In Git, files don't have history. Instead, commits are history. The history in a Git repository is found by:
- picking a branch name (or some other label that lets you find a commit's hash ID);
- observing that commit, as found by its hash ID;
- using that commit, which contains the hash ID of one or more earlier commits, to step back to an earlier commit.
It's this stepping-back, from commit to earlier commit, that is the act of exploring history. Hence the commits are the history.
Now, each commit contains files: in fact, each commit has a full copy of every file, saved in a read-only format, compressed and—since most commits mostly copy some previous commit's files unchanged—de-duplicated. So given some set of commits that are history, we can have Git filter away some or even most of the history, showing us only those commits that Git thinks we'll find particularly interesting. This is how git log -- path and git log --follow -- filename work. They look at some subset of the actual history—some set of commits—and simply lie to you, using carefully selected, programmatically-controllable lies that give you the useful information.
You must keep in mind that these are lies that you control with options like --simplify-by-decoration or --full-history. Otherwise you'll be fooled by the lies. The actual history is the commits and Git is filtering some away.
Making a new commit leaves all the existing commits unchanged. It must because no part of any commit can ever be changed, not even by Git itself. (This is how Git manages to do the "distributed" part of being a Distributed Version Control System.)
