On 6 January 2016 at 20:44, Thomas Martitz kugel@rockbox.org wrote:
Am 06.01.2016 um 04:32 schrieb Matthew Brush:
Agree, I sometimes avoid putting LGTM when I think something is a good idea, because I don't want to give the impression that I have (or even will) reviewed or tested it. Maybe just a "thumbs up" could mean "good idea, though I haven't reviewed or tested it"?
Github is a terrible code review platforms. Other platforms do much better:
- Differentiate between [x] I have reviewed [x] I have tested [x] I like
the change (no test or review)
Just say it in plain old English :) "I (do/do not) like the idea, I (have/have not) reviewed the implementation and I (have/have not) tested it on (win/lin/both)." Thats not a committer/dev prerogative, anybody can comment.
- Handle updated changesets without losing comments, to the point that you
can even browse older revisions
Yeah that sucks.
- is a problem for reviewers and 2) is a problem for everyone
There are a lot of other, much better code review systems, but I guess we're stuck with github (bought into proprietary solution, anyone?)
To give an example of such a "functionality review", the fractional font sizes patch
https://github.com/geany/geany/pull/407
LGTM - even though I probably won't use it myself, I understand it may be useful for someone and it modifies just 23 lines so it's nothing intrusive and can go in from my point of view.
Agree, and I've actually wanted for this before.
Such a review can be done in a few seconds so if everyone goes through the new pull requests from time to time, the patches will receive some feedback and it will be clearer whether it's something others want it in Geany.
- Unclear status of some patches. Sometimes it might not be clear in
what state the pull request is - I'd suggest adding at least the following two tags:
needs-work (reviewed with some comments that need to be addressed) work-in-progress (not meant for review in the current state)
This will help to distinguish pull requests awaiting merge and pull requests that aren't there yet.
Sounds like a good idea, I just added those labels. We also already have "reviewed", which I believe means whoever added the label has fully reviewed a PR and it's ready to be merged.
I can't put labels on github. The feature seems to be limited to those with write access to the repository, so it's useless if we want more reviewers.
Please just add a comment, its actually clearer anyway since we don't have a well defined semantics for the labels :)
- Fear that the pull request isn't tested enough. I believe that if a
patch did undergo a review and there doesn't seem anything obviously wrong with it, it can be merged to master. I think there's no need for some long-term private testing of a patch before it gets merged - people using the development versions of Geany should be aware it may contain bugs and should know how to deal with them. There are also more eyes so a potential bug is spotted earlier. Of course it makes sense getting more conservative towards the end of the development cycle to stabilise things.
This is a big one too. Since we don't have a "development" branch (or rather "master" is the development branch), we often have too high standards, only merging stuff that is 100% finished, debugged and ready to release. This kind of defeats the purpose of a development branch.
Assuming there's a general consensus that a PR is a good idea, I agree we should start integrating it early, even if there's some possibility it could break things. This way it can get tested in real-life by more than just the person who initiated the PR and others could/would make their own PRs to fixup any perceived issues. When a PR sits off on someone else's repo, it doesn't get real testing, and people (rarely) make PRs against the repo where it came from, where they more likely would if it was in the main repo.
I agree. Currently there is a culture that everything must be perfect prior to merging. This is fine in theory but simply doesn't scale. It requires endless iterations which in turn require a lot of re-testing, and results in PRs sitting for way longer than necessary. Up to the point where the rate of new PRs can't be handled.
I agree that PRs should be merged earlier, with possible fix-up/follow-up commits in a new PR. This way changes acutally get the testing they need.
The problem with a development branch and with making master more buggy/less stable is that they will be used less, since its dangerous to use them day to day. So the testing will go down even more.
OK, these are things that come into my mind regarding the zombie pull requests but there may be other problems. What do you think? Do the points above look reasonable to you and do you think they might help a bit? Is there anything else that might help killing the bloody zombies?
- Lack of collaboration; We tend not to use Git to its full potential by
making pull requests against active pull requests. I don't know if it's a lack of Git knowledge or that it's too much hassle, but we rarely seem to do this (I've only done it a couple times, Colomban does it more often I think). Instead we make comments like "you should do this or that", assuming that the only person who must make changes is the pull requester. If instead of making "do this" comments, we made pull requests to show/implement it, it would give something concrete to discuss and increase collaboration on the code.
