Jonas Lindqvist

I begin with the three trees and the staging area, because every confusing command is easier once you know which one it moves. The working tree is what you edit, the index is what the next commit will record, and the repository is what is already committed. That model settles a surprising number of arguments. A commit takes what is in the index rather than everything in the working directory, which is why half-finished work survives a commit. Files added to .gitignore only stop being tracked once they are untracked explicitly, and that step is the one people miss. The commands whose names mislead are the ones I spend time on. Reset moves the current branch pointer, which means the default form can discard committed work, while restore moves files without touching history. Checkout combines both jobs, which is why it does things nobody expects when switching branches with local modifications. Rebase versus merge is a decision about history shape rather than correctness, and it interacts badly with commits that have already been published. Merge commits preserve the record. Rebase rewrites it, and rewriting published history breaks everyone else working from the same branch. The reflog is my safety net. It records where each branch reference pointed, and it means almost every destructive mistake is recoverable for weeks. Cherry-pick applies one commit elsewhere and creates a new one with a new identity. Bisect walks history to find a first bad commit, which turns a vague report into a specific range. I am deliberately conservative about advising rewrites on shared branches, because that is where a helpful command turns into somebody's lost afternoon.