OI2 applications, and the 'sample' directory all the templated configuration and other files for creating a new package and website.

Later I added a management task to create this source directory in a public place ('/usr/local/oi2' or something), but that was just a band-aid. I didn't think much about this just because there were so many resources -- 110 files or so -- and, to be honest, I wasn't really thinking from a user's point of view./p>

So recently I had the idea to get rid of this source directory (see OIN-121 for more). My first thought was to use Inline::Files and treat all of those 110 resources as a file within a class. But I had an odd problem with that (something about a scalar not being promotable) and in the course of investigating decided it would be easier to just make every resource a subroutine that returned the contents of a heredoc.

After some (surprisingly small) reworking of the tools that manipulated those 110 resource files that worked reasonably well. Then I had to do the packages -- a set of zip files distributed with the OI2 CPAN package. I figured I could use the same process but just base-64 the zip contents and inline them the same way. And it worked! We'd now rid ourselves of the hated source directory!

As I was doing that I remembered another issue (OIN-72) about making OI2 packages distributable on CPAN. I figured it would be just as easy to use this same process to distribute a package. And it wasn't that difficult, but it opened a whole slew of opportunities I didn't foresee that forced me to rethink what a 'package' actually is.

First, some more background. A package is a full application, consisting of:

  • Perl modules for actions, supporting classes, and declarations for installing your SQL structures
  • Templates used to generate your content
  • Web resources (html pages, images, etc.)
  • Configuration for content-generating actions and object-relational mapping
  • SQL data definitions for tables, sequences, whatever else you need
  • Any initial data you want stored when your application is installed, including security settings
  • Documentation

OI2 has tools for installing a package to a website, creating its SQL structures to a supported database (there are quite a few) and initializing them with initial data.

CPAN doesn't care about any of these pieces except the first and last items. It also cares about some metadata like authorship, what other modules yours depends on, the license you're using, etc. What that means is that we don't have to put them in some CPAN-compatible format. All we have to do is bundle those resources into some way that we can find them whenever we're installing the package.

So I created a management task for bundling an existing application into a CPAN distribution, generating the necessary supporting files like Makefile.PL, and creating an uploadable tarball. Works great.

But that got me thinking: why do we even need zip packages any more? Instead of thinking about CPAN distributions as an alternate way for packaging applications, why not think of them as the way? Not only do they make a number of things easier, but they also get rid of a longstanding problem we had when at startup we copied all the package modules to a single directory called the 'temporary lib dir' since we add it to @INC -- some people stored their directories on NFS mounts and recreating this directory and copying all the files there took quite a few seconds.

Since the modules would be installed to the normal @INC: no problem. And since we're using a standard tool with support for copying modules to a specific location, we can just give it a 'LIB' to ensure that one website's packages don't get in the way of another's. (This was one of the motivations behind packages, but from what I've heard in practice few people use more than one OI application per web server.)

Posted from cwinters.com; read original