On 1 June 2016 at 19:36, Aaron Ballman <aaron at aaronballman.com> wrote:> Despite people's reservations of a git-only repository?Hi Aaron, not at all! I was especially vague on my first email to make sure SVN folks would be shoved on the side, but John had asked for a full plan *in the case we move*, and I was just completing the picture. Having said that, I can't take that decision alone, and my own opinion is irrelevant on the grand scheme. Right now, our main repo is in SVN with most people using Git. If the vast majority vote for the move, it wouldn't be fair to continue to force SVN on them, and it would be overall less effort for the few people that prefer SVN to have a bit more work than they have today, to save the majority of Git users the extra work. I have no idea how much people is enough to move to Git, but unless we fix the sub-module problem, there's no point in even trying. So, my personal points are: 1. We can only move IFF the Git solution is technically equivalent or superior than what we have today. 2. We should only move IFF the vast majority will see benefits from it, even if a small minority will see some increased effort. Of course, the balance of efforts has to be overall positive. 3. We should not move if there is no replacement for SVN users at the moment. We should try to encourage SVN users to move to Git, to speed up the move, though. I'm assuming the SVN vs. Git argument is not just a personal thing, but a tooling / infrastructure issue. The bigger picture here is not which VCS is better, but getting rid of a huge infrastructure cost from our part, which nowadays means moving to Git or using SourceForge. cheers, --renato
On Wed, Jun 1, 2016 at 3:10 PM, Renato Golin <renato.golin at linaro.org> wrote:> On 1 June 2016 at 19:36, Aaron Ballman <aaron at aaronballman.com> wrote: >> Despite people's reservations of a git-only repository? > > Hi Aaron, not at all! > > I was especially vague on my first email to make sure SVN folks would > be shoved on the side, but John had asked for a full plan *in the case > we move*, and I was just completing the picture.Oh! That makes perfect sense to me. Thank you for the clarification!> Having said that, I can't take that decision alone, and my own opinion > is irrelevant on the grand scheme. > > Right now, our main repo is in SVN with most people using Git.Our main repo is in SVN; I would say we don't know what most people are using (aside from "svn for write access because it's the only option").> If the > vast majority vote for the move, it wouldn't be fair to continue to > force SVN on them, and it would be overall less effort for the few > people that prefer SVN to have a bit more work than they have today, > to save the majority of Git users the extra work. I have no idea how > much people is enough to move to Git, but unless we fix the sub-module > problem, there's no point in even trying.Fair points, but with the caveat that people using git today have a workable solution (as I understand it, and I could be totally wrong) using the git mirrors. That's not a reason to not transition from svn to git, however.> So, my personal points are: > > 1. We can only move IFF the Git solution is technically equivalent or > superior than what we have today. > > 2. We should only move IFF the vast majority will see benefits from > it, even if a small minority will see some increased effort. Of > course, the balance of efforts has to be overall positive.Agreed with both of these points.> 3. We should not move if there is no replacement for SVN users at the > moment.Agreed.> We should try to encourage SVN users to move to Git, to speed > up the move, though.This is implying that we will move, which I think should still be left as a vague question mark until we have more answers. Based on that, I think it's premature to encourage anyone to switch to git.> I'm assuming the SVN vs. Git argument is not just a personal thing, > but a tooling / infrastructure issue. The bigger picture here is not > which VCS is better, but getting rid of a huge infrastructure cost > from our part, which nowadays means moving to Git or using > SourceForge.I agree that there are infrastructure costs to consider. I just hope we don't consider those at the expense of a functioning system that people are used to using and already have workflows based on. Git may be the new shiny, but it's not what we use today for LLVM. It's sometimes easy to forget there's also cost with telling the community "please go learn a new, very different toolset so that you can continue to contribute to the project." I'm not making a claim that the costs aren't worth the gains (because this might very well be the correct time to switch VCS), but I am worried when emails make it sound like switching to git-only is a foregone conclusion, which is a bit of a strange way to start a discussion about whether the community wants to switch. That being said, I like that we're discussing what a switch would look like were one to occur so that we can suss out all the pros and cons! ~Aaron
On 1 June 2016 at 20:31, Aaron Ballman <aaron at aaronballman.com> wrote:> Our main repo is in SVN; I would say we don't know what most people > are using (aside from "svn for write access because it's the only > option").If the LLVM Meetings are any indication, and they are at least related to the most active developers, everyone I've asked (and I did ask a lot over the last 2 years) was using Git for development. Most people were using Git-SVN for commits, with some few still using a separate SVN repository. I'm not really too worried about personal preferences. The infrastructure cost alone is more important than any of our preferences. That is why I have been using SVN all these years, even though I hate it more than I once hated CVS. The admin cost is high, and we're not sysadmins, so GitHub would provide an *immense* value for the price of $0. I can't argue we that. But there are also downstream infrastructure issues that need to be taken into account. As James said, knowing Git well is almost a required skill nowadays, and everything is done thinking about Git these days, so the likelihood that you'll find a replacement for your company's infrastructure to work with Git is higher than a new process will work well with SVN. In the long term, SVN will be just like CVS was 10 years ago or RCS 20. It'll be impossible to work with them, and companies depending on that infrastructure will be in *serious* trouble.> Fair points, but with the caveat that people using git today have a > workable solution (as I understand it, and I could be totally wrong) > using the git mirrors. That's not a reason to not transition from svn > to git, however.Indeed. It's all about the overall costs for the whole community. Personal preferences will be mostly diluted, with strong technical arguments being the driving forces.> This is implying that we will move, which I think should still be left > as a vague question mark until we have more answers. Based on that, I > think it's premature to encourage anyone to switch to git.I will (only slightly and personally) disagree, based on James' point of how important Git is today. Ever since I discovered Git I have always encouraged people to use it, regardless, and I'll keep encouraging. :) But that's irrelevant to the discussion.> but I am worried when emails make it > sound like switching to git-only is a foregone conclusion, which is a > bit of a strange way to start a discussion about whether the community > wants to switch.I think that's just the result of people enthusiastic with the opportunity to move to a better working environment. Not everyone is, but those that are, are showing. This is another indication that there are more Git users than SVN users, but not an argument to force the move. The only arguments we should accept are technical ones. cheers, --renato