Illegal Byte sequence in a file name while checking out git branch

Question

I am using git in cli to change the current branch:

git checkout dev

and it produces:

fatal: cannot create directory at 'app/src/androidTest/java?com': Illegal byte sequence

As answered in this question and this one, I tried:

LC_ALL=C git checkout dev

or

LC_CTYPE=C git checkout dev

but I am getting the same error as shown above.

Running:

git status

shows that some of the files were changed by the checkout, but I am still on the master branch.

How can I remove the file causing the problems or how can I checkout the branch without getting this error?

Answer 1

I have found a few files in various github projects that are not compatible with one or another operating system. Files with a ".nul" or ".con" extension are a real pain on windows, for example. It isn't a problem exclusive to git. For example Subversion will abort nastily if it can't restore a file for local naming reasons.

In some cases the file may have been uploaded in error. If that is the case for your own projects it should be possible to use the git tools to list the archive and perform a delete of the file from the archive without actually instancing the file locally.

In other cases perhaps that particular file is not significant, and perhaps can be ignored. Perhaps a test will fail if it is missing?

One trick I have used is to stop the whole folder containing that file from being synced by manually creating the directory path, but for the last element, create an empty file instead of a folder. Of course, now the whole test suite will fail.

When the version control tries to do the checkout, it will simply fail to restore the folder, rather than giving a fatal error.

Of course, that only works if the folder is non-critical, e.g. some test files.

The alternative is to piecemeal check-out all but the problem file, but that can be a tedious sequence of checkouts. But you can use this attack to restore the rest of the folder that you omitted using the above technique. Alternatively, locally drag the files from the zip download if they are non-critical.

Answer 2

The locale only affects how things are displayed. If the file name contains a character which isn't allowed by the file system, no amount of locale tweaking can fix that.

I can't think of a way to force a file system to let you create a file which then cannot be used, or a good reason to want to be able to do that.

Probably as a workaround, create a virtualized host with a bare-bones Linux system formatted to permit old-style 8-bit file names (Latin-1 or CP1252 if you can live with the unsavory Windows flavor of that), check out the file there, rename and commit the rename back to git. You still won't be able to check out versions of the source tree from before the rename.