How do I make a Git commit in the past?

I'm converting everything over to Git for my own personal use and I found some old versions of a file already in the repository. How do I commit it to the history in the correct order according the file's "date modified" so I have an accurate history of the file?

I was told something like this would work:

git filter-branch --env-filter="GIT_AUTHOR_DATE=... --index-filter "git commit path/to/file --date " --tag-name-filter cat -- --all  

I know this question is quite old, but that's what actually worked for me:

git commit --date="10 day ago" -m "Your commit message" 

The advice you were given is flawed. Unconditionally setting GIT_AUTHOR_DATE in an --env-filter would rewrite the date of every commit. Also, it would be unusual to use git commit inside --index-filter.

You are dealing with multiple, independent problems here.

Specifying Dates Other Than “now”

Each commit has two dates: the author date and the committer date. You can override each by supplying values through the environment variables GIT_AUTHOR_DATE and GIT_COMMITTER_DATE for any command that writes a new commit. See “Date Formats” in git-commit(1) or the below:

Git internal format = <unix timestamp> <time zone offset>, e.g.  1112926393 +0200
RFC 2822            = e.g. Thu, 07 Apr 2005 22:13:13 +0200
ISO 8601            = e.g. 2005-04-07T22:13:13

The only command that writes a new commit during normal use is git commit. It also has a --date option that lets you directly specify the author date. Your anticipated usage includes git filter-branch --env-filter also uses the environment variables mentioned above (these are part of the “env” after which the option is named; see “Options” in git-filter-branch(1) and the underlying “plumbing” command git-commit-tree(1).

Inserting a File Into a Single ref History

If your repository is very simple (i.e. you only have a single branch, no tags), then you can probably use git rebase to do the work.

In the following commands, use the object name (SHA-1 hash) of the commit instead of “A”. Do not forget to use one of the “date override” methods when you run git commit.

---A---B---C---o---o---o   master

git checkout master
git checkout A~0
git add path/to/file
git commit --date='whenever'
git tag ,new-commit -m'delete me later'
git checkout -
git rebase --onto ,new-commit A
git tag -d ,new-commit

---A---N                      (was ",new-commit", but we delete the tag)
        \
         B'---C'---o---o---o   master

If you wanted to update A to include the new file (instead of creating a new commit where it was added), then use git commit --amend instead of git commit. The result would look like this:

---A'---B'---C'---o---o---o   master

The above works as long as you can name the commit that should be the parent of your new commit. If you actually want your new file to be added via a new root commit (no parents), then you need something a bit different:

B---C---o---o---o   master

git checkout master
git checkout --orphan new-root
git rm -rf .
git add path/to/file
GIT_AUTHOR_DATE='whenever' git commit
git checkout -
git rebase --root --onto new-root
git branch -d new-root

N                       (was new-root, but we deleted it)
 \
  B'---C'---o---o---o   master

git checkout --orphan is relatively new (Git 1.7.2), but there are other ways of doing the same thing that work on older versions of Git.

Inserting a File Into a Multi-ref History

If your repository is more complex (i.e. it has more than one ref (branches, tags, etc.)), then you will probably need to use git filter-branch. Before using git filter-branch, you should make a backup copy of your entire repository. A simple tar archive of your entire working tree (including the .git directory) is sufficient. git filter-branch does make backup refs, but it is often easier to recover from a not-quite-right filtering by just deleting your .git directory and restoring it from your backup.

Note: The examples below use the lower-level command git update-index --add instead of git add. You could use git add, but you would first need to copy the file from some external location to the expected path (--index-filter runs its command in a temporary GIT_WORK_TREE that is empty).

If you want your new file to be added to every existing commit, then you can do this:

new_file=$(git hash-object -w path/to/file)
git filter-branch \
  --index-filter \
    'git update-index --add --cacheinfo 100644 '"$new_file"' path/to/file' \
  --tag-name-filter cat \
  -- --all
git reset --hard

I do not really see any reason to change the dates of the existing commits with --env-filter 'GIT_AUTHOR_DATE=…'. If you did use it, you would have make it conditional so that it would rewrite the date for every commit.

If you want your new file to appear only in the commits after some existing commit (“A”), then you can do this:

file_path=path/to/file
before_commit=$(git rev-parse --verify A)
file_blob=$(git hash-object -w "$file_path")
git filter-branch \
  --index-filter '

    if x=$(git rev-list -1 "$GIT_COMMIT" --not '"$before_commit"') &&
       test -n "$x"; then
         git update-index --add --cacheinfo 100644 '"$file_blob $file_path"'
    fi

