I always use a soft reset when I accidentally commit in the wrong order, because I like keeping my history tidy. If the change is live, I always use revert.
On another note, I love how HN can go from complex research papers I struggle to keep up with, to explaining git commands the next day — something I learned within my first month of Git and never had a problem with again. How bipolar.
I do already do! :) I use interactive rebase to squash multiple commits together when I need it.
However, I just find it easier to start from scratch if I changed 4 files at once, wanted the changes spread out over 3 commits, and I did those in the wrong order or something.
Interactive rebasing works fine for teasing apart commits as well (maybe slightly clumsier, but c'mon... it's git, it's _supposed_ to be a frustrating UX)
Even if you're on your own branch you can easily get a DAG after rebasing or resetting. The reflog keeps track of the history of all refs (ie. branch names) that you change. After you rebase your old commits are still there! It's actually quite hard to remove commits.
Trusting the reflog is key to becoming good at git.
I use interactive rebasing as the backdrop for teasing apart commits that are not the HEAD commit; it provides the navigation through the commits.
The cherry-pick operations that underpin rebase are not what actually does the teasing-apart, though; I do that with with git reset --interactive HEAD^ when I've landed using "edit" on the commit I want to separate.
I think the main thing is, if you've already pushed to somewhere where others will pull from then use revert. Otherwise, use rebase (or reset+cherry-pick or amend, etc). If you push a patchset that has a patch and also a revert patch for that patch then you're just adding completely unnecessary noise for anyone who is reviewing the patches.
The chunk about `git reset --hard` should be prefixed with a big warning label saying "Don't use this unless you know exactly what you're doing". There are very few legitimate reasons to use `git reset --hard <commit>` in the way that the article describes and even fewer if you're working with remote. Now `git reset` (No commit or --hard) can be useful to retry a screwed staging, but that's about it. As to squashing commits... that's a whole different set of problems when you're dealing with auditors and code review requirements and "public" commits.
Very few? I find myself having to use it a lot (usually targetting HEAD or HEAD~1), and actually it's kind of handy for clearing out any cruft when switching between project branches. The important thing to know is that it only acts on files that are added to git, and will pass over any that are not.
I've used `reset` that way a few times, but pretty much never `reset --hard` Still, this is not something that should be suggested to someone unfamiliar with git. It's just too easy to screw up your codebase.
In the last example (which is actually the most common case in practice), they give wrong instructions. They should only revert the bad commit, not all the following commits. Reverting everything is impractical and useless, and if you want to do that you may as well use revert (which is still a bad idea).
I love how the illustrations are actually the obvious default interface that everyone should be using to edit nodes in a graph. Click "Delete" while the box is around C and D: done. (Or right clicking and choosing "commit a reversion" or whatever.)
If it is about memory, why not generate 4 or so words as the 'visual' hash, which is probably good enough for most teams to use for branching/merging without experiencing clashes.
On another note, I love how HN can go from complex research papers I struggle to keep up with, to explaining git commands the next day — something I learned within my first month of Git and never had a problem with again. How bipolar.