I agree PRs are easy to work with for developers. Here's the same question I was trying to ask put a different way:
If a user discovers that feature X is broken, how easy it for him to find out whether someone before him has found this bug and then figure out what version of the software he needs to fix it. Bug tracking systems are optimized for answering these queries, and I don't think source control systems are. Maybe calcite is enough developer focussed that anyone searching for issues will be comfortable skimming through the git log to figure it out, so maybe my concerns are moot in this case. Alan. > On Feb 8, 2017, at 12:30 PM, Julian Hyde <[email protected]> wrote: > > In principle it is hard to compute the delta of a pull request, but in > practice it is easy. A well-formed pull request is a branch that is a small > number of commits away from the master branch at the time, and the pull > requests that we tend to accept are well-formed. > > Since we don’t rewrite the master branch, you can easily apply the pull > request using “git rebase”. Because git knows where where the pull request’s > branch meets the master branch, it can do a better job than “patch” could. > > Julian > > >> On Feb 8, 2017, at 12:21 PM, Alan Gates <[email protected]> wrote: >> >> I agree that PRs are easier to manage than attaching patches to JIRA. And >> now days most contributors seem to prefer them as well. >> >> One question I have is about traceability and findability. It is very nice >> for people to be able to come to JIRA and figure out if others have had the >> same problem they have, and if so if and where it's fixed, and exactly which >> commits they need to pick up if they want the fix. Can all this be achieved >> with just PRs? >> >> If the answer is that PRs can't achieve that, I'd still vote for moving to >> them. But I would also suggest continuing to open JIRAs that point to the >> PRs. >> >> Alan. >> >>> On Feb 8, 2017, at 11:33 AM, Julian Hyde <[email protected]> wrote: >>> >>> Our current policy is that we accept patches attached to JIRA case and pull >>> requests to https://github.com/apache/calcite >>> <https://github.com/apache/calcite>. I would like to propose that we no >>> longer support patches. >>> >>> Why? I argue that it makes the process easier for the committer. The pull >>> request implicitly does “git add” and “git remove”, whereas when applying a >>> patch you have to remember to apply these. The pull request comes in a >>> branch, so if I modify the code as I am reviewing it, I can easily save and >>> restore my state. Also, a pull request is “valid” as a contribution, from >>> an IP standpoint, even when not accompanied by a JIRA case. >>> >>> Recently I went through 5 rounds of patches for a particular feature. I >>> couldn’t tell what had changed between one iteration of the patch and the >>> next (you can’t “diff" patches - you need to apply the patches to separate >>> git branches and diff the branches - yuck!). And I went through 3 test >>> cycles and 24 hours before I managed to “git add” all of the files. Yes, I >>> did “git status” and I missed the 2 new files among all of the “.orig” and >>> “.rej” files in my sandbox. >>> >>> In summary. I propose that we accept contributions only as pull requests to >>> https://github.com/apache/calcite <https://github.com/apache/calcite>. If >>> they are non-trivial they should be accompanied by a JIRA case. Committers >>> can propose changes any way they like, as long as they commit the changes >>> themselves, but if they want to make it easier for others to review, they >>> should use either a personal git branch or a pull request. >>> >>> Julian >>> >> >
