On Tuesday, 24. July 2012 at 13:26, Junio C Hamano wrote: > Jens Lehmann <jens.lehm...@web.de (http://web.de)> writes: > > > Am 24.07.2012 21:01, schrieb Justin Spahr-Summers: > > > This occurs on Mac OS X 10.7.4, on git versions 188.8.131.52 (Apple Git-33) > > > and 184.108.40.206. > > > > > > Steps: > > > 1. Create or clone a repository to an absolute path that contains spaces. > > > 2. Add a submodule to the repository, if it does not already have one. > > > 3. Within that submodule, attempt to add another submodule. > > > > > > The result is an error "fatal: Not a git repository", followed by the > > > relative path to the submodule directory within .git/modules of the > > > top-level repository. > > > > > > Similarly, using "git submodule update --init --recursive" in a > > > freshly-cloned repository that matches the above configuration will fail > > > with the same error. "git clone --recursive" does not seem to suffer from > > > the same problem at clone time, but will still fail to add recursive > > > submodules. > > > > Hmm, I don't understand how that is different from what t7407 does, it uses > > "git submodule update --init --recursive" in to populate recursive > > submodules > > in a freshly cloned repository whose path contains a space (in the trash > > directory name) in test number 8. > > > > I can see one codepath that would behave incorrectly, especially if > the submodule path relative to the superproject has whitespaces in > it. In module_clone(), you have: > > # We already are at the root of the work tree but cd_to_toplevel will > # resolve any symlinks that might be present in $PWD > a=$(cd_to_toplevel && cd "$gitdir" && pwd)/ > b=$(cd_to_toplevel && cd "$sm_path" && pwd)/ > ... > # Turn each leading "*/" component into "../" > rel=$(echo $b | sed -e 's|[^/][^/]*|..|g') > echo "gitdir: $rel/$a" >"$sm_path/.git" > > I _think_ $sm_path is computed correctly by the codeflow leading to > this place, and $b is also computed correctly, but notice the lack > of quoting around $b when you echo it? It will be split at $IFS, so > if b='/Program Files/My Stupidity/', the sed script will see a > single SP between My and Stupidity, which is different from what you > wanted to feed, I presume. > > Having said that, I do not think git-submodule is prepared to take > paths with path-unsafe characters in it, given that many part of it > has loops like "while read mode sha1 stage sm_path" that reads from > ls-files/ls-tree output without -z (which means it cannot handle > pathnames with LF in them). > > My recommendation at this point (i.e. not a long term) for people > with problems Justin saw is "Don't do it then".
I appreciate the debugging work. Unfortunately, none of the relative submodule paths have had whitespace in them, so I'm not sure that's the issue. Here's some real output, with a couple specific names removed, starting from the root of the top-level repository (where External/twui is a submodule): $ cd External/twui $ git submodule add git://github.com/petejkim/expecta.git TwUITests/expecta Cloning into 'TwUITests/expecta'... remote: Counting objects: 988, done. remote: Compressing objects: 100% (404/404), done. remote: Total 988 (delta 680), reused 842 (delta 535) Receiving objects: 100% (988/988), 156.30 KiB, done. Resolving deltas: 100% (680/680), done. fatal: Not a git repository: ../../../../../../../../Volumes/drive name with spaces/Users/justin/Documents/Programming/project name with spaces/.git/modules/External/twui/modules/TwUITests/expecta -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html