Monthly Archives: November 2008

I am currently nailing down dependencies for JBoss. It is a lot slower than I originally had hoped. I am being impeded by my work flow as well as the size of the list of dependencies. My goal for my 0.2 release was to have all the dependencies fleshed out for JBoss but it is a lot slower than I thought it would be. The first obstacle is my workflow. Currently I find that the easiest way to work with the dependency table is to take the HTML from the wiki then load it into OpenOffice.org Calc then do my changes. After this I have to take the HTML output of OpenOffice.org then convert it back into wiki. I find that this is a very error prone procedure, so I try to do it as little as possible. Another problem with my workflow is that i am working on my laptop as my desktop is out of service right now. I am used to a lot larger of a screen and as a result I have to use virtual desktops. This slows me down as I forget where everything is and causes me to forget things when switching.

The actual task of finding dependencies is a slow task intrinsically. I find that there is no set package naming convention for jar files. One package will be vendor-project-package-version.jar and the next will be package-version.jar. Because of this it is very difficult to write a parser. To get the status of each dependency I am search in packagekit (or yum) for each package. It can be difficult sometimes to go from a jar file to figuring out where the upstream is. It is also confusing because there will be multiple versions of the same package (i.e. Apache Commons Collections). Fedora does include a specific version. I can see this will be an issue because it rarely matches with the version that Fedora has and there are multiple versions I will have to figure out which version JBoss is coded to itself and which versions it uses for the web applications it hosts. Once I have this information, I am trying to see which packages are repeated and naming them in a way which reflects this.

I am also still working on my ario package from my 0.1. I have gotten the package imported into the Fedora CVS and I am currently in the process of mock building it on F-8 through F-10 and once I finish this, I will push it live with bodhi. My second package which is required for me to become a full fledged packager, links, is currently at a bit of a road block because of conflicts. There are three versions of the “Links” browser, links1, links2 and elinks. Fedora only has elinks and as a result the package provides and obsoletes an unversioned links package. Because of Fedora’s naming conventions I wanted to name the package I was doing ‘links’ as it is the original links author continued efforts. Another contributor was saying that because a lot of scripts rely on elinks for processing on X-less servers my package should be links2 or glinks. Currently my package builds with X.org as a dependency for graphical mode. I am thinking of creating a sub package called ‘links-graphical’ which would allow for someone to have links with or without graphical support. I am wondering how this would work with the Fedora policies. The reason I am skeptical is that links compiles down to a single dynamically linking binary, so to have X capable and X incapable versions I’d have to ship two different binaries, one per package.

0.2 stuff:

• Updated Dependency Chart
• Ario Bug on Bugzilla
• Ario on Fedora CVS
• Links Bug on Bugzilla