373

I would like to use this workflow:

  1. Stage some changes.
  2. Save the unstaged changes to the stash.
  3. Do some stuff with the things in stage (build, test, etc.).
  4. Commit.
  5. Restore the unstaged changes.

Is there a way to do step 2?

Example:

git init
echo one >file
git add file
git commit
echo two >>file
git add file
echo three >>file
git stash push
test
git commit
git stash pop
Géry Ogam
  • 6,336
  • 4
  • 38
  • 67
Unapiedra
  • 15,037
  • 12
  • 64
  • 93
  • Why not commit your changes after staging them? – Shizzmo Oct 04 '11 at 16:08
  • 4
    IIRC --keepindex does exactly that – sehe Oct 04 '11 at 16:11
  • 5
    Because if, say, the build fails I don't want to have a commit of this. I know I can delete the commit but I'd like to do this without a commit if possible. – Unapiedra Oct 04 '11 at 16:12
  • Sehe, thanks. I can confirm this works. Gee, I looked at the manual at http://linux.die.net/man/1/git-stash which is out of date. `man git stash` is much better. – Unapiedra Oct 04 '11 at 16:17
  • it's --keep-index, fwiw. – jaf0 Jun 24 '14 at 21:34
  • this helped me learn that 'index' == staging area – JonnyRaa Apr 19 '21 at 16:26
  • There is no need to limit the stash data at `git push [-k|--keep-index]` time. Instead you can decide at apply time and use `git cherry-pick -m2 -n stash` to only pick the unstaged changes. See detailed answer. – kxr Sep 21 '22 at 12:23
  • Handy alias how to `stash only unstaged` is [here](https://stackoverflow.com/a/44824325/4632019). – Eugen Konkov Apr 07 '23 at 21:39

16 Answers16

439

git stash push has an option --keep-index that does exactly what you need, so run:

git stash push --keep-index
tdy
  • 36,675
  • 19
  • 86
  • 83
vhallac
  • 13,301
  • 3
  • 25
  • 36
  • 12
    True. I keep using `save` with `git stash`. Maybe it is the programmer in me insisting on honoring the symmetry with apply/pop. :) – vhallac Oct 04 '11 at 16:22
  • 178
    Note: this still stashes _all_ your changes; the only difference from regular `git stash save` is that it leaves the already-staged changes in your working copy as well. In the workflow above this would work fine since you're just applying the stash on top of a local copy that already has half of the stash's changes (which git is smart enough to ignore). But if you edit the code before re-applying the stash, you could potentially see merge conflicts when you go to apply. Fyi. – peterflynn Mar 04 '14 at 06:09
  • 4
    @ytpete That has bitten me so many times. I really wish there was a way for git to only stash the things you are not keeping... I often commit stuff, then do a full git stash, knowing that I can `git commit --ammend` if there are problems in what I committed. – rjmunro Apr 08 '14 at 10:11
  • 2
    `--amend` (rather than `--ammend`) – Rhubbarb Jan 21 '15 at 12:46
  • I used the commit, stash save, amend-commit-if-necessary, stash pop method. This is fine so long as you don't amend after a push (i.e. only push after you're happy with the final amend). I'm going to try the --keep-index option on the basis that it's easy to opt to take 'upstream' or 'stashed' changes upon merge conflict. However, a properly considered resolution is probably best. As a slight aside, I find resolving conflicts to be very much easier when 3-way conflict annotations are enabled: set the "merge.conflictstyle" configuration variable to "diff3". – Rhubbarb Jan 21 '15 at 12:54
  • I also prefer to explicitly specify `save` on a stash; that is helpful when you also want to provide a message string, which might otherwise be ambiguously interpreted by git. (Unlikely but possible.) – Rhubbarb Jan 21 '15 at 12:58
  • 33
    This solution does not work for me because of the problems described by peterflynn. It is not a good answer to the question since it still stashes the staged changes. Anybody got a better solution? – user643011 Mar 18 '17 at 14:29
  • 5
    Docs seem to say that `stash save` is deprecated now: "This option is deprecated in favour of `git stash push`. It differs from "stash push" in that it cannot take pathspecs, and any non-option arguments form the message." – jocull Aug 27 '19 at 20:14
  • 4
    It does not do what the question asked. It creates a stash that contains all modifications INCLUDING the index. What it does differently is that it does not 'resets' the index. But if you reset the index yourself and then 'stash pop', the index will be brought back. The question is about creating a stash that does NOT INCLUDE indexed files. – Alex Oct 24 '19 at 14:50
  • I'm a bit frustrated. I began a long cherry-pick operation last night which involves some necessary conflict management. I didn't finish, and today I collaborated with my colleagues on an unrelated section of code (since I still had the build available and running, why not). Now i've got a bunch of unstaged changes I need to save, I don't need to commit in the course of continuing a range cherry-pick, so I want to avoid that, but I cannot stash because it would slurp up the unrelated changes i've staged. Seems like I will need to commit and hope for the best. – Steven Lu Dec 11 '20 at 21:56
  • Follow up, i was able to get through the situation by using `git stash --patch` – Steven Lu Dec 11 '20 at 22:21
  • 2
    @peterflynn You can use `git diff` in this scenario. Use `git diff > ~/patch` to export the changes between the staging area and working area, `git checkout` to discard the changes in the working area, do what you need to do with the staging area, then `git apply ~/patch` to restore the saved changes. – whenov Mar 31 '21 at 03:29
  • @whenov The problem with your approach is that `git checkout` won’t discard anything in the working tree after `git reset`. – Géry Ogam Jan 30 '22 at 01:30
  • 1
    @user643011 see my answer here: https://stackoverflow.com/a/71150883/5469587 – citysurrounded Feb 17 '22 at 00:10
  • This did not work for me as I was in the middle of cherry-pick and needed to quickly stash some changes that shouldn't go into the commit. The command aborted the cherry-pick for some reason. – pooya13 Jan 24 '23 at 20:13
89

