[SM-Discuss] stable-rc quality

seth at swoolley.homeip.net seth at swoolley.homeip.net
Wed May 24 18:23:01 EDT 2006


Prometheus has been chugging along and is about two days into what looks 
like an eight day run since we are a quarter of the way along.

While uncast is still rather high, we consider spells winners if they've 
cast successfully once, even as dependencies.  Uncast is only 
decremented for each run even if it contains successful spells built as 
dependencies .  This makes sure we catch the basic "cast straight from 
install" dependency issues.  I've built xorg, cvs, and g++ in this run 
because we have a known bug in stable sorcery with a cvs spell download 
plus cvs dependency race and many spells tend to not describe their X 
and g++ dependencies, and I don't care about that issue for this run.

77 installed.lst 1735 winner.lst 71 skipped.lst 174 loser.lst 2057 total 3302 uncast

I'm happy about the low number of "skipped" spells.  Those are spells 
that are losers or are installed.  The lower this number is the more it 
is that the losers are not critical applications that are dependended on 
by a lot of things.  At this rate, we'll see less than three hundred 
bugs filed mostly in corner cases.  That's a > 90% good ratio so far.  
The last run, I had a pretty steady 75% good ratio, but that includes 
some setup issues that messed up most of the qt-based spells for 
skipping and the x86_64 issues.  Some of the failures will be that uname 
-m is x86_64 even though this is a 32-bit chroot.  In fact, that 
happened with the atlas spell as I typed this...

About 1/4 of the failures are due to the php-pear spell creation script 
that had an issue with BUILD (thanks sandalle for that) for the api2 
conversion.  These are not true regressions, just broken enhancements, 
so they will not be holding up the stable push.

I've only been able to look at a very small number of failures in detail 
though, so when I get more time, probably later this week, I'll go 
through and mark spells that need to be fixed in stable-rc and develop a 
new test list to make sure the fixes are correct.  Things, however are 
looking good, and I figure we might make the very low end of the two 
weeks to two months prediction.



More information about the SM-Discuss mailing list