On 12/22/2017 10:22 AM, George Joseph wrote:
On Thu, Dec 21, 2017 at 1:44 PM, Corey Farrell <g...@cfware.com <mailto:g...@cfware.com>> wrote:George asked that I post some scenarios where this would be useful. 1. You are about to create updated asterisk package and want to quickly scan the changes to 13 since latest 13.x.0 release to see if anything is a 'must patch' for your deployments. You can use 'tig' to review changes for critical fixes until you reach the tag '13.x.0-rc1' (which you can see in the list because the tag was merged).I didn't even know about tig. There's always something to learn about the git ecosystem. :)I know the end result can be accomplished by other means, but not as easily. 2. You've created an automated test to try finding a performance regression. The test runs asterisk under a profiler and stores results. Each revision you test needs a new file to store results - 'git describe' would provide an excellent filename that is unique per revision. One thing I'm not sure about is if we should only merge rc1 tags or if we should merge all new release tags. At first all release tags seem reasonable, but the order of tags other than rc1 would be off. rc1 is special because it would be merged back to mainline before anything else. Even 13.x.0 does not get cut until after other commits are merged to 13, so if we merged 13.19.0 to 13, the commits made since 13.19.0-rc1 would appear out of order (before 13.19.0). The difference between rc1 and final release is always small, but the number of new commits to mainline between that time can be quite large. So we would merge rc1's back to mainline but how about the point releases? 13.19.0-rc1 13.19.1 ?? 13.19.2 ?? 13.18-cert1-rc1 13.18-cert2 ?? 13.18-cert3 ??
Just to be clear certified branches would be untouched by this proposal. Look at 'git log certified/13.13-cert9' - you will see the previous tagged releases in the certified/13.13 release series.
I think we should not do anything different with the point releases (including 13.19.0) because of the commit order. Think about when 13.18.4 were released. If we merged it to 13 it would be in the wrong place on 'git log 13'. Easily 200 non-merge commits would appear after the 13.18.4 tag in the 13 branch, when in fact they are not part of the 13.18.4 tag. My hope is that we can provide additional information, but only if the information is accurate.
One last detail I don't know exactly how we deal with new major releases (ie 16.0.0). If I remember correctly we will release 16.0.0-beta1, but I don't remember if that is the start of the 16.0 branch or if 16.0 starts with rc1. However it works my current proposal would be to merge the first commit of 16.0 back to 16.
Tzafrir I haven't heard from you since I joined your new thread. If we were to merge the rc1's back to mainline so that mainline knew about the "split point", would this satisfy your request? Also was your intent to say "we should do this first before deleting minor branches"? Do you object if we proceed with removal of minor branches or does your request need to be completed first?
On 12/21/2017 10:45 AM, Corey Farrell wrote:I just read `git help merge` again and I think the solution is 'git checkout 13 && git merge --strategy ours 13.19.0-rc1'. This would effectively tell git that '13' already contains 13.19.0-rc1, but without actually trying to pull any changes to 13. This merge would be the final step of mkrelease.py. No changes will be needed to our handling of '.lastclean', please ignore those comments as I was wrong. On 12/21/2017 08:19 AM, George Joseph wrote:On Wed, Dec 20, 2017 at 3:14 PM, Corey Farrell <g...@cfware.com <mailto:g...@cfware.com>> wrote: One thing that might improve this is if releases were merged back to the major branch. Currently the commit "Update for 13.19.0-rc1" is on the 13.19 branch and tagged as 13.19.0-rc1. I believe that if we added 'git checkout 13 && git merge 13.19.0-rc1' we would get better information from 'git describe 13' and tags would appear in 'git log 13 --oneline'. This would continue working even after we delete the minor branches. Sounds reasonable. As a test I just ran 'git merge 13.18.4' from the current 13 branch. The merge did have 2 conflicts but that's because 13.18 was branched so long ago and a couple files that were modified in minor releases have since been modified again or deleted. Then I ran 'git describe 13', it said '13.18.4-404-gd5d67bb1f4'. This tells us that my local branch had about 404 commits (including merges) that are not part of 13.18.0-rc1 (the point where 13.18 diverged from 13 because 13.18.3 was not merged back). Merging each tag as soon as it's created would make the results more accurate. and (almost always) eliminate merge conflicts. "almost always" will be an issue since it's the scripts that do the work. It's kinda