This may be done in 3 steps: save staged changes, stash everything else, restore index with staged changes. Which is basically:

git commit -m 'Save index'
git stash push -u -m 'Unstaged changes and untracked files'
git reset --soft HEAD^

This will do exactly what you want.

ma11hew28
  • 121,420
  • 116
  • 450
  • 651
Alesya Huzik
  • 1,520
  • 14
  • 17
  • 4
    Note: `-u` also stashes untracked files. – ma11hew28 Dec 17 '18 at 22:28
  • This approach essentially duplicates what `git stash save --keep-index` does with a lot more work. I don't see any advantages. – Inigo Feb 01 '20 at 22:55
  • 19
    @vas No, the approach does not duplicate that. See peterflynn's comment to the accepted answer. – Alexander Klauer Mar 10 '20 at 17:30
  • Excellent, but the purpose of stashing unstaged changes was to test staged changes *before* committing them. So there is no point in executing your third command `git reset --soft HEAD^`. – Géry Ogam Jan 30 '22 at 02:33
  • Pro tip: You can generally combine single-character flags together: `-u -m` can become `-um` e.g. `git stash push -um "Greatest stashing ever"` – Adi H Apr 21 '22 at 16:31
  • Yes, this did what the op asked for and what I was looking for. The edits made in 2018 help. – WeakPointer Sep 16 '22 at 10:59
36

From Git 2.35+ (Q1 2022) you can now use the --staged flag (man) on git stash push to only stage the changes in your index.

Since your question asks the exact opposite, we have 2 choices:

  1. Reverse the operation like so:
git stash push --staged            # Stash staged changes
git stash                          # Stash everything else
git stash pop stash@{1}            # Restore staged changes stash
  1. Stage the changes you want to stash instead of the ones you want to keep. Now you can just run:
git stash push --staged

I got this information from this answer on another S/O post.

pd93
  • 666
  • 1
  • 7
  • 15
  • 1
    Good point. I have updated [my ow answer on this page](https://stackoverflow.com/a/55799184/6309), in order to reference [my other answer](https://stackoverflow.com/a/70231955/6309). – VonC Feb 22 '22 at 14:01
  • 1
    Amazing! This is way more elegant than my [`git diff` solution](https://stackoverflow.com/a/70914962/2326961). – Géry Ogam Nov 15 '22 at 01:02
  • This was a great solution! My only change was I needed to stash new files as well, so I ran `git stash --include-untracked` instead of just `git stash` for step 2. – ADataGMan Jul 04 '23 at 12:32
36
git stash save --keep-index

