Re: git fast-export not preserving executable permissions?
From: Taylor Blau <hidden>
Date: 2020-05-01 22:32:45
On Fri, May 01, 2020 at 09:35:28AM -0400, Doug Glidden wrote:
Taylor,
(My full response is below, but please in general do not top-quote mail here.)
Thanks for your response! It looks like git does not actually
recognize the file as executable:
$ git ls-tree HEAD
100644 blob 7d2f57b2381766924e1e4ffcc62615c637bbd784 executable_script.sh
100644 blob d1d7cf309e091f54f268503b31653d8eba42fe88
non_executable_file.txt
Now you have me wondering if the real problem here is that I'm working
in git-bash on a Windows machine, which means the file permissions
aren't completely native.
I was wondering if that was the case ;-). If you are using NTFS or
FAT32, neither of these filesystems support execute permission bits. (I
am certainly not an expert here, but I know that Dscho (cc'd) would be
able to answer authoritatively here.)
That said, *Git* understands executable permissions, even if your
filesystem doesn't. You can tell Git to mark a file as executable by
the following:
$ git update-index --chmod=+x /path/to/file
and then committing the result. Round-tripping this through 'git
fast-{im,ex}port' should preserve the permissions from Git's
perspective, and ditto for checking out the contents of a repository on
a filesystem that does support the executable permission bit.
I'm going to run a similar experiment in a native Linux environment and see if I get the same results. I'll let you know what I find.
Sounds good. I'll be very surprised if it doesn't work as you expect.
Thanks, Doug On Wed, Apr 29, 2020 at 2:49 PM Taylor Blau [off-list ref] wrote:quoted
Hi Doug, On Wed, Apr 29, 2020 at 09:36:31AM -0400, Doug Glidden wrote:quoted
Hello Git world! I have run into an issue that I cannot seem to resolve with git fast-export. When running a fast-export on a repo that contains scripts with executable permissions (e.g. a gradlew script), the resulting export does not properly reflect the executable permissions on the script files.Interesting. fast-import and fast-export both understand executable modes (although Git only understands the modes 644 and 755 for blobs), so this should be working. I can not reproduce the issue as-is. Round-tripping a fast-import and fast-export preserves executable bits for me: #!/bin/bash set -e rm -rf repo client git init -q repo git init -q client ( cd repo printf "x" >x printf "y" >y chmod +x x git add x y git commit -q -m "initial commit" ) git -C repo fast-export HEAD | git -C client fast-import diff -u <(git -C repo ls-tree HEAD) <(git -C client ls-tree HEAD)quoted
To illustrate this issue, I created a small sample repo, with one executable file and one non-executable file. From the output below, you can see that the mode in the output from fast-export is the same for both files; according to the documentation for fast-import, the mode for the executable file should be 100755 instead of 100644. $ ls -gG total 2 -rwxr-xr-x 1 106 Apr 29 09:13 executable_script.sh* -rw-r--r-- 1 63 Apr 29 09:12 non_executable_file.txt $ git fast-export --all blob mark :1 data 106 #!/bin/bash # This is a shell script that should be executable. echo 'The script executed successfully!' blob mark :2 data 63 This file is a simple text file that should not be executable. reset refs/heads/dev commit refs/heads/dev mark :3 author Doug [off-list ref] 1588167102 -0400 committer Doug [off-list ref] 1588167102 -0400 data 25 Adding some sample files M 100644 :1 executable_script.sh M 100644 :2 non_executable_file.txt Please let me know if there is any further information I can provide about this issue.Does Git think that the file is executable? Please run 'git ls-tree HEAD' to find out.quoted
Thank you, DougThanks, Taylor
Thanks, Taylor