[External] : Backporting a set of patches in a single commit

[Langer, Christoph]

Hi Kevin,

thanks for the information and the link to Erik's mail. I'll request the feature for jdk11u-dev then via a separate mail to ops at o.j.n<mailto:ops at o.j.n>.

Best regards
Christoph

From: Kevin Rushforth <kevin.rushforth at oracle.com>
Sent: Freitag, 23. Juli 2021 00:45
To: Langer, Christoph <christoph.langer at sap.com>; skara-dev at openjdk.java.net; erik.joelsson at oracle.com
Cc: Lindenmaier, Goetz <goetz.lindenmaier at sap.com>
Subject: Re: [External] : Backporting a set of patches in a single commit

The only drawback I'm aware of is the bit of extra noise caused by the automatically created "pr/NNNN" branches. If you don't have a dependent PR (which most aren't in my experience), then you can ignore it. If you do, then the branch is there and ready for you to use.

I will look for a write-up, but I don't see one, so I don't know for sure whether Erik H finished docuementing it. The best source of documentation I know of is the "New Skara feature: dependent pull requests" [1] email that Erik H sent in March announcing the feature.

If you want us to enable it, we can enable this some time next week.

-- Kevin

[1] https://mail.openjdk.java.net/pipermail/skara-dev/2021-March/004528.html

On 7/22/2021 2:37 PM, Langer, Christoph wrote:
Hi Kevin,

well, I think then we should try the dependent pull request feature on jdk11u-dev. Is there any drawback besides the additional branches created? Also, is there any documentation?

Thanks
Christoph

From: Kevin Rushforth <kevin.rushforth at oracle.com><mailto:kevin.rushforth at oracle.com>
Sent: Mittwoch, 21. Juli 2021 23:37
To: Langer, Christoph <christoph.langer at sap.com><mailto:christoph.langer at sap.com>; skara-dev at openjdk.java.net<mailto:skara-dev at openjdk.java.net>; erik.joelsson at oracle.com<mailto:erik.joelsson at oracle.com>
Cc: Lindenmaier, Goetz <goetz.lindenmaier at sap.com><mailto:goetz.lindenmaier at sap.com>
Subject: Re: [External] : Backporting a set of patches in a single commit

Hi Christoph,

Yes, I think one solution would be to enable the dependent pull request feature of Skara for your repo. It's already being used by the jdk project. Since it is already implemented and tested (at least for original fixes, but I can't think of anything that would preclude it for backports), it would be easy to set up so you could try it and see if it met your needs.

FWIW, I ran into this in a couple cases myself when backporting fixes to jfx11u after the Skara switch. I ended up testing them in a branch where I pulled in the relevant fixes I needed, and then did them as separate PRs in the right order. Since they were clean it took only a few extra minutes. It was the testing that took all the time anyway, but I couldn't help thinking that dependent pull requests would have made it easier.

-- Kevin


On 7/21/2021 2:20 PM, Langer, Christoph wrote:
Hi Skara Team,

after gathering the first significant experiences with the backport process using GitHub/Skara, I find that there is one use case which works considerably worse than before. It is when you want to backport a set of patches that must be pushed together. Imagine you have a patch, then a build fix and then maybe another test fix. So as to not break builds and tests, you will want to push them together in one go. You could now add those 3 commits into one PR. But when you integrate, the 3 commits get squashed which is contrary to the usual way of backporting which tries to preserve individual commits. Alternatively, you could file 3 separate PRs. This, however, often bears the problem that the PRs might be dependent on each other. So any individual backport PR could possibly only be opened and reviewed after its predecessor was integrated. Which will leave the repo in a bad state for a while.

So what can we do here? Shall we change the paradigm and allow more squashing? Or did you already think of a better workflow? I think there is a feature of Skara called dependent PRs. Could that be used here? Or could maybe the backport bot recognize individual commits to preserve?

Thanks in advance for sharing your thoughts.

Best regards
Christoph