In the FreeBSD ports tree I restored the previously removed directory
math/dynare:
1. git checkout 3fb72e318f7be~1 -- math/dynare
2. git mv math/dynare science/dynare
3. rm -rf science/dynare && cp -r science/dynare.new science/dynare
4. git add science/dynare
5. git commit science/dynare
After this 'git log science/dynare' only shows my last commit, and not
the previous history of 'math/dynare' that existed before its removal.
The above sequence of commands should have kept and migrated the history
of math/dynare to science/dynare, but it didn't.
The corresponding subversion command sequence keeps history.
Yuri
On Sat, Aug 14, 2021 at 2:21 AM Yuri [off-list ref] wrote:
In the FreeBSD ports tree I restored the previously removed directory
math/dynare:
1. git checkout 3fb72e318f7be~1 -- math/dynare
[remaining complex shuffling action that mimics what's needed
in SVN, snipped]
Git is not SVN; do not expect it to behave *like* SVN.
In particular, Git does not have file history. Git has only commits.
SVN has commits, but SVN commits contain file history. In Git, the
commits *are* the history, and are the *only history*. Each commit has
a full snapshot of every file, under the names you told Git to use, with
the contents that you told Git to save.
The command `git log science/dynare` tells Git: Look through
commits, one by one, and whenever there's a file whose name
starts with `science/dynare` that's different in *this* commit, vs the
previous one, tell me about this commit. Since the old commits
have files named `math/dynare/something` instead, Git doesn't
tell you about them.
The `git log` command has a flag, `--follow`, that has a trick up its
sleeve (wait: does `git log` have sleeves? well, anyway). This trick
is a bit limited though. Let's say there's a README file in the
science/dynare/ directory, as of the latest commit. If you run:
git log --follow science/dynare/README
Git will look at *this* commit and find the file, and will look at this
commit's *parent* and *not* find the file, because the parent does
not have `science/dynare/README` in it. But the parent does
have `math/dynare/README` in it *and that file exactly matches*
this commit's copy. So `git log` will tell you about this commit and
say that the file was *renamed*. From this point backwards,
`git log` will be looking, not for `science/dynare/README`, but
instead for `math/dynare/README`.
Unfortunately, the `--follow` trick only works on *one file*. But it
will let you know that if you re-run your `git log` with `math/dynare`
you can see what happened before the renaming.
A future `git log` really should be smarter about detecting the
whole-subtree renaming, and be able to follow it. This won't
require any change in the repository, just a smarter `git log`.
But someone has to write that smarter `git log`.
Chris
I should add: to make this work in that future smarter
`git log` command, be sure to:
1. commit the un-removal first; then
2. commit the rename (you can just `git mv` the
directory and commit)
so that the future smarter `git log` has an easy way of
finding this. Ideally, it would find this even without that,
but having all the files / sub-trees have identical hash IDs
will make this much faster and easier.