On Mon, 27 May 2013 20:41:08 +0100 Sérgio Basto <sergio(a)serjux.com&gt; wrote: Done. Note that you should next time request this by filing a ticket in releng trac. ;) kevin

On May 27, 2013 11:17 AM, "Sérgio Basto" <sergio(a)serjux.com&gt; wrote: I've been using "git checkout $targetbranch; git checkout $sourcebranch -- relative/path/to/file" to pull $file from $sourcebranch to $targetbranch. Globbing with * works, too. I have no idea if this is the best approach, but it works for my purposes. --Pete

On Mon, 2013-05-27 at 23:04 +0300, Oron Peled wrote: The ideal way to do it is to make your changes in 'master' and merge them down to any branches which have not yet diverged from 'master' at all: i.e. as long as your Rawhide and F19 builds are the same, you can just make changes in 'master' and merge them to 'f19'. As soon as your f19 build diverges from master at all, merging becomes inappropriate (and probably impossible) and you should instead use 'cherry-pick'. It helps to have at least a vague overview of how git is designed to be used, and what is the actual _point_ of the commands you're using in the intended git workflow. Just because the *content* of two branches is identical does not mean changes can be merged from one to the other with no issues. Merging will only work without issues if the change history of each branch is the same. In practice, you want to look at 'git log' for the two branches and check the commit IDs. If the last few commit IDs differ - even if the content of the commits is the same - you are not going to be able to merge without issues. At any point, if you're not sure what the hell is happening and you're getting error messages, *don't* just whack it with a hammer until it seems to be right, push that, and hope nobody notices; *do* yell for help from someone who really understands git (note: not me) in #fedora-devel or on here. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | identi.ca: adamwfedora http://www.happyassassin.net

On 05/28/2013 11:30 AM, Paolo Bonzini wrote: That's because there's little support in VCS for backporting. The VCS doesn't know if the new development in master is part of the fix, or unrelated new development. The only system I've ever used which had some support for this was darcs, but more often than not, I had unintentional patch dependencies on new development I didn't want to backport, so it didn't work out that well. The forward-porting approach has the risk that it stops before reaching master, so users will encounter regressions when updating. And both approaches do not really mix that well because merging from the stable branch with cherry-picked and massages backports tends to conflict a lot. Better tool support for backporting and the more general issue of patch stacks (like we have in RPM and Debian packages) is very desirable, but it's a really difficult problem. -- Florian Weimer / Red Hat Product Security Team