On 05/04/2010 02:14 PM, Kevin Kofler wrote:
Jesse Keating wrote:
> This involved doing another build of the package, which could
> involve changes in the buildroot and anomalies in the build
> process. Ask DaveJ some time about what happened to his kernel
> builds when the build host did a clock adjustment during the build.
> Shit happens, and making assumptions that just because the build
> completed that nothing went wrong is a great way to make a fool of
> yourself.
Some risks are so low that they're basically negligible.
Sure, but if the chance of any given thing going wrong - i.e. doing a
build that successfully produces a package that doesn't do what you
meant for it to - is .005, and there are 20 of those things which are
serious enough to be a problem, that's a 10% chance of a major problem.
(to say nothing of the myriad of things that can cause minor problems.)
If the 2 options are keeping an existing regression (which missed
testing) in updates for a few more days or risking the off chance
that there MAY be another regression with a probability of 1 in a
million or something in that order of magnitude, I'll take the risk
any day! If that kind of risk is too high for you, I hope you don't
ever use a car, it might crash, you know?
It's nice that you'll take the risk, but it sure would be nice to
shield the *users* of our software from this irrational bravado.
--
Peter
RFC 882 put the dots in .com.