Also, Re:

Why not commit your changes after staging them? – Shin

A: Because you should always checkin tested code :) That means, you need to run the tests with only the changes you are about to commit

All this apart from the fact that of course, as an experienced programmer, you have the innate urge to test and review just those changes -- only partly kidding

sehe
  • 374,641
  • 47
  • 450
  • 633
35

Stash Without the Staged Changes

The Problem with --keep-index / -k

Stashing just the working tree (unstaged changes) in Git is more difficult than it should be. The accepted answer, and quite a few other answers, stashes the unstaged changes and leaves the stage alone as requested via --keep-index.

However what isn't obvious is that --keep-index also stashes the staged changes. The staged changes end up in both the stage AND the stash. This is rarely what one wants because any interim changes to the stash are likely to result in conflicts when popping the stash later.

Alias Solution

This alias works well to stage just the working copy changes:

stash-working = "!f() { \
  git commit --quiet --no-verify -m \"temp for stash-working\" && \
  git stash push \"$@\" && \
  git reset --quiet --soft HEAD~1; }; f"

It commits the staged changes temporarily, creates a stash from the remaining changes (and allows additional arguments such as --include-untracked and --message to be passed as alias arguments), and then resets the temporary commit to get back the staged changes.

It is similar to @Simon Knapp's answer, but with a few minor differences -- it uses --quiet on the temporary actions taken, and it accepts any number of parameters for the stash push, rather than hard-coding the -m, and it does add --soft to the final reset so that the index remains as it started. It also uses --no-verify on the commit to avoid changes to the working copy from pre-commit hooks (HT: @Granfalloner).

For the opposite problem of stashing just the staged changes (alias stash-index) see this answer.

Raman
  • 17,606
  • 5
  • 95
  • 112
  • 4
    As a further improvement to this snippet, its worth to add [`--no-verify`](https://git-scm.com/docs/githooks) option to `git commit`, otherwise implicit temporary commit might seriously mess up working directory because of pre-commit hook. – Granfalloner Dec 16 '20 at 17:52
  • I had more luck using it as a function, not sure why but zsh choked on the above syntax: ```stash-working() { git commit --quiet --no-verify -m "temp for stash-working" && git stash push "$@" && git reset --quiet --soft HEAD~1; }``` – Lotus Jun 08 '22 at 00:49
  • 2
    @Lotus Try copying it exactly as-is directly into the `[alias]` section of your `.gitconfig`. That way, `zsh` is not involved at all. – Raman Jun 08 '22 at 13:10
26

TL;DR; Since at git 2.35 flag was implemented: git stash [push [-S|--staged]], you can do:

git stashu

after adding this alias:

git config --global alias.stashu '!git stash push -S; git stash; git stash pop --index stash@{1}'"

Explanation:

git stash push -S                  # Stash staged changes
git stash                          # Stash everything else
git stash pop --index stash@{1}    # Restore staged changes into index

$ git config --global alias.stashu '!git stash push -S; git stash; git stash pop --index stash@{1}'

$ git diff 
```diff
diff --git a/src/js/modal.js b/src/js/modal.js
index d07c085..766e39a 100644
--- a/src/js/modal.js
+++ b/src/js/modal.js
@@ -6,10 +6,12 @@
 import "jquery-validation/dist/jquery.validate";
 import "bootstrap/dist/js/bootstrap.bundle";
 
+staged
 const FormDataJson = require('form-data-json-convert');
 FormDataJson.defaultOptionsToJson.uncheckedValue =  false;
 FormDataJson.defaultOptionsToJson.skipEmpty      =  true;
 
+unstaged
 import { TabulatorFull as Tabulator } from 'tabulator-tables/dist/js/tabulator';
 
 import { ajax_query } from "./common/ajax";
``` 

