In my previous job, we used to use some Java framework, Mule ESB, for our clients. The project is distributed with a lot of jar files, supposedly for the convenience of users. When we work on a project, we just add all those jars to the classpath (we use Eclipse, an excessively powerful and ugly IDE). Now, me being a sucker for the Debian way of doing things, I'd install packages providing all those jars from a Debian repository, then remove those Mule-distributed files, then add symlinks in place of them pointing back to the actualy files. Imagine how much time it took! Either that, or even easier (but still time-consuming), skip the symlink thing entirely and just add those jars (found in /usr/share/java directory) directly.
Beyond just wasting time, the are other potential problems with my approach:
- Debian's jar versions did tend to be different to Mule-supplied versions. You can imagine what problems this can cause, especially because the versions included there are most likely the ones used for testing Mule.
- Even if the versions were the same, what if there are slight changes from Debian's side. Debian's strict software guidelines implies that they'll strip out some stuff that doesn't adhere. The good thing is that Debian appends dfsg to the version number, but not everybody knows that. And even if they did, they would then need to spend time checking exactly what changed.
- Debian doesn't merely re-distribute the jar files. They actually rebuild them, with Debian-supplied compilers. Maybe this isn't an issue for Java projects, but who knows.
This is not to knock the Debian way of doing things. It's actually quite excellent (hence my love), but it can't fit all scenarios. See Matt Zimmerman's excellent post, where he touches on this issue.
I did wise up a bit by simply doing things the way the makers of Mule intended.