frustrating already when you're trying to get releases out the door and something goes wrong with the script. What conditions do you think might still cause merge conflicts? The only wrinkle in this plan is that the '.lastclean' file is created on the releases but it's listed in .gitignore. I think we might be able to just get rid of the .lastclean and .cleancount files. This Makefile hack predates the use of SVN and I don't think it's necessary. One thing it does do is try to enable the astdb2sqlite3 utility, but Berkely DB was last used in Asterisk 1.8. The default is for that utility to be enabled, that's enough. In addition the mkrelease script actually copies .cleancount to .lastclean, I think that means it's disabled for releases. These kind of things we can alter to suite our needs so there shouldn't be an issue. On 12/20/2017 12:58 PM, George Joseph wrote:On Wed, Dec 20, 2017 at 8:14 AM, Tzafrir Cohen <tzafrir.co...@xorcom.com <mailto:tzafrir.co...@xorcom.com>> wrote: On Wed, Dec 20, 2017 at 07:50:03AM -0700, George Joseph wrote: > On Wed, Dec 20, 2017 at 2:44 AM, Tzafrir Cohen <tzafrir.co...@xorcom.com <mailto:tzafrir.co...@xorcom.com>> > wrote: > > > Off-topic: > > > > On Tue, Dec 19, 2017 at 01:50:03PM -0700, George Joseph wrote: > > > > > Thankfully we tag EVERYTHING! :) > > > > asterisk(13)$ git describe > > 13.15.0-rc1-908-ge31e3b581b > > > > asterisk(14)$ git describe > > fatal: No tags can describe 'fb18797ae09a685ec71101499fb1c1c606b16397'. > > Try --always, or create some tags. > > > > asterisk(15)$ git describe > > fatal: No tags can describe 'd312068ee93ff8ce97b464f3c2ff3304e15cb3fe'. > > Try --always, or create some tags. > > > > > > I wasted half an hour yesterday trying to find out why a build sis not > > switch from master to 13, only to realize that the name of the git > > branch in the version string is always "master". > > > > We tag everything. But only well after it was branched from the main > > > branch. > > > > I'm not following you. > > We tag every release... > > $ git checkout 13.18.4 > HEAD is now at f4644317b7... Update for 13.18.4 > $ git describe > 13.18.4 > $ git checkout 13.18 > Switched to branch '13.18' > Your branch is up-to-date with 'gerrit/13.18'. > $ git describe > 13.18.4 > $ > > We have to create the minor release branch (13.18) and do the work there so > that patch releases (13.18.4) are based on the minor release branch, not > the major branch. Those branches are likewise short-lived branches (at least with respect to the number of commits). Real work is done on master, 13, 15 and such. But when I'm on such a branch, I can't ask git on which branch I am (not to mention: at which stage in it). I _think_ I understand now. For instance: maybe whenever you tag a new release branch (e.g. 13.18), tag the split point as something like "13.18.base" or "base.13.18"? Well, that's easy enough. Toss us an issue for it. But maybe it's just me and branches 13 and 15 are not widely used (for master it is irrelevant anyway). -- Tzafrir Cohen +972-50-7952406 mailto:tzafrir.co...@xorcom.com <mailto:tzafrir.co...@xorcom.com> http://www.xorcom.com -- _____________________________________________________________________ -- Bandwidth and Colocation Provided by http://www.api-digital.com -- asterisk-dev mailing list To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-dev <http://lists.digium.com/mailman/listinfo/asterisk-dev>-- George JosephDigium, Inc. | Software Developer 445 Jan Davis Drive NW - Huntsville, AL 35806 - US Check us out at: www.digium.com <http://www.digium.com/> & www.asterisk.org <http://www.asterisk.org/>-- _____________________________________________________________________ -- Bandwidth and Colocation Provided by http://www.api-digital.com -- asterisk-dev mailing list To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-dev <http://lists.digium.com/mailman/listinfo/asterisk-dev>-- George JosephDigium, Inc. | Software Developer 445 Jan Davis Drive NW - Huntsville, AL 35806 - US Check us out at: www.digium.com <http://www.digium.com/> & www.asterisk.org <http://www.asterisk.org/>-- _____________________________________________________________________ -- Bandwidth and Colocation Provided by http://www.api-digital.com -- asterisk-dev mailing list To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-dev <http://lists.digium.com/mailman/listinfo/asterisk-dev> -- George Joseph Digium, Inc. | Software Developer 445 Jan Davis Drive NW - Huntsville, AL 35806 - USCheck us out at: www.digium.com <http://www.digium.com/> & www.asterisk.org <http://www.asterisk.org/>
-- _____________________________________________________________________ -- Bandwidth and Colocation Provided by http://www.api-digital.com -- asterisk-dev mailing list To UNSUBSCRIBE or update options visit: http://lists.digium.com/mailman/listinfo/asterisk-dev