$ git add -p
```diff
@@ -6,7 +6,8 @@
 import "jquery-validation/dist/jquery.validate";
 import "bootstrap/dist/js/bootstrap.bundle";
 
+staged
 const FormDataJson = require('form-data-json-convert');
 FormDataJson.defaultOptionsToJson.uncheckedValue =  false;
 FormDataJson.defaultOptionsToJson.skipEmpty      =  true;
``` 

(1/2) Stage this hunk [y,n,q,a,d,j,J,g,/,e,?]? y
```diff
@@ -9,7 +10,8 @@
 const FormDataJson = require('form-data-json-convert');
 FormDataJson.defaultOptionsToJson.uncheckedValue =  false;
 FormDataJson.defaultOptionsToJson.skipEmpty      =  true;
 
+unstaged
 import { TabulatorFull as Tabulator } from 'tabulator-tables/dist/js/tabulator';
 
 import { ajax_query } from "./common/ajax";
```
(2/2) Stage this hunk [y,n,q,a,d,K,g,/,e,?]? n

$ git status
On branch dev-Modal
Changes to be committed:
  (use "git restore --staged <file>..." to unstage)
modified:   modal.js

Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git restore <file>..." to discard changes in working directory)
modified:   modal.js

$ git diff
```diff
diff --git a/src/js/modal.js b/src/js/modal.js
index ac48408..766e39a 100644
--- a/src/js/modal.js
+++ b/src/js/modal.js
@@ -11,6 +11,7 @@ const FormDataJson = require('form-data-json-convert');
 FormDataJson.defaultOptionsToJson.uncheckedValue =  false;
 FormDataJson.defaultOptionsToJson.skipEmpty      =  true;
 
+unstaged
 import { TabulatorFull as Tabulator } from 'tabulator-tables/dist/js/tabulator';
 
 import { ajax_query } from "./common/ajax";
``` 

$ git stashu 
Saved working directory and index state WIP on dev-Modal: 9f6e760 Imported Tabulator module
Saved working directory and index state WIP on dev-Modal: 9f6e760 Imported Tabulator module
On branch dev-Modal
Changes to be committed:
  (use "git restore --staged <file>..." to unstage)
modified:   src/js/modal.js

Dropped stash@{1} (509be8cf1afa340faf77aa6a2b8d008aa82a980a)

$ git stash show -p
```diff
diff --git a/src/js/modal.js b/src/js/modal.js
index d07c085..ed509ed 100644
--- a/src/js/modal.js
+++ b/src/js/modal.js
@@ -10,6 +10,7 @@ const FormDataJson = require('form-data-json-convert');
 FormDataJson.defaultOptionsToJson.uncheckedValue =  false;
 FormDataJson.defaultOptionsToJson.skipEmpty      =  true;
 
+unstaged
 import { TabulatorFull as Tabulator } from 'tabulator-tables/dist/js/tabulator';
 
 import { ajax_query } from "./common/ajax";
```

**BONUS for those who read till here. The alias to stash staged changes:
```
$ git config --global alias.stashs 'stash push -S'
```
And now you can `stash staged`:
```
git stashs
```

With git version 2.7.4 you may use the --patch option:

git stash save --patch

Then git will ask you to add or not your changes into stash.
And you then just answer y or n.

You can restore the working directory as you always do like:

git stash pop

or, if you want to keep saved changes in stash:

git stash apply
Eugen Konkov
  • 22,193
  • 17
  • 108
  • 158
23

To add the unstagged (not added to commit) files to stash, run the following command:

git stash -k

If you want to include newly added files(which are not staged - not in green) also to the stash, do the following:

git stash -k -u

Then you can commit the staged files. After that you can get back the last stashed files using the command:

git stash pop
Eugen Konkov
  • 22,193
  • 17
  • 108
  • 158
srth12
  • 873
  • 9
  • 16
  • 1
    I think this should be accepted answer because the one accepted does not work if you have some edits that are staged and unstaged on the same file, you get `patch does not apply`. – iRestMyCaseYourHonor May 20 '22 at 13:25
  • If you modify a file after staging it, you will lose these changes when doing `git stash -k -u` followed by `git stash pop` - See https://stackoverflow.com/questions/73128025/how-can-i-create-a-temporary-clean-working-tree-for-running-unit-tests-in-a-gi – Tobias Bergkvist Jul 26 '22 at 18:52
  • This does not work because it will still stash everything (BOTH staged and unstaged files). `-k` or '--keep-index' does not prevent staged files from being added to the stash; it is merely for the convenience of leaving the stash intact after the operation (instead of removing it from the stash) – Adi H Jul 06 '23 at 01:28
7

Git doesn't have a command that stashes only your unstaged changes.

