git encoding trouble: recursively removing a directory where git prints out a different name than it accepts

The story so far:

A few years back I put all my conferences material in a GitHub repository https://github.com/jpluimers/Conferences/. There were a lot directories and files so I didn’t pay much attention to the initial check-in list. The files had been part of copy.com syncing between Windows and Mac machines.

Often git on a Mac is a bit easier than on Windows (on a Mac you can install them with the xcode-select --install trick which installs only the Command Line Tools without having to install the full Xcode [WayBack]).

I choose a Mac because it is closer to a Linux machine than Widows so I expected no encoding trouble (as git has a Linux origin: it “was created by Linus Torvalds in 2005 for development of the Linux kernel“).

Boy I was wrong:

Recently I cloned the repository in a different place and found out a few strange things:

1. Directories with accented characters had been duplicated, for instance in https://github.com/jpluimers/Conferences/tree/master/2011
   1. …/EKON15-Renaissance-Hotel-D%FCsseldorf
   2. …/EKON15-Renaissance-Hotel-Düsseldorf
2. Beyond Compare would show the same content
3. After a check-out git would not understand the %FC encoded directory name (%FC is IEC_8859-1 encoding for ü and \374 is the octal representation of 0xFC [WayBack]) and a git status would show stuff like this:

   • Untracked files:
       (use "git add ..." to include in what will be committed)
     
         EKON15-Renaissance-Hotel-D%FCsseldorf/

     or

     deleted: "EKON15-Renaissance-Hotel-D\374sseldorf/Delphi-XE2-Debugging/BO-EKON15-Delphi-XE2-Debugging.pdf"

4. A git rm -r --cached call [WayBack] would not work, as both these would fail:

   • $ git rm -r --cached EKON15-Renaissance-Hotel-D%FCsseldorf
     fatal: pathspec 'EKON15-Renaissance-Hotel-D%FCsseldorf' did not match any files
     

     and

     $ git rm -r --cached "EKON15-Renaissance-Hotel-D\374sseldorf"
     fatal: pathspec 'EKON15-Renaissance-Hotel-D\374sseldorf' did not match any files
     

5. a

So git could:

• detect the directories and files
• display the names of the detected directories and files
• not translate back the specified names into directories and files

All if this was with:

$ git --version
git version 1.9.5 (Apple Git-50.3)

This is how I fixed it

First I created an alias:

alias git-config="echo global: ; git config --list --global ; echo local: ; git config --lis --local ; echo system: ; git config --list --system"


That allowed me to view the git settings on various levels in my system.

It revealed I didn’t have the core.precomposeunicode setting at all (valid values are true or false). I also read various stories about one or both being the correct value: osx – Git and the Umlaut problem on Mac OS X – Stack Overflow [WayBack].

–jeroen

Result of git status: