I use both Mercurial and Git, and honestly, they aren't that different. Unless you're doing something very specific I don't see why you would choose one over the other.
I've never used mercurial outside of TortoiseHg. In fact, the only time I've used mercurial is when I was attempting to build Firefox from source before I realized that they do a git mirror thing so you don't have to touch mercurial at all.
I don't know much about mercurial. I use git because somebody said it is easier to rewrite history in git.
Do you ever write code, test it, and then edit that code some more? If so you just "rewrote" history. The history in the VCS is whatever you decide it to be when you actually commit. You can carefully shape what the history looks like with any VCS by being careful about what you commit. Git and mercurial simply allow you to shape the history after committing as well.
> Subversion uses a centralized revision control model. Ben Collins-Sussman, one of the designers of Subversion, believes a centralised model would help prevent "insecure programmers" from hiding their work from other team members.
Note that Ben (a personal friend) has been an enthusiastic user of Mercurial for several (4? 5?) years, and is pretty happy with it. He was even using hgsubversion for a while I think.
Note that there's a lot of social problems around DVCS, and the thing Ben worried about is still a _real problem_. I get potential contributors showing up with months of work, and it's an enormous pain in the ass (both for them to rewrite and for me to beg them to work with me) when they did something architecturally unacceptable early in the process that a simple 2 minute review could have caught.
But would the insecure people have done that work at all if they knew they would be forced to commit it publicly? Maybe so, but would it all be one huge commit to svn instead of maybe a string of more concise commits to git or mercurial?
I just envision someone insecure (like I feel at times) saying, well, I'll just clone this and play around on my own machine. Then after making some commits and building up courage, saying, OK, I'll try and go public with this. With centralized VCS that seems less likely to me. The insecure programmer just won't even try. I don't know, it's probably different for different people.
First of all, the history re-writing of both git and mercurial do not throw any history away[1], let's be clear on that. The history re-writing allows you to commit willy-nilly as often as you want, and then allows you to clean up those commits to make them readable, useful, and presentable to your team by combining commits, reordering commits, deleting commits, etc. The freedom to commit (and have an undo-point to go back to) at any time is very liberating and anxiety reducing :-)
Using VCS's other than git and mercurial people still end up doing essentially the same thing, they are simply required to do all the cleanup outside of the VCS before they commit. Almost nobody commits every sloppy, intermediate, untested edit of their files to svn or perforce. Users of those VCS's "rewrite" and "throw away" history just the same.
Footnote:
1. They both keep the original commits around after they are replaced by "rewritten" or "edited" commits. git does eventually delete the old history with its garbage collection, but it takes a number of weeks for that to happen. Mercurial saves backup bundle files (or obsolete changesets if using the evlove plugin) indefinitely. You have to manually delete them.
For me, I tend to commit frequently, sort of like my obsession with constantly hitting CTRL-S while working in an IDE. Before I push my changes I like to squash the commits into more cohesive commits. If anything, I think it makes it easier on my colleagues for code review.
I do the same in hg. I do `hg amend` all the time to keep adding changes to my commit. At the end I may selectively undo some changes with `hg uncommit --interactive` or `hg uncommit --all` and redo the whole thing piecemeal with `hg commit --interactive` in order to slowly split up my work into several commits. Evolve makes it really easy to keep (and ignore!) a meta-history of all of my editions, with a clear lineage of which new commit replaced which prior commit.
I may also rebase my work onto the latest head at the end (not to be confused with git HEAD).
And all of this with a very nice interface. It's always --interactive, not sometimes --patch and sometimes --interactive.
Mercurial was my first DCVS and I loved using it, but I only touched on the basic features. I stopped using it when I switched jobs and to become proficient with git. I'll have to give it another try with one of my side projects.
No it's not, it rewrites and destroys history. The problem it solves is "keeping the history clean", when actually it's not what you want. You don't see your history, you see logs and diffs. So you want to keep your logs and diffs clean. There should be an other solution to this other than simply removing/merging commits and eventually removing information with it.
We had the same discussion in my office.
If I am developing a feature (in a separate branch) - why shouldn't I rebase to make a single commit out of this feature ?
Is this really interesting for you that it took me 10 commits to do sO ?
Even if one of the commits was just a "I have to switch an fix a bug and want to commit before switching to another branch" - commit ?
I think there is nothing wrong with rebasing and to merge commits before pusing them.
I don't think it's wrong when using git. However I think it shouldn't be designed like this though. Git rebase should add metadata not remove them. Essentially you rebase because you want to hide certain details because they are not important. Rebase could do one or more actual merges instead and mark certain commits not important so they wouldn't show up in logs by default.
AFAIK rebase is the only operation that happily destroys your history without any --force or --hard option. It always disturbed me.
There is a legitimate usage of rebase though: you really want something removed from history. Like you accidentally committed a secret, inappropriate message or you just want to fix a typo in the last commit.
The way I understand it, I should only rebase (for squashing, not in the case of a spilled secret or inappropriate message) if I haven't pushed it to a shared remote branch. Is that correct?
if you read my post, "history" on local branches is irrelevant. They do not exist and "destroy" here is totally fine. It's not like it's meant to be a grand book detailing all the work done by you, in every point in time, as a database record. What you want here instead is just a clean patch coming off your topic branch -- which this does perfectly.
This is SUPREMELY valuable for OSS patch workflow and I absolutely love git for having it.
Some of the arguments against merging made in this article seem a little strange, for example saying "If folks on topic branches rebase, there will not be any conflicts on merge" after previously stating "It’s also a great idea to rebase periodically - several times a day", as there would also be no merge conflicts if the branch was merged in periodically as well. If merge commits are so abhorrent, just alias
git log
to
git log --no-merges
I use both Mercurial and Git, and honestly, they aren't that different. Unless you're doing something very specific I don't see why you would choose one over the other.