Git does, however, let you specify which files you want to stash.

git stash push --message 'Unstaged changes' -- app/controllers/products_controller.rb test/controllers/products_controller_test.rb

If you only want to stash specific changes in those files, add the --patch option.

git stash push --patch --message 'Unstaged changes' -- app/controllers/products_controller.rb test/controllers/products_controller_test.rb

The --include-untracked option lets you stash untracked files.

git stash push --include-untracked --message 'Untracked files' -- app/controllers/widgets_controller.rb test/controllers/widgets_controller_test.rb

Run git help stash (or man git-stash) for more info.

Note: If your unstaged changes are rather disoganized, @alesguzik's answer is probably easier.

ma11hew28
  • 121,420
  • 116
  • 450
  • 651
5

Extending previous answers, I sometimes have a complex set of changes staged, but wish to commit a separate change first. For example, I might have spotted a bug or otherwise incorrect code that I'd like to fix ahead of my staged changes. One possible route to take is this:

first stash everything, but leave the staged changes intact

$ git stash save --keep-index [--include-untracked]

now stash the staged changes separately too

$ git stash save

make changes for fix; and test; commit them:

$ git add [--interactive] [--patch]

$ git commit -m"fix..."

now restore the previously staged changes:

$ git stash pop

resolve any conflicts, and note that if there were conflicts, git will have applied but not dropped that top stash entry.

(... Then commit the staged changes, and restore the stash of all the other changes, and continue ...)

Community
  • 1
  • 1
Rhubbarb
  • 4,248
  • 6
  • 36
  • 40
3

Another tip, related to the question:

When you effectively stash your unstaged changes using

$ git stash save --keep-index

you might wish to give the stash a message, so that when you to do a git stash list it's more obvious what you have stashed before, especially if you follow that stash operation by further saves. For example

$ git stash save --keep-index "changes not yet staged"

(although actually it does contain all the changes as noted in other answers).

For example, the above might be followed immediately by:

$ git stash save "staged changes for feature X"

Beware, though, that you can't then use

$ git stash apply "stash@{1}" ### ✘ doesn't quite do what you might want

to restore just the unstaged changes.

Rhubbarb
  • 4,248
  • 6
  • 36
  • 40
3

I use an an alias, which accepts a string to use as a message to the stash entry.

mystash = "!f() { git commit -m hold && git stash push -m \"$1\" && git reset HEAD^; }; f"

Which:

  • commits everything in the index,
  • stashes what is changed in the working tree (could of course add -u or -a),
  • resets the last commit back to the working try (may want to use --soft to keep it in the index).
Simon Knapp
  • 305
  • 2
  • 8
  • thanks I had some success with it with the `--soft` flag, except I added 2 twists: `git commit --allow-empty` in case nothing is staged, and a default value for the stash message `$1`. I had to add `#` to the end as well to avoid message being repeated. which gives me: `git config --global alias.save '!git commit --no-verify --allow-empty -m hold && git stash push -um "${1-unstaged}" && git reset --soft HEAD^ #'` – Simon Baslé Nov 17 '22 at 17:21
2

2022: I mention in "Stashing only staged changes in git - is it possible?", Git 2.35 (Q1 2022) comes with "git stash push --staged"(man):

This option is only valid for push and save commands.

Stash only the changes that are currently staged.
This is similar to basic git commit except the state is committed to the stash instead of current branch.


2019: The modern form of that command is git stash push [--] [<pathspec>...], since Git 2.16+ (git stash save is deprecated)

You can combine that with a wildcard form, for example:

git stash push --all --keep-index ':(glob)**/*.testextension' 

But that does not work well with Git for Windows, until Git 2.22 (Q2 2019), see issue 2037, considering git stash has been re-implemented in C (instead of a shell script)

See commit 7db9302 (11 Mar 2019) by Thomas Gummerer (tgummerer).
See commit 1366c78, commit 7b556aa (07 Mar 2019) by Johannes Schindelin (dscho).
(Merged by Junio C Hamano -- gitster -- in commit 0ba1ba4, 22 Apr 2019)

built-in stash: handle :(glob) pathspecs again

When passing a list of pathspecs to, say, git add, we need to be careful to use the original form, not the parsed form of the pathspecs.

