ChatGPT解决这个技术问题 Extra ChatGPT

How do I add an empty directory to a Git repository?

How do I add an empty directory (that contains no files) to a Git repository?

While it's not useful, there is a way to hack an empty (really empty) directory into your repo. It won't checkout with current versions of Git, however.
@tiwo I for one disagree that it's not useful. Your directory hierarchy is part of your project, so it should be version controlled.
In my case, I'd like to add a directory structure for tmp files, but not the tmp files themselves. By doing this, my tester has the correct structure (otherwise there are errors) but I don't clog my commits with tmp data. So yes, it's useful to me!
@AdamMarshall I think tiwo was saying that the hack is not useful, since it is ignored by checkout. Tmp dirs do sound like a useful feature for a VCS.
Why not have the procedure that creates the tmp files also create the tmp directory?

H
HelloGoodbye

Another way to make a directory stay (almost) empty (in the repository) is to create a .gitignore file inside that directory that contains these four lines:

# Ignore everything in this directory
*
# Except this file
!.gitignore

Then you don't have to get the order right the way that you have to do in m104's solution.

This also gives the benefit that files in that directory won't show up as "untracked" when you do a git status.

Making @GreenAsJade's comment persistent:

I think it's worth noting that this solution does precisely what the question asked for, but is not perhaps what many people looking at this question will have been looking for. This solution guarantees that the directory remains empty. It says "I truly never want files checked in here". As opposed to "I don't have any files to check in here, yet, but I need the directory here, files may be coming later".


I think the README solution proposed by @JohnMee should be used together with this one; the .gitignore file provides an explanation of what we want to keep out of version control, while the README file explains what is the purpose of the directory, which are both very important pieces of information.
@pedromanoel I write the documentation you would put in the README inside the .gitignore file (as comments).
spot the 1 difference: 1.) an empty folder, 2.) a folder with .gitignore file in it. ;-)
This is perfect for cache folders.
Unfortunately, this results in a non-empty directory, it has a single hidden file.
G
Gilles 'SO- stop being evil'

You can't. See the Git FAQ.

Currently the design of the git index (staging area) only permits files to be listed, and nobody competent enough to make the change to allow empty directories has cared enough about this situation to remedy it. Directories are added automatically when adding files inside them. That is, directories never have to be added to the repository, and are not tracked on their own. You can say "git add

" and it will add files in there. If you really need a directory to exist in checkouts you should create a file in it. .gitignore works well for this purpose; you can leave it empty, or fill in the names of files you expect to show up in the directory.


Below answer is MUCH better. The fact that git the low level software doesn't allow it doesn't matter to me as much as HOW to actually use Git when I need an empty directory. Adding a 2 line .gitignore seems acceptable to me.
You can read "it's impossible, you can't, etc." all over the Internet for this frequent question. The .gitignore trick is a frequent answer, and satisfies many needs. However it IS possible to make git track an truly empty directory, see my answer
Though the more I think of it, the more it feels like "SHA hash of the empty string", if it exists, actually would be a well-defined identifier for an empty tree, unless it would be impossible to tell whether that object is a tree or a blob.
I've seen a lot of repos that use an empty file called .gitkeep for this purpose.
@Amala: what are you referring to, when you say "below". The order of answers is changing...
M
Mark Amery

Create an empty file called .gitkeep in the directory, and add that.


.gitkeep has not been prescribed by Git and is going to make people second guess its meaning, which will lead them to google searches, which will lead them here. The .git prefix convention should be reserved for files and directories that Git itself uses.
@t-mart "The .git prefix convention should be reserved..." Why? Does git request this reservation?
In this case a README or ABOUT file would be just as good or better. Leaving a note for the next guy, just like we all used to do it before URLs.
Doesn't work if you're writing a unit test that should test code on an empty directory...
@szablica I don't think it's confusing at all. In fact I think it's very intuitive to call it .gitkeep. Calling it .gitignore is what sounds contradictory to me. So this is just a matter of personal taste.
W
Will

You could always put a README file in the directory with an explanation of why you want this, otherwise empty, directory in the repository.


+1, Good suggestion, an empty directory does not make any sense unless it is going to be used in the future. So create a README file inside it and write what this directory is for, and what files will be put there in the future. That solves both two problems.
@ilius Nonsense. A directory structure containing empty directories may be highly desirable in many situations (like an MVC app where you want a models directory but haven't gotten around to creating any models yet, or a shared views directory you plan to add shared views to, later). Moreover, putting a README in each one of these is overkill as it's obvious what they're there for, and it's easy to forget to put a README in each one of them. AND you have to remember to remove the README when you add some other files to them. Basically, git should definitely allow empty directories.
@Jez: I disagree. The point is that git is designed to control (and index) source-code. Importantly, the id of a commit is a hash of the contents. That is to say, it must have contents. You don't need a README in every part of the tree, only leaf nodes. If you have places you intend to put code, but no code, and you won't even take the time to echo "place for models" >> README, then what you have is an idea not a commit. It is not of interest to git. Saying "I want the running app to have XYZ empty directories" is a runtime problem, not a source problem. Handle it w/ your installer.
@JoeAtzberger It's a missing feature, not an intentional limitation. From the Git FAQ: Currently the design of the Git index (staging area) only permits files to be listed, and nobody competent enough to make the change to allow empty directories has cared enough about this situation to remedy it.
@jbo5112 Yes, the "special code" you refer to is the "installer" I mentioned. Your webapp installation already has to handle creating a database, local config, pulling dependencies or 100 other operations, but a couple empty directories are beyond it? Try gradle, passenger, chef, a primitive Makefile, etc. There is no security difference between creating directories and the other (potentially far more complicated/dangerous) work of installing an app. And if you really have no deps, config, DB, etc., and no installer, then just use the README. No case requires you to do both.
A
Asclepius
touch .placeholder

On Linux, this creates an empty file named .placeholder. For what it's worth, this name is agnostic to git, and this approach is used in various other places in the system, e.g. /etc/cron.d/.placeholder. Secondly, as another user has noted, the .git prefix convention can be reserved for files and directories that Git itself uses for configuration purposes.

Alternatively, as noted in another answer, the directory can contain a descriptive README.md file instead.

Either way this requires that the presence of the file won't cause your application to break.


This is good for an initial bare directory, but what if it starts to fill with files? Then Git will notice them and claim them as untracked files. The selected answer here works far more elegantly to allow one to keep a directory but then safely ignore the contents.
The question and the predominant general concern is about adding an empty directory. If it later has a resident file, obviously delete the .keep file or just disregard it. If instead the files in the directory are to be ignored, that's a different question altogether.
It was suggested that git clean -nd | sed s/'^Would remove '// | xargs -I{} touch "{}.keep" will do this in all untracked empty directories.
Don't like this solution, it is tough to guess what this file does. Also, if you are generating files in your dev environment (like logs or images, etc.), this isn`t keeping those file from being versioned and making their way into production, which is not nice.
Elegant: the .keep file together with the commit message shows the intent of "keeping" the project structure. Adding Readme or Abouts I think will cause more confusion...
M
Manu Manjunath

Why would we need empty versioned folders

First things first:

An empty directory cannot be part of a tree under the Git versioning system.

It simply won't be tracked. But there are scenarios in which "versioning" empty directories can be meaningful, for example:

scaffolding a predefined folder structure, making it available to every user/contributor of the repository; or, as a specialized case of the above, creating a folder for temporary files, such as a cache/ or logs/ directories, where we want to provide the folder but .gitignore its contents

related to the above, some projects won't work without some folders (which is often a hint of a poorly designed project, but it's a frequent real-world scenario and maybe there could be, say, permission problems to be addressed).

Some suggested workarounds

Many users suggest:

Placing a README file or another file with some content in order to make the directory non-empty, or Creating a .gitignore file with a sort of "reverse logic" (i.e. to include all the files) which, at the end, serves the same purpose of approach #1.

While both solutions surely work I find them inconsistent with a meaningful approach to Git versioning.

Why are you supposed to put bogus files or READMEs that maybe you don't really want in your project?

Why use .gitignore to do a thing (keeping files) that is the very opposite of what it's meant for (excluding files), even though it is possible?

.gitkeep approach

Use an empty file called .gitkeep in order to force the presence of the folder in the versioning system.

Although it may seem not such a big difference:

You use a file that has the single purpose of keeping the folder. You don't put there any info you don't want to put. For instance, you should use READMEs as, well, READMEs with useful information, not as an excuse to keep the folder. Separation of concerns is always a good thing, and you can still add a .gitignore to ignore unwanted files.

Naming it .gitkeep makes it very clear and straightforward from the filename itself (and also to other developers, which is good for a shared project and one of the core purposes of a Git repository) that this file is A file unrelated to the code (because of the leading dot and the name) A file clearly related to Git Its purpose (keep) is clearly stated and consistent and semantically opposed in its meaning to ignore

A file unrelated to the code (because of the leading dot and the name)

A file clearly related to Git

Its purpose (keep) is clearly stated and consistent and semantically opposed in its meaning to ignore

Adoption

I've seen the .gitkeep approach adopted by very important frameworks like Laravel, Angular-CLI.


You missed one thought - whats the reason for keeping and empty folder (e.g. /logs, /tmp, /uploads)? Yes - its to keep the folder empty. :) So if you want to keep a folder empty, you have to ignore the files inside it.
@RomanAllenstein: not necessarily. It could be that you create a repo with a given structure which can become populated later. Those files will be added to the repo as soon as they are created, and it will be annoying to start deleting or editing .gitignore files (and dangerous, because probably you do not even realize that they are not being tracked: git is ignoring them)
If you edit your answer to replace .gitkeep with any other non git-prefixed file name you get my upvote, I think this one is the best and most informative answer. Reason: I think ".git*" should be reserved for git prescribed files, while this is just a mere placeholder. My first guess when I saw that is that for example a ".gitkeep" file would be auto-ignored (that would be a nice feature) but that is not the case, right?
@Santosh You could edit my post and be useful to the community instead of childishly bragging against a non-native speaker and uselessly polluting the comments, which is [IN]consistent with average intelligent behaviour. That's why edits are for, btw. Thanks for the free lesson anyway, duly appreciated :)
I wonder why people have such a hard time to understand why one wants to add "empty" folders to git. You have to start somewhere, right? So, usually you start with your projects folder structure and - alas - at the start of the project there is nothing there yet. Once your project repo is done, team workers can clone and start working on the SAME structure.
P
Peter Mortensen

As described in other answers, Git is unable to represent empty directories in its staging area. (See the Git FAQ.) However, if, for your purposes, a directory is empty enough if it contains a .gitignore file only, then you can create .gitignore files in empty directories only via:

find . -type d -empty -exec touch {}/.gitignore \;

You may want to ignore the .git directory: find . -name .git -prune -o -type d -empty -exec touch {}/.gitignore \;
A simpler variation for most situations is find * -type d -empty -exec touch {}/.gitignore \;
Since OS X creates a .DS_Store file in almost every directoy, this does not work there. The only (DANGEROUS!) workaround i found, was to delete all the .DS_Store files first via find . -name .DS_Store -exec rm {} \; and then use the preferred variant from this answer. Be sure to only execute this in the correct folder!
Does anyone know a way to do this in Windows from the command line? I've seen some solutions here in Ruby and Python, but I'd like a barebones solution if it can be managed.
@akhan Adding something to .gitignore has no influence on the -empty flag of the find command. My comment is about removing the .DS_Store files in a directory tree, so the -empty flag can be applied.
P
Peter Mortensen

Andy Lester is right, but if your directory just needs to be empty, and not empty empty, you can put an empty .gitignore file in there as a workaround.

As an aside, this is an implementation issue, not a fundamental Git storage design problem. As has been mentioned many times on the Git mailing list, the reason that this has not been implemented is that no one has cared enough to submit a patch for it, not that it couldn’t or shouldn’t be done.


r
rogerdpack

The Ruby on Rails log folder creation way:

mkdir log && touch log/.gitkeep && git add log/.gitkeep

Now the log directory will be included in the tree. It is super-useful when deploying, so you won't have to write a routine to make log directories.

The logfiles can be kept out by issuing,

echo log/dev.log >> .gitignore

but you probably knew that.


What does that have to do with Ruby on Rails?
L
Lesmana

Git does not track empty directories. See the Git FAQ for more explanation. The suggested workaround is to put a .gitignore file in the empty directory. I do not like that solution, because the .gitignore is "hidden" by Unix convention. Also there is no explanation why the directories are empty.

I suggest to put a README file in the empty directory explaining why the directory is empty and why it needs to be tracked in Git. With the README file in place, as far as Git is concerned, the directory is no longer empty.

The real question is why do you need the empty directory in git? Usually you have some sort of build script that can create the empty directory before compiling/running. If not then make one. That is a far better solution than putting empty directories in git.

So you have some reason why you need an empty directory in git. Put that reason in the README file. That way other developers (and future you) know why the empty directory needs to be there. You will also know that you can remove the empty directory when the problem requiring the empty directory has been solved.

To list every empty directory use the following command:

find -name .git -prune -o -type d -empty -print

To create placeholder READMEs in every empty directory:

find -name .git -prune -o -type d -empty -exec sh -c \
  "echo this directory needs to be empty because reasons > {}/README.emptydir" \;

To ignore everything in the directory except the README file put the following lines in your .gitignore:

path/to/emptydir/*
!path/to/emptydir/README.emptydir
path/to/otheremptydir/*
!path/to/otheremptydir/README.emptydir

Alternatively, you could just exclude every README file from being ignored:

path/to/emptydir/*
path/to/otheremptydir/*
!README.emptydir

To list every README after they are already created:

find -name README.emptydir

P
Peter Mortensen

WARNING: This tweak is not truly working as it turns out. Sorry for the inconvenience.

Original post below:

I found a solution while playing with Git internals!

Suppose you are in your repository. Create your empty directory: $ mkdir path/to/empty-folder Add it to the index using a plumbing command and the empty tree SHA-1: $ git update-index --index-info 040000 tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904 path/to/empty-folder Type the command and then enter the second line. Press Enter and then Ctrl + D to terminate your input. Note: the format is mode [SPACE] type [SPACE] SHA-1hash [TAB] path (the tab is important, the answer formatting does not preserve it). That's it! Your empty folder is in your index. All you have to do is commit.

This solution is short and apparently works fine (see the EDIT!), but it is not that easy to remember...

The empty tree SHA-1 can be found by creating a new empty Git repository, cd into it and issue git write-tree, which outputs the empty tree SHA-1.

EDIT:

I've been using this solution since I found it. It appears to work exactly the same way as creating a submodule, except that no module is defined anywhere. This leads to errors when issuing git submodule init|update. The problem is that git update-index rewrites the 040000 tree part into 160000 commit.

Moreover, any file placed under that path won't ever be noticed by Git, as it thinks they belong to some other repository. This is nasty as it can easily be overlooked!

However, if you don't already (and won't) use any Git submodules in your repository, and the "empty" folder will remain empty or if you want Git to know of its existence and ignore its content, you can go with this tweak. Going the usual way with submodules takes more steps that this tweak.


After putting the empty folder into the index and committing, is it then possible to git svn dcommit it with the desired result?
It's unlikely that this tweak will work with any other tool. Like stated in the warning and the edit, I discourage using it unless in a quite restricted case.
I've created a better solution based on this that doesn't have these drawbacks: stackoverflow.com/a/58543445/277882
G
GAMITG

Let's say you need an empty directory named tmp :

$ mkdir tmp
$ touch tmp/.gitignore
$ git add tmp
$ echo '*' > tmp/.gitignore
$ git commit -m 'Empty directory' tmp

In other words, you need to add the .gitignore file to the index before you can tell Git to ignore it (and everything else in the empty directory).


Two things: You could just "echo '*' > tmp/.gitignore" instead of touching, and "git commit -m" does not commit changes done after you've added the files to the index.
If you just do echo bla > file you will not get file: File exists because > will overwrite the file if it's already there or create a new one if it doesn't exist.
/bin/sh cultural assumption!* If "here" is csh and the variable noclobber is set, you will indeed get file: File exists. If someone says "I get this", don't assume they're an idiot and reply "No you don't". * c2.com/cgi/wiki?AmericanCulturalAssumption
@clacke If someone decides to use a different shell than everyone else, they should state that expressly if they are encountering problems. Unlike with nationality, everyone has their free choice of shell.
@SeldomNeedy Maybe they are looking for help because they don't even know they are using a different shell than everybody else.
B
Brent Bradburn

Maybe adding an empty directory seems like it would be the path of least resistance because you have scripts that expect that directory to exist (maybe because it is a target for generated binaries). Another approach would be to modify your scripts to create the directory as needed.

mkdir --parents .generated/bin ## create a folder for storing generated binaries
mv myprogram1 myprogram2 .generated/bin ## populate the directory as needed

In this example, you might check in a (broken) symbolic link to the directory so that you can access it without the ".generated" prefix (but this is optional).

ln -sf .generated/bin bin
git add bin

When you want to clean up your source tree you can just:

rm -rf .generated ## this should be in a "clean" script or in a makefile

If you take the oft-suggested approach of checking in an almost-empty folder, you have the minor complexity of deleting the contents without also deleting the ".gitignore" file.

You can ignore all of your generated files by adding the following to your root .gitignore:

.generated

Note: The symbolic link that I suggested is "broken" in a clean checkout because the .generated directory does not initially exist. It will no longer be broken once you do your build.
I agree in some cases this is a very good idea, but in others (such as distributing a project where you have an otherwise empty skeleton with folders such as models/ and views/ ) you would want the user to have these directories at hand rather than manually having to read read the docs, and it could be a bit much to expect them to run some sort of installation script after cloning the repo. I think this answer in combination with @john-mee's README answer should cover most if not all cases.
v
vidur punj

Add a .gitkeep file inside that directory and commit it.

touch .gitkeep 

it is the standard followed by git.


M
Mig82

I like the answers by @Artur79 and @mjs so I've been using a combination of both and made it a standard for our projects.

find . -type d -empty -exec touch {}/.gitkeep \;

However, only a handful of our developers work on Mac or Linux. A lot work on Windows and I could not find an equivalent simple one-liner to accomplish the same there. Some were lucky enough to have Cygwin installed for other reasons, but prescribing Cygwin just for this seemed overkill.

Edit for a better solution

So, since most of our developers already have Ant installed, the first thing I thought of was to put together an Ant build file to accomplish this independently of the platform. This can still be found here

However, I later thought It would be better to make this into a small utility command, so I recreated it using Python and published it to the PyPI here. You can install it by simply running:

pip3 install gitkeep2

It will allow you to create and remove .gitkeep files recursively, and it will also allow you to add messages to them for your peers to understand why those directories are important. This last bit is bonus. I thought it would be nice if the .gitkeep files could be self-documenting.

$ gitkeep --help
Usage: gitkeep [OPTIONS] PATH

  Add a .gitkeep file to a directory in order to push them into a Git repo
  even if they're empty.

  Read more about why this is necessary at: https://git.wiki.kernel.org/inde
  x.php/Git_FAQ#Can_I_add_empty_directories.3F

Options:
  -r, --recursive     Add or remove the .gitkeep files recursively for all
                      sub-directories in the specified path.
  -l, --let-go        Remove the .gitkeep files from the specified path.
  -e, --empty         Create empty .gitkeep files. This will ignore any
                      message provided
  -m, --message TEXT  A message to be included in the .gitkeep file, ideally
                      used to explain why it's important to push the specified
                      directory to source control even if it's empty.
  -v, --verbose       Print out everything.
  --help              Show this message and exit.

I hope you find it useful.


for what it's worth, a windows powershell one-liner could be ls -r -dir | ?{$_.getFileSystemInfos().Count -eq 0} | %{ni -p $_.FullName -n .gitkeep}
G
GAMITG

You can't and unfortunately will never be able to. This is a decision made by Linus Torvald himself. He knows what's good for us.

There is a rant out there somewhere I read once.

I found Re: Empty directories.., but maybe there is another one.

You have to live with the workarounds...unfortunately.


I know you posted this as an example of a bad argument, but I appreciate the link because it's actually a well-reasoned argument against tracking directories. ;-)
This answer seems to be inconsistent, since in the next post on the referenced thread, Linus Torvald says he expects that they will need to add directory tracking: markmail.org/message/libip4vpvvxhyqbl . In fact, he says he "would welcome patches that [add support for tracking empty directories]"
Patrick, he also uses the word "idiotic" there. I suspect his wording adresses the people here in this thread and so I assume he will not implement something "idiotic" into Git by himself.
佚名

I've been facing the issue with empty directories, too. The problem with using placeholder files is that you need to create them, and delete them, if they are not necessary anymore (because later on there were added sub-directories or files. With big source trees managing these placeholder files can be cumbersome and error prone.

This is why I decided to write an open source tool which can manage the creation/deletion of such placeholder files automatically. It is written for .NET platform and runs under Mono (.NET for Linux) and Windows.

Just have a look at: http://code.google.com/p/markemptydirs


D
DevonDahon

This solution worked for me.

1. Add a .gitignore file to your empty directory:

*
*/
!.gitignore

* ignore all files in the folder

*/ Ignore subdirectories

!.gitignore include the .gitignore file

2. Then remove your cache, stage your files, commit and push:

git rm -r --cached .
git add . // or git stage .
git commit -m ".gitignore fix"
git push

This is great. I used to do the .gitkeep file and a similar few lines in the main .gitignore, but I prefer this. Just seems more git-y. Thanks.
G
GAMITG

When you add a .gitignore file, if you are going to put any amount of content in it (that you want Git to ignore) you might want to add a single line with just an asterisk * to make sure you don't add the ignored content accidentally.


n
ntninja

Reading @ofavre's and @stanislav-bashkyrtsev's answers using broken GIT submodule references to create the GIT directories, I'm surprised that nobody has suggested yet this simple amendment of the idea to make the whole thing sane and safe:

Rather than hacking a fake submodule into GIT, just add an empty real one.

Enter: https://gitlab.com/empty-repo/empty.git

A GIT repository with exactly one commit:

commit e84d7b81f0033399e325b8037ed2b801a5c994e0
Author: Nobody <none>
Date: Thu Jan 1 00:00:00 1970 +0000

No message, no committed files.

Usage

To add an empty directory to you GIT repo:

git submodule add https://gitlab.com/empty-repo/empty.git path/to/dir

To convert all existing empty directories to submodules:

find . -type d -empty -delete -exec git submodule add -f https://gitlab.com/empty-repo/empty.git \{\} \;

Git will store the latest commit hash when creating the submodule reference, so you don't have to worry about me (or GitLab) using this to inject malicious files. Unfortunately I have not found any way to force which commit ID is used during checkout, so you'll have to manually check that the reference commit ID is e84d7b81f0033399e325b8037ed2b801a5c994e0 using git submodule status after adding the repo.

Still not a native solution, but the best we probably can have without somebody getting their hands really, really dirty in the GIT codebase.

Appendix: Recreating this commit

You should be able to recreate this exact commit using (in an empty directory):

# Initialize new GIT repository
git init

# Set author data (don't set it as part of the `git commit` command or your default data will be stored as “commit author”)
git config --local user.name "Nobody"
git config --local user.email "none"

# Set both the commit and the author date to the start of the Unix epoch (this cannot be done using `git commit` directly)
export GIT_AUTHOR_DATE="Thu Jan 1 00:00:00 1970 +0000"
export GIT_COMMITTER_DATE="Thu Jan 1 00:00:00 1970 +0000"

# Add root commit
git commit --allow-empty --allow-empty-message --no-edit

Creating reproducible GIT commits is surprisingly hard…


Z
Zaz

There's no way to get Git to track directories, so the only solution is to add a placeholder file within the directory that you want Git to track.

The file can be named and contain anything you want, but most people use an empty file named .gitkeep (although some people prefer the VCS-agnostic .keep).

The prefixed . marks it as a hidden file.

Another idea would be to add a README file explaining what the directory will be used for.


G
GAMITG

As mentioned it's not possible to add empty directories, but here is a one liner that adds empty .gitignore files to all directories.

ruby -e 'require "fileutils" ; Dir.glob(["target_directory","target_directory/**"]).each { |f| FileUtils.touch(File.join(f, ".gitignore")) if File.directory?(f) }'

I have stuck this in a Rakefile for easy access.


I'd rather use find . -type d -empty -print0 | xargs --null bash -c 'for a; do { echo "*"; echo "!.gitignore"; } >>"$a/.gitignore"; done' --
C
Community

Many have already answered this question. Just adding a PowerShell version here.

Find all the empty folders in the directory Add a empty .gitkeep file in there

Get-ChildItem 'Path to your Folder' -Recurse -Directory | Where-Object {[System.IO.Directory]::GetFileSystemEntries($_.FullName).Count -eq 0} | ForEach-Object { New-Item ($_.FullName + "\.gitkeep") -ItemType file}

Nice.‌‌ ༼ ͡☉ ͜ʖ ͡☉ ༽
also, ni -p $_.FullName -n .gitignore -v "*`r`n!.gitignore"
C
Community

The solution of Jamie Flournoy works great. Here is a bit enhanced version to keep the .htaccess :

# Ignore everything in this directory
*
# Except this file
!.gitignore
!.htaccess

With this solution you are able to commit a empty folder, for example /log, /tmp or /cache and the folder will stay empty.


He wants to keep a empty directory and not a file.
And i have mentioned that it will keep the .htaccess, too. Example: if a software has a directory for log-files (like oxid eshop) that should not be accesible via web, there is a .htaccess in the directory. If you put the above mentioned .gitignore in the folder, the .htaccess will not be comitted and the folder will be accessible via web.
If you have a .htaccess file that's under version control, then you already have the directory containing it under version control. Thus, the problem is already solved - the .gitignore file becomes irrelevant.
@Wallacoloo Related to the question you're right, nevertheless the file is useful, I'll use it for an upload-directory like that where files shall be protected by .htaccess. Contrary to Romans explanation the .htaccess-file will be committed as it's excluded by the ignore-rule. [old thread, I know]
P
Peter Mortensen

I always build a function to check for my desired folder structure and build it for me within the project. This gets around this problem as the empty folders are held in Git by proxy.

function check_page_custom_folder_structure () {
    if (!is_dir(TEMPLATEPATH."/page-customs"))
        mkdir(TEMPLATEPATH."/page-customs");    
    if (!is_dir(TEMPLATEPATH."/page-customs/css"))
        mkdir(TEMPLATEPATH."/page-customs/css");
    if (!is_dir(TEMPLATEPATH."/page-customs/js"))
        mkdir(TEMPLATEPATH."/page-customs/js");
}

This is in PHP, but I am sure most languages support the same functionality, and because the creation of the folders is taken care of by the application, the folders will always be there.


Just so we're all on the same page, I do not do this anymore. It's a waste of time. The .gitkeep convention is a much better practise.
I can't see how this can be a waste of time. When your TEMPLATEPATH is obviously dynamic you can't use the .gitkeep solution. And even with a nondynamic folder structure you should add some more stuff instead of removing the very good solution of checking directories e.g. check for permissions and chmod the files. Adding a way to mark directories inside a global .gitignore would be perfect for me. Something like #keep /path/to/dir
P
Peter Mortensen

Here is a hack, but it's funny that it works (Git 2.2.1). Similar to what @Teka suggested, but easier to remember:

Add a submodule to any repository (git submodule add path_to_repo)

This will add a folder and a file .submodules. Commit a change.

Delete .submodules file and commit the change.

Now, you have a directory that gets created when commit is checked out. An interesting thing though is that if you look at the content of tree object of this file you'll get:

fatal: Not a valid object name b64338b90b4209263b50244d18278c0999867193

I wouldn't encourage to use it though since it may stop working in the future versions of Git. Which may leave your repository corrupted.