  ' \
  --tag-name-filter cat \
  -- --all
git reset --hard

If you want the file to be added via a new commit that is to be inserted into the middle of your history, then you will need to generate the new commit prior to using git filter-branch and add --parent-filter to git filter-branch:

file_path=path/to/file
before_commit=$(git rev-parse --verify A)

git checkout master
git checkout "$before_commit"
git add "$file_path"
git commit --date='whenever'
new_commit=$(git rev-parse --verify HEAD)
file_blob=$(git rev-parse --verify HEAD:"$file_path")
git checkout -

git filter-branch \
  --parent-filter "sed -e s/$before_commit/$new_commit/g" \
  --index-filter '

    if x=$(git rev-list -1 "$GIT_COMMIT" --not '"$new_commit"') &&
       test -n "$x"; then
         git update-index --add --cacheinfo 100644 '"$file_blob $file_path"'
    fi

  ' \
  --tag-name-filter cat \
  -- --all
git reset --hard

You could also arrange for the file to be first added in a new root commit: create your new root commit via the “orphan” method from the git rebase section (capture it in new_commit), use the unconditional --index-filter, and a --parent-filter like "sed -e \"s/^$/-p $new_commit/\"".

You can create the commit as usual, but when you commit, set the environment variables GIT_AUTHOR_DATE and GIT_COMMITTER_DATE to the appropriate datetimes.

Of course, this will make the commit at the tip of your branch (i.e., in front of the current HEAD commit). If you want to push it back farther in the repo, you have to get a bit fancy. Let's say you have this history:

o--o--o--o--o

And you want your new commit (marked as "X") to appear second:

o--X--o--o--o--o

The easiest way would be to branch from the first commit, add your new commit, then rebase all other commits on top of the new one. Like so:

$ git checkout -b new_commit $desired_parent_of_new_commit
$ git add new_file
$ GIT_AUTHOR_DATE='your date' GIT_COMMITTER_DATE='your date' git commit -m 'new (old) files'
$ git checkout master
$ git rebase new_commit
$ git branch -d new_commit

To make a commit that looks like it was done in the past you have to set both GIT_AUTHOR_DATE and GIT_COMMITTER_DATE:

GIT_AUTHOR_DATE=$(date -d'...') GIT_COMMITTER_DATE="$GIT_AUTHOR_DATE" git commit -m '...'

where date -d'...' can be exact date like 2019-01-01 12:00:00 or relative like 5 months ago 24 days ago.

To see both dates in git log use:

git log --pretty=fuller

This also works for merge commits:

GIT_AUTHOR_DATE=$(date -d'...') GIT_COMMITTER_DATE="$GIT_AUTHOR_DATE" git merge <branchname> --no-ff

In my case over time I had saved a bunch of versions of myfile as myfile_bak, myfile_old, myfile_2010, backups/myfile etc. I wanted to put myfile's history in git using their modification dates. So rename the oldest to myfile, git add myfile, then git commit --date=(modification date from ls -l) myfile, rename next oldest to myfile, another git commit with --date, repeat...

To automate this somewhat, you can use shell-foo to get the modification time of the file. I started with ls -l and cut, but stat(1) is more direct

git commit --date="`stat -c %y myfile`" myfile

This is an old question but I recently stumbled upon it.

git commit --date='year-month-day hour:minutes:seconds' -m "message"

So it would look something like this: git commit --date='2021-01-01 12:12:00' -m "message" worked properly and verified it on GitHub and GitLab.

The following is what I use to commit changes on foo to N=1 days in the past:

git add foo
git commit -m "Update foo"
git commit --amend --date="$(date -v-1d)"

If you want to commit to a even older date, say 3 days back, just change the date argument: date -v-3d.

That's really useful when you forget to commit something yesterday, for instance.

UPDATE: --date also accepts expressions like --date "3 days ago" or even --date "yesterday". So we can reduce it to one line command:

git add foo ; git commit --date "yesterday" -m "Update"

In my case, while using the --date option, my git process crashed. May be I did something terrible. And as a result some index.lock file appeared. So I manually deleted the .lock files from .git folder and executed, for all modified files to be commited in passed dates and it worked this time. Thanx for all the answers here.

git commit --date="`date --date='2 day ago'`" -am "update"

Or just use a fake-git-history to generate it for a specific data range.

1. git --date changes only GIT_AUTHOR_DATE but many git apps, e.g., GitHub shows GIT_COMMITTER_DATE. Make sure to change GIT_COMMITTER_DATE too.
2. git does not aware time zone in default UNIX date output. Make sure to format the datetime format in a compatible format such as ISO8601.

Complete example in OS X (Change both GIT_COMMITTER_DATE and GIT_AUTHOR_DATE to 4 hours ago):

x=$(date -v -4H +%Y-%m-%dT%H:%M:%S%z); export GIT_COMMITTER_DATE=$x; git commit --amend --date $x

You can always change a date on your computer, make a commit, then change the date back and push.

Pre-Step.

• Pull all data from the remote to the local repository.

• we are using the --amend and --date switches.
  

The exact command is as follows:

$ git commit --amend --date="YYYY-MM-DD HH:MM:SS"