It's been a bad week for update sites and Galileo contribution from Modeling... and I confess I'm partly to blame. That and the fact that despite documenting processes, workarounds, tips, tricks, and advice... no one Reads The Fine Mediawiki (Category:Releng or Modeling Project Releng).
- The mysterious appearance of a new version of org.eclipse.osgi_*.jar in releng.basebuilder's R35_M5 tag, which caused an ant <copy/> used to rename a file to fail because copy can't merge two jars into one file. Still no idea why an old basebuilder tag would magically grow a new jars, but I've worked around the now-faulty assumption w/ smarter Ant code.
- A change to the way our sites are created, in an attempt to workaround what I believe (but can't yet prove) is a flaw in the way content.xml is produced - namely, if the xml file is > 21M, it gets truncated or corrupted. We used to cache 2 or 3 releases of a given project (eg., M5 and M6) on the same site, in order to give people a way to "back up" to the previous release; now, you only get the latest (bug 271486). I confess I screwed up here and instead of replacing a folder with new contents, I was copying INTO that folder - `cp -r one two` resulted in folder one/two/ instead of two/. I fixed that by changing to a move instead of a copy, but a downstream process failed because of the assumption that both one/ and two/ would exist. The lessons here are: a) shotgun debugging sucks, and b) don't change the way stuff is created after M7.
- People publishing two updates to a site at the same time, resulting in the appearance of two </site> tags in a site.xml file, causing p2 metadata generation to be incomplete or fail entirely; unfortunately, no error is logged when this happens so it's rather difficult to decipher the tea leaves. This may be the real source of the metadata corruption, if not the "file is too big" issue above.
- Observations about obsolete jars corrupting metadata, but no one taking it upon themselves to clean up the site or do some troubleshooting
- People inconsistently naming their milestones (it's 2.0.0M7, not just M7!) and corrupting our Release Notes database. This one amazes me the most since it takes seconds to see what was done last time (check any of the following: RSS feeds, release notes, downloads pages, update sites) and follow suit. And, of course, the conventions are documented, along with the rationale (consistent patterns == simpler code).
Or, to put it another way...
Sick of this life
Not that you'd care
I'm not the only one with
whom these feelings I share
Quite why we're here
We're searchin' for answers
That never appear
But maybe if I looked real hard I'd
I'd see your tryin' too
To understand this life,
That we're all goin' through
Sometimes I feel like I'm beatin' a dead horse
And I don't know why you'd be bringin' me down
I'd like to think that your love's
Worth a tad more
It may sound funny but you'd think by now
I'd be smilin'
I guess some things never change