This actually works but I think confuses the heck out of IntelliJ... :|
I've created a better solution based on this that doesn't have these drawbacks: stackoverflow.com/a/58543445/277882
a
ajmedway

If you want to add a folder that will house a lot of transient data in multiple semantic directories, then one approach is to add something like this to your root .gitignore...

/app/data/**/*.* !/app/data/**/*.md

Then you can commit descriptive README.md files (or blank files, doesn't matter, as long as you can target them uniquely like with the *.md in this case) in each directory to ensure that the directories all remain part of the repo but the files (with extensions) are kept ignored. LIMITATION: .'s are not allowed in the directory names!

You can fill up all of these directories with xml/images files or whatever and add more directories under /app/data/ over time as the storage needs for your app develop (with the README.md files serving to burn in a description of what each storage directory is for exactly).

There is no need to further alter your .gitignore or decentralise by creating a new .gitignore for each new directory. Probably not the smartest solution but is terse gitignore-wise and always works for me. Nice and simple! ;)

https://i.stack.imgur.com/xNBNN.png


T
Trendfischer

Sometimes you have to deal with bad written libraries or software, which need a "real" empty and existing directory. Putting a simple .gitignore or .keep might break them and cause a bug. The following might help in these cases, but no guarantee...

First create the needed directory:

mkdir empty

Then you add a broken symbolic link to this directory (but on any other case than the described use case above, please use a README with an explanation):

ln -s .this.directory empty/.keep

To ignore files in this directory, you can add it in your root .gitignore:

echo "/empty" >> .gitignore

To add the ignored file, use a parameter to force it:

git add -f empty/.keep

After the commit you have a broken symbolic link in your index and git creates the directory. The broken link has some advantages, since it is no regular file and points to no regular file. So it even fits to the part of the question "(that contains no files)", not by the intention but by the meaning, I guess:

find empty -type f

This commands shows an empty result, since no files are present in this directory. So most applications, which get all files in a directory usually do not see this link, at least if they do a "file exists" or a "is readable". Even some scripts will not find any files there:

$ php -r "var_export(glob('empty/.*'));"
array (
  0 => 'empty/.',
  1 => 'empty/..',
)

But I strongly recommend to use this solution only in special circumstances, a good written README in an empty directory is usually a better solution. (And I do not know if this works with a windows filesystem...)


a
arcseldon

An easy way to do this is by adding a .gitkeep file to the directory you wish to (currently) keep empty.

See this SOF answer for further info - which also explains why some people find the competing convention of adding a .gitignore file (as stated in many answers here) confusing.


J
Jörg W Mittag

You can't. This is an intentional design decision by the Git maintainers. Basically, the purpose of a Source Code Management System like Git is managing source code and empty directories aren't source code. Git is also often described as a content tracker, and again, empty directories aren't content (quite the opposite, actually), so they are not tracked.


I contest this view. Structure is content, and everything you name contributes to content.
An empty file isn't source code or content either. It's just a name. Yet Git will happily track empty files. I don't think it was an intentional design decision to make Git refuse to track empty directories. I think tracking empty directories is a feature that simply isn't needed 99% of the time, so they didn't bother to do the extra work required to make it work properly. Git can do it if someone wants the feature badly enough to implement it. I doubt the Git maintainers would be opposed to such a patch if it were done correctly.
@TobyAllen here is the updated FAQ link The top answer is also what is recommended by the FAQ with more precise instructions.
It's a missing feature (and low priority), not an intentional limitation. From the Git FAQ: Currently the design of the Git index (staging area) only permits files to be listed, and nobody competent enough to make the change to allow empty directories has cared enough about this situation to remedy it.
Don't really agree. I can find various reasons why I want to track an empty folder. For example, I am developing a very lightweight PHP MVC framework for my projects. I have specific folders for placing models, views, etc. When I make a new site based on my framework, those folders are empty since there are no models or views by default, but I do need the folder to exist, else my framework won't work!