On Mon, Jul 06, 2009 at 02:13:46PM +0100, Mark McLoughlin wrote: We can't really drop the existing ChangeLog file since it contains much much more data than the historical commit messages. We need to really auto-generate only new changelog entries, Preferably only generating 'ChangeLog' when doing a 'make dist'. Could probably just do a rename of existing file to ChangeLog.old, and then just generate a new ChangeLog file from the switch over date. Daniel -- |: Red Hat, Engineering, London -o- http://people.redhat.com/berrange/ :| |: http://libvirt.org -o- http://virt-manager.org -o- http://ovirt.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: GnuPG: 7D3B9505 -o- F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 :|

On Mon, Jul 06, 2009 at 03:41:34PM +0200, Jim Meyering wrote: yeah, I think moving ChangeLog to ChangeLog.CVS and generating the logs starting from today at "make dist" time is probably the best, it will avoid copying over redundant informations. On the other hand we should try to keep the git log output coherent with the previous format. Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/

On Mon, 2009-07-06 at 16:19 +0200, Daniel Veillard wrote: Cool. FWIW, I think it's best to switch to a new format that better suits git. e.g. "git log --pretty=oneline" can be very useful, but only where people properly summarise the commit on the first line. If we used the typical ChangeLog format, it would be useless. Also, we don't need the date and author information twice. II think git commit messages generally more genuinely useful information than a ChangeLog entry - it seems to encourage people to more fully explain what they're doing. If we switch formats, 'git log --stat' is fine for Changelog - you get author, date, files changed etc. Cheers, Mark.

On Mon, 2009-07-06 at 17:22 +0200, Jim Meyering wrote: Looking at the git commits e.g. http://git.savannah.gnu.org/gitweb/?p=coreutils.git;a=commit;h=24c727d3 it doesn't duplicate date/author and has a good first line summary, so it's pretty good. The one question I'd have is whether listing of per-file changes has any value. IMHO, it tends to restrict the explanation people give about about their commits. Looking at projects that don't do this, I think you tend to get much more background on why the change is being made, not just what changed. Cheers, Mark.

