4 ms·
In the sense that git was noticeably "better" than the centralized systems that came before it, is enough wrong with git to not use it these days? Although ther
by mhh__ 6y ago
In the sense that git was noticeably "better" than the centralized systems that came before it, is enough wrong with git to not use it these days? Although there are pain points with git, a large majority of git activities are fairly cookie-cutter
- gnufx 6y agoThat seems to suggest there weren't distributed systems before git (like the proprietary one git was written to replace). I think there's enough wrong with git after long experience with revision control in collaborations; I've been quote happy with Darcs.
- stevekemp 6y agoDarcs was technically interesting, and also a joy to use. However it suffered from the exponential merge problem, which I think is still not fully resolved. Once you hit that all the theory and nicities went out of the window.
- gnufx 6y agoThe merge complexity in darcs 2 really hasn't been an issue in my usage. As I understand it, it is resolved for darcs 3, but maybe I misunderstood that from skimming what was going on. (The new theory seems to be different to Pijul's, and perhaps inferior.) I'd be happy with either as a joy to use.
- stevekemp 6y agoI/we definitely hit it in real life, sufficiently often that it became obvious we needed to move to another solution. At the time it was probably CVS to darcs, then it became darcs -> hg, and giving in to the inevitable it became git.
- chriswarbo 6y ago> Darcs was technically interesting, and also a joy to use. However it suffered from the exponential merge problem, which I think is still not fully resolved. That's kind of the point of Pijul
- stevekemp 6y agoYeah if it can be solved then that's great news. But having tried "everything" I think it would be a tough sell to switch at this point. The momentum is clearly behind git, and has been for quite some time. None of the more recent contenders have ever become popular (bzr, fossil), though of course they have their advocates. Having the code be reworked in-private, for a long time, too doesn't make it seem like it is ready to become usable in the short-term.
- peteretep 6y agoIrritating that this is downvoted. I agree with you though — I simply don’t run into enough (any?) git related pain that would motivate a switch. git is really really simple and can be made to do whatever I want
- alwillis 6y agoMany Git users act as if they have Stockholm Syndrome[1]; because of their dependence on it, they don't see all of the issues with it and make excuses for its shortcomings. I’m old enough to remember the same thing with MS-DOS when GUIs first started to become mainstream. Git is extremely flexible and allows you to implement almost any workflow you want. But its command line interface isn't intuitive and there are still enough opportunities to shoot yourself in the foot. It's still too easy to get yourself in a situation where you need the Git expert on your team to unwedge the repo. Sure, I'm a Git user, but not because I wanted to be one; it's the de facto standard and there's a large ecosystem of support for it, starting with GitHub and GitLab. I use Mercurial for personal projects because I don't want or need the cognitive overhead of Git when I'm doing something for myself. Out of the box, Mercurial doesn't give you access to a foot-gun unless you really want it. But when you're starting out, you don't want to shoot your foot off before you know what you're doing. I enjoy Mercurial’s concept of changeset evolution[2] and Phases[3] which helps avoid all of the coordination and potential drama of rebasing on top of changes that shouldn't have been changed, etc. The Evolve plugin[4] allows for intelligent and sane history rewriting, using the underlying support for changeset evolution. The Evolve plugin also provides support topic branches[5], reducing the overhead of managing them. There's a great description of how this works in Mercurial on the Git Minutes podcast[6]. [1]: Stockholm syndrome is a condition in which hostages develop a psychological alliance with their captors during captivity. Emotional bonds may be formed between captors and captives, during intimate time together, but these are generally considered irrational in light of the danger or risk endured by the victims. [2]: https://www.mercurial-scm.org/wiki/ChangesetEvolution https://www.mercurial-scm.org/wiki/ChangesetEvolution [3] https://www.mercurial-scm.org/wiki/Phases https://www.mercurial-scm.org/wiki/Phases [4]: https://www.mercurial-scm.org/doc/evolution/user-guide.html https://www.mercurial-scm.org/doc/evolution/user-guide.html [5]: https://www.mercurial-scm.org/doc/evolution/tutorials/topic-tutorial.html https://www.mercurial-scm.org/doc/evolution/tutorials/topic-... [6]: https://episodes.gitminutes.com/2013/05/gitminutes-07-martin-geisler-on.html https://episodes.gitminutes.com/2013/05/gitminutes-07-martin...
- greggman3 6y agoIt probably depends on your domain. AFAIK AAA games all use Perforce. Git is just not good managing terabytes of source game assets nor at managing access to those assets so two people don't edit them at the same time since they are usually not mergable. git lfs was supposed to help but it only solves the storage side, not the permission side, and it's pretty slow compared to p4. I doubt pijul will make a dent in solving AAA game devs vcs issues though
- kortex 6y agoLfs is frustrating. The clean/smudge model means files could be in the state of pointer or actual data. Lost count of how many times I didn't have everything smudged, or accidentally commited a 500mb model, or gotten into some wacked-out state where I couldn't clean, smudge, push, or pull without errors and had to nuke and start fresh from remote. Data Version Control (dvc.org) is a far better model.
- pmeunier 6y ago> I doubt pijul will make a dent in solving AAA game devs vcs issues though Pijul has the potential to do that, since patches provide a very easy interface for doing partial checkouts: - just pull patches related to a particular part of the repository. - don't pull patches related to artwork frequently. Since Pijul patches always commute, this could work. Additionally, I hear the new (unreleased) Pijul can split patches between a "transform part" (e.g. "I inserted 2Tb in file f, and deleted 1") and a "contents part" (e.g., "here, have your 2+1Tb: …").
- mleonhard 6y agoGit works well for large distributed teams. Very little software is developed like that. Most software is made by small teams. For small teams, git's model adds extra overhead. Small teams can be more productive with a simpler source repository. Release processes benefit from incremental commit numbers. They are easy to reason about in release planning. They are useful for gating features. Example: Wait until api server deploys commit number N or higher before enabling app feature A. They provide an easy way to track the age of binaries running in production. Example: We can turn down API B once every deployed production binary is built from commit number N or newer, since that is when we deleted the API B client library. Merging and rebasing is easier to understand with incremental commit numbers. Git has no support for code reviews. For example, it provides no functions for distributing versions of uncommitted changesets, distributing comments on uncommitted changesets, or linking comments to particular versions of changesets. These are core parts of the software development process. Code review comments are crucial context for future readers to understand the code. They belong in the repository with the code. Git's "stash" function is full of footguns and is therefore mostly unusable. Git's nomenclature is inexact. A git "commit" is actually a set of changes to files. It would be better called a "change set", "change list", or even "diff". Git's documentation needs work. It needs to explain common use cases. It needs to warn of more pitfalls. It needs more common use case examples. For example, https://git-scm.com/docs/git-revert https://git-scm.com/docs/git-revert has no example of specifying a particular commit or a set of commits. Another example, https://git-scm.com/docs/git-stash https://git-scm.com/docs/git-stash is missing a link to the article https://git-scm.com/book/en/v2/Git-Tools-Stashing-and-Cleaning https://git-scm.com/book/en/v2/Git-Tools-Stashing-and-Cleani... and that article needs a big fat warning about the biggest footgun: using 'stash' and then 'reset' and accidentally erasing untracked files that you thought were stashed. Git sees every source file as plain text, ignoring structure. For example, reformatting destroys blame info. Changeset metadata includes lines added/deleted/modified but treats whitespace changes the same as changes to comments, data, and logic. Git provides no access control functions. Users must rely on third-party tools. Users need to apply ACLs to source, code reviews, and test output. This is complicated to set up with third-party tools. I use git and I look forward to switching to a better tool in the future. Maybe Pijul will be it?