I'd like to have "lenient hash check for noarch subpackages" discussed. Basically, we do an rpmdiff between the noarch subpackages that are produced on different arches. If the checks fail, the build is failed. However, the checks currently don't check every difference between subpackages. Notably, the hash check is not performed.
I've modified the rpmdiff we're using to check hashes on everything except %doc files, .pyo and .pyc. I think that enabling this check and adding to it if we find other file types that should be whitelisted is preferable to not doing hash checks at all.
This is the results of running with --lenient-hash on packages that have current been rebuilt with noarch subpackages::
https://www.redhat.com/archives/fedora-devel-list/2009-February/msg02185.html
This sounds good to me - and fixing dbus :) How difficult is this to get upstream?
The rpmdiff that koji uses hasn't been upstreamed but I've talked to Ville about getting it and my additional changes in there. He's looking at the source code now.
Bear in mind that while I'm for this, mbonnet and dgilmore are both against doing the hash check.
I'm also persuaded that adding '/usr/share' to the whitelist along with %doc and .pyc and .pyo is a good idea.
This was voted against at the 2009/03/13 FESCo meeting, on the basis that if we do this for noarch subpackages, we should do the same for noarch base packages as well.
X crashed as FESCo was voting... I want to make sure I understand this. Was mbonnet and dgilmore on board with this idea? Or were they just opposed to doing a hash check? Is FESCo for doing a hash check of all noarch packages or only against doing a hash check if it only applies to subpackages?
To do checks on noarch packages in addition to noarch subpackages we need to modify koji. So I need to know before I start working on code whether it's something that will be accepted because it's seen as a wanted feature or if it's a waste of time because hash checks themselves are not wanted.
Login to comment on this ticket.