Daniel Veillard wrote: To create that new repository, I rewrote the git repository from et.redhat.com to correct a few minor problems. In case anyone is interested, here are the details. --------------------------------------- Some commits had incomplete names. Run this to see all 64 commits: $ git log --pretty=format:%h:%an|grep -v ' ' Here's a summary: $ git log --pretty=format:%an|grep -v ' '|sort|uniq -c 1 aliguori 13 clalance 3 crobinso 8 danms 5 dlesko 23 meyering 11 sakaia ============================ Note that rewriting changes nearly all SHA1 values, so any reference to an SHA1 in a commit log should be updated to point to the new SHA1. There are a few: $ git log|grep -E '\b[0-9a-f]{7,40}\b'|grep -v '^commit' This regressed via commit 4c3f3b4d. [043d702f] "use virAsprintf instead of asprintf" introducted The 'getVer' fix introducted in d88d459d [Allow remote://hostname/ Bug introduced in 895d0fdf5bef358fafb91c672609190b3088097b. ==================================================== Run the following long(!) command to rewrite the "master" branch, preserving tags (but not signed ones), fixing bogus author names and preserving SHA1 references like the above: git filter-branch -d $TMPDIR/.git-rewrite \ --tag-name-filter cat \ --env-filter ' case $GIT_AUTHOR_NAME in agx) n="Guido Günther" e=&quot;agx(a)sigxcpu.org&quot; ;; aliguori) n="Anthony Liguori" e=&quot;aliguori(a)us.ibm.com&quot; ;; clalance) n="Chris Lalancette" e=&quot;clalance(a)redhat.com&quot; ;; crobinso) n="Cole Robinson" e=&quot;crobinso(a)redhat.com&quot; ;; danms) n="Dan Smith" e=&quot;danms(a)us.ibm.com&quot; ;; dlesko) n="David L. Leskovec" e=&quot;dlesko(a)linux.vnet.ibm.com&quot; ;; meyering) n="Jim Meyering" e=&quot;meyering(a)redhat.com&quot; ;; sakaia) n="Atsushi SAKAI" e=&quot;sakaia(a)jp.fujitsu.com&quot; ;; *) n=$GIT_AUTHOR_NAME e=$GIT_AUTHOR_EMAIL ;; esac export GIT_AUTHOR_EMAIL=$e GIT_AUTHOR_NAME=$n export GIT_COMMITTER_EMAIL=$e GIT_COMMITTER_NAME=$n ' --msg-filter ' cat > t.msg ref=$(perl -ne "/\b([0-9a-f]{7,40})\b/ and print \$1" t.msg) test -n "$ref" && sha=$(git rev-parse $ref) && { len=$(printf $ref|wc -c) new_sha=$(map $sha) short_sha=$(printf %.${len}s $new_sha) perl -pi -e "s/$ref/$short_sha/" t.msg } cat t.msg ' master I tested it on this tiny repository: git init -q; : > j; git add j; git ci -q --author='meyering <meyering>' -m. -a c=$(git log --pretty=%h) echo a > j; git ci -q -m "fix bug in $c" -a ==================================================== The above preserves regular tags, but not the signed ones. Reapply the signed vM.N.O tags: for i in $(git tag -l 'LIBVIRT_*'); do v=$(echo $i|sed 's/LIBVIRT_/v/;s/_/./g') echo $v git tag -f -s -m$v $v $i done ==================================================== With the above changes, we've bloated the repository by ~50%, increasing its size from 22M to 34M. Remove the cruft: git for-each-ref --format="%(refname)" refs/original/ \ | xargs -n 1 git update-ref -d git reflog expire --expire=now --all git gc --prune=now And that brings it back down to 23M. Run this function git-repo-compress() { local d=$1 du -sh "$d"; start=$(date); /usr/bin/time \ git --git-dir=$d repack -afd --window=250 --depth=250 echo started $start; date; du -sh "$d" } to compress a few remaining bits. Final size: 21M. Push to libvirt.org: git push libvirt.org:/git/libvirt.git master:master git push --tags libvirt.org:/git/libvirt.git ==================================================== After I'd pushed all of the above, Dan Berrange noticed that the new signed tags were out of order and had all been created today. Back-date the signed tags so that each has the same date as its unsigned counterpart: for i in $(git tag -l 'LIBVIRT_*'); do v=$(echo $i|sed 's/LIBVIRT_/v/;s/_/./g') date=$(git log --pretty=%ci -1 $i) echo $v GIT_COMMITTER_DATE="$date" git tag -f -s -m$v $v $i done # And push (now we need the -f option): git push -f --tags libvirt.org:/git/libvirt.git

Daniel P. Berrange wrote: ... On a similar note, the libvirt.git repository on et.redhat.com is no longer being updated. I expect to move it aside (rename to libvirt-old.git, and eventually to remove it) and to make libvirt.git a mirror of the official repository.

On Mon, Jul 06, 2009 at 02:42:59PM +0200, Daniel Veillard wrote: Now that we're using git, we can make a distinction between Author and Committer, automatically. When people create patches with 'git format-patch' and send them to this list for inclusion, the patch will include them as the Author. When Daniel, et. al. then applies the patch (with 'git am' or similar method), they will be the Committer. This is nice because 'git log', 'git shortlog', etc... will give the appropriate credit to the author, and the committer doesn't need to keep writing in the commit msg who the patch came from. Just an FYI to those new to git. -- Garry Dolley ARP Networks, Inc. | http://www.arpnetworks.com | (818) 206-0181 Data center, VPS, and IP Transit solutions Member Los Angeles County REACT, Unit 336 | WQGK336 Blog http://scie.nti.st

On Tue, Jul 07, 2009 at 10:08:12PM +0900, Atsushi SAKAI wrote: Hum, I need to switch libvirt-java , for other projects that depends on their maintainers :-) Daniel -- Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/ daniel(a)veillard.com | Rpmfind RPM search engine http://rpmfind.net/ http://veillard.com/ | virtualization library http://libvirt.org/