If there is not enough effort to keep up with the commits, there is even less effort available to make pull requests on top of other peoples.
It's awkward to use because it's not visible on the original PR. It's often hard enough to monitor one PR, let alone recursively monitoring PRs in other repos. Also if there are more than 2 people involved at least some of them don't see the PR-ontop-of-PR. Lastly, if you actually merge those PRs you get a weird history with "Merged Pull Request #XX" from different repos (of course XX changes widely).
I think gists linked to from the original PR would be more natural.
Don't know a good solution to this one, maybe if you want to actually make changes, just mail the patches to the OP and let them put them up on the PR.
- This one applies to many of your PRs Jiří :) Speaking only for myself
here, any PRs made against CTags or TagManager I pretty much ignore. Not only that I'm completely unfamiliar with their code, but also the changes are often very subtle and require in-depth knowledge of those code bases, and I'm neither qualified nor inclined to make any judgments about the PR, let alone review or merge them.
We need other means to ensure quality if nobody is up to reviewing them properly. These code areas could make good use unit tests so that merge changes there can be merged with confidence even if they are not deeply reviewed.
Yes, better tests for the non-GUI areas would make them easier to change. Your offer to write the tests is accepted :)
- Monster PRs. This goes along with #3. Sometimes PRs are just too big to
be digested by volunteer developers all at once. If we started integrating such larger changes earlier, and in smaller pieces, I think it would increase the likelihood of getting the feature/changes merged into the master branch. When a PR is so big, it basically takes as much time to fully review and test it as it did to make the changes in the first place. I personally rarely merge pull requests unless I've reviewed and understand each line of code that changed, and tested all the cases I can think of. As volunteer like the rest of us, I just simply can't afford to spend days/weeks reviewing, understanding, and testing such large PRs, especially if I'm rather indifferent on the feature/improvement they implement.
Yes, monster PRs are difficult to deal with. But on the other hand I get the response that "I don't like to merge this if nothing uses it right now" (when making a groundwork change required by a later one) when attempting to make smaller, incremental changes. Both views are reasonable. Plus, changes that merely lay the groundwork and don't have any impact itself seem to have very low review priority and are merged with large delay, if at all.
I also have learned that breaking up large changes into lots of small commits doesn't help either. It still makes the PR as a whole rather large and lots of commits are not handy to deal with either.
You are right that individual commits on a monster PR don't help much unless they can be reviewed, tested and committed individually.
The way massive changes are handled on Julia is for the OP to put up a PR (or just an issue) with a list of the steps required for the whole change. Individual PRs that make preparatory changes then refer to this so there is a context for them and they can be smaller and easier to review/test.
And wherever possible breaking changes are made in such a way that they can operate in parallel with the existing capability so it can be deprecated for a decent period before removal. On that topic I would like to congratulate you on accepting the challenge of making your recent big PRs retain backward compatibility, its not always easy, and can seem like just an annoyance, so well done.
- Lack of committers/commits. There's 7 people who have write access to
the repo. Of them, Colomban probably does > 95% of the actual merges/commits. IMO, we either need more people able to assign themselves and merge pull requests, or we need to spread the workload out between those who already can. Personally, when I get some time, I go through and cherry pick the very simple/trivial PRs and try and merge them, but it's not very fair to leave all the really hard ones for Colomban (not to put words in his mouth). If this existing committers don't have enough time or interest to keep on top of pull requests, then all we can do (besides status quo) is to have more interested, trusted developers able to merge pull requests.
I agree. The more the merrier, if it helps spreading the workload. Jiří is an awesome candidate, especially since he's the official MAC guy.
I'd volunteer as well. However, given how I'm constantly failing to do good PRs initially or even after 2 or 3 revisions (by Colomban's standards), I'm not sure I'm suitable.
The MOST important thing for a commiter is not killer programming skills, its maturity. Can the person be trusted to not commit their own PRs without others agreement? Can the person be patient enough to wait for comments and other people to test, not everybody is available every day. Does the person understand their own limits, will they ask before committing to an area they have never touched before? Having commit rights is not about getting the persons own PRs into Geany, its about getting others changes in.
So if you think you can handle the above, I don't see why both you and Jiri would not be acceptable (assuming Jiri is interested, I don't know if he was asked :).
Cheers Lex
Best regards.
Devel mailing list Devel@lists.geany.org https://lists.geany.org/cgi-bin/mailman/listinfo/devel