Git - Difference Between 'assume-unchanged' and 'skip-worktree'

I have local changes to a file that I don't want to commit to my repository. It is a configuration file for building the application on a server, but I want to build locally with different settings. Naturally, the file always shows up when I do 'git status' as something to be staged. I would like to hide this particular change and not commit it. I won't make any other changes to the file.

To clarify, using .gitignore is not what I want since that only prevents new files from being added. I want to ignore changes to a file already in the repository.

After some digging around, I see 2 options: assume-unchanged and skip-worktree. A previous question here talks about them but doesn't really explain their differences.

How are the two commands different? Why would someone use one or the other?

You want skip-worktree.

assume-unchanged is designed for cases where it is expensive to check whether a group of files have been modified; when you set the bit, git (of course) assumes the files corresponding to that portion of the index have not been modified in the working copy. So it avoids a mess of stat calls. This bit is lost whenever the file's entry in the index changes (so, when the file is changed upstream).

skip-worktree is more than that: even where git knows that the file has been modified (or needs to be modified by a reset --hard or the like), it will pretend it has not been, using the version from the index instead. This persists until the index is discarded.

There is a good summary of the ramifications of this difference and the typical use cases here: http://fallengamer.livejournal.com/93321.html .

From that article:

• --assume-unchanged assumes that a developer shouldn’t change a file. This flag is meant for improving performance for not-changing folders like SDKs.
• --skip-worktree is useful when you instruct git not to touch a specific file ever because developers should change it. For example, if the main repository upstream hosts some production-ready configuration files and you don’t want to accidentally commit changes to those files, --skip-worktree is exactly what you want.

Note: fallengamer did some tests in 2011 (so they may be outdated), and here are his findings:

Operations

• File is changed both in local repository and upstream
  git pull:
  Git preserves local changes anyway.
  Thus you wouldn’t accidentally lose any data that you marked with any of the flags.
  • File with assume-unchanged flag: Git wouldn’t overwrite local file. Instead it would output conflicts and advices how to resolve them
  • File with skip-worktree flag: Git wouldn’t overwrite local file. Instead it would output conflicts and advices how to resolve them

• File is changed both in local repository and upstream, trying to pull anyway
  git stash
git pull
  Using skip-worktree results in some extra manual work but at least you wouldn’t lose any data if you had any local changes.
  • File with assume-unchanged flag: Discards all local changes without any possibility to restore them. The effect is like ‘git reset --hard’. ‘git pull’ call will succeed
  • File with skip-worktree flag: Stash wouldn’t work on skip-worktree files. ‘git pull’ will fail with the same error as above. Developer is forced to manually reset skip-worktree flag to be able to stash and complete the failing pull.

• No local changes, upstream file changed
  git pull
  Both flags wouldn’t prevent you from getting upstream changes. Git detects that you broke assume-unchanged promise and choses to reflect the reality by resetting the flag.
  • File with assume-unchanged flag: Content is updated, flag is lost.
    ‘git ls-files -v’ would show that flag is modified to H (from h).
  • File with skip-worktree flag: Content is updated, flag is preserved.
    ‘git ls-files -v' would show the same S flag as before the pull.

• With local file changed
  git reset --hard
  Git doesn’t touch skip-worktree file and reflects reality (the file promised to be unchanged actually was changed) for assume-unchanged file.
  • File with assume-unchanged flag: File content is reverted. Flag is reset to H (from h).
  • File with skip-worktree flag: File content is intact. Flag remains the same.

He adds the following analysis:

• It looks like skip-worktree is trying very hard to preserve your local data. But it doesn’t prevent you to get upstream changes if it is safe. Plus git doesn’t reset the flag on pull.
  But ignoring the ‘reset --hard' command could become a nasty surprise for a developer.

• Assume-unchanged flag could be lost on the pull operation and the local changes inside such files doesn’t seem to be important to git.

See:

• Junio's (current git maintainer) comment regarding intent of assume-unchanged,

  In particular, Junio points out that changes to assume-unchanged files could accidentally be committed: "if Git can determine a path that is marked as assume-unchanged has changed without incurring extra lstat(2) cost, it reserves the right to report that the path has been modified (as a result, git commit -a is free to commit that change)."

• difference between assume-unchanged and skip-worktree as discussed in git mailing list upon addition of skip-worktree patch.

He concludes:

Actually neither of the flags is intuitive enough.

• assume-unchanged assumes that a developer shouldn’t change a file. If a file was changed – then that change is not important. This flag is meant for improving performance for not-changing folders like SDKs.
  But if the promise is broken and a file is actually changed, git reverts the flag to reflect the reality. Probably it’s ok to have some inconsistent flags in generally not-meant-to-be-changed folders.

• On the other hand skip-worktree is useful when you instruct git not to touch a specific file ever. That is useful for an already tracked config file.
  Upstream main repository hosts some production-ready config but you would like to change some settings in the config to be able to do some local testing. And you don’t want to accidentally check the changes in such file to affect the production config. In that case skip-worktree makes perfect scene.

With Git 2.25.1 (Feb. 2020), the "Actually neither of the flags is intuitive enough" mentioned above is further clarified:

See commit 7a2dc95, commit 1b13e90 (22 Jan 2020) by brian m. carlson (bk2204).
^{(Merged by Junio C Hamano -- gitster -- in commit 53a8329, 30 Jan 2020)}
(Git Mailing list)

doc: dissuade users from trying to ignore tracked files

^{Signed-off-by: Jeff King
Signed-off-by: brian m. carlson}

It is quite common for users to want to ignore the changes to a file that Git tracks.

Common scenarios for this case are IDE settings and configuration files, which should generally not be tracked and possibly generated from tracked files using a templating mechanism.

However, users learn about the assume-unchanged and skip-worktree bits and try to use them to do this anyway.

This is problematic, because when these bits are set, many operations behave as the user expects, but they usually do not help when git checkout needs to replace a file.

There is no sensible behavior in this case, because sometimes the data is precious, such as certain configuration files, and sometimes it is irrelevant data that the user would be happy to discard.

Since this is not a supported configuration and users are prone to misuse the existing features for unintended purposes, causing general sadness and confusion, let's document the existing behavior and the pitfalls in the documentation for git update-index so that users know they should explore alternate solutions.

In addition, let's provide a recommended solution to dealing with the common case of configuration files, since there are well-known approaches used successfully in many environments.

The git update-index man page now includes:

Users often try to use the assume-unchanged and skip-worktree bits to tell Git to ignore changes to files that are tracked. This does not work as expected, since Git may still check working tree files against the index when performing certain operations. In general, Git does not provide a way to ignore changes to tracked files, so alternate solutions are recommended.

For example, if the file you want to change is some sort of config file, the repository can include a sample config file that can then be copied into the ignored name and modified. The repository can even include a script to treat the sample file as a template, modifying and copying it automatically.

That last part is what I describe a typical content filter driver based on smudge/clean scripts.