This makes a difference e.g. when calling

git stash -- ':(glob)**/*.txt'

where the original form includes the :(glob) prefix while the parsed form does not.

However, in the built-in git stash, we passed the parsed (i.e. incorrect) form, and git add would fail with the error message:

fatal: pathspec '**/*.txt' did not match any files

at the stage where git stash drops the changes from the worktree, even if refs/stash has been actually updated successfully.

VonC
  • 1,262,500
  • 529
  • 4,410
  • 5,250
1

Here's (in my opinion) the best solution, that does exactly what the OP has asked for. It stashes only the unstaged, tracked files – without an unnecessary commit or stashing all changed files with --keep-index

It lists all the unstaged, tracked changes (git diff --name-only) converts newlines to spaces (| tr '\n' ' ') and stashes all these files using git stash push:

git stash push $(git diff --name-only | tr '\n' ' ')
citysurrounded
  • 664
  • 4
  • 8
0

I was interested in how the Python program pre-commit does it. Here is the code. https://github.com/pre-commit/pre-commit/blob/3fe38dff05957f609cf7b97f471b35a8d9e0659a/pre_commit/staged_files_only.py#L50

It is functionally equivalent to something like:

git diff-index --ignore-submodules --binary --exit-code --no-color --no-ext-diff $(git write-tree) -- >stash.patch
git checkout -- .

# Do stuff now

git apply stash.patch && rm stash.patch
Noé Rubinstein
  • 786
  • 6
  • 17
0

Rethink: It is not necessary to limit the stash data to the working tree changes only, but it can be decided later at apply time to only apply the working tree changes of the stash.

Thus at stash time just do as usual:

git stash [-k|--keep-index]

At apply time do

git cherry-pick -m2 -n stash

Explanation: The -m2 selects the changes to the 2nd parent of the stage commit, which is the stored index state. -n|--no-commit blocks auto-commit. stash@{1} would be the ref for the 2nd stash on the stack ...

kxr
  • 4,841
  • 1
  • 49
  • 32
-1

To my knowledge, it is currently impossible to save only unstaged changes in the working tree with git stash push, i.e. to save changes from the index state. This command saves all changes in the working tree (staged and unstaged changes), i.e. changes from the HEAD state, even with the option --keep-index which also sets the working tree state to the index state instead of the HEAD state (thereby creating conflicts when restoring the changes from the HEAD state with git stash pop). It would be very convenient if git stash push had an option -U|--unstaged for saving only unstaged changes (to me the option --keep-index is flawed), since it has already an option -S|--staged for saving only staged changes.

So for the moment you have to emulate

git stash push --unstaged

git stash pop

with a temporary file:

git diff >unstaged
git restore .

git apply unstaged
rm unstaged

Your use case is testing before committing partial changes and it is already in the reference documentation, but with the flawed option --keep-index which creates conflicts. Here is the version with the emulated option -U|--unstaged:

git init
echo one >file
git add file
git commit
echo two >>file
git add file
echo three >>file
git diff >unstaged
git restore .
test
git commit
git apply unstaged
rm unstaged

Visualising states

For a better understanding of stashing, I think it is important to look at the states of the working tree, index and HEAD at each step. Let’s take your use case.

git init

working index HEAD

echo one >file

working index HEAD
one

git add file

working index HEAD
one one

git commit

working index HEAD
one one one

echo two >>file

working index HEAD
one one one
two

git add file

working index HEAD
one one one
two two

echo three >>file

working index HEAD
one one one
two two
three

git diff >unstaged

git restore .

working index HEAD
one one one
two two

test

git commit

working index HEAD
one one one
two two two

git apply unstaged

rm unstaged

working index HEAD
one one one
two two two
three
Géry Ogam
  • 6,336
  • 4
  • 38
  • 67