I have been seeking it on Google for 2 hours and still can't find the solution.
I want to checkout all files with wildcard to a specific revision. I used following command:
git checkout 663090b1c 'src/app/**/*.spec.ts'
But it says:
error: pathspec 'src/app/**/*.spec.ts' did not match any file(s) known to git.
Can anyone help me?
I believe that this should work, but clearly it doesn't. Here's the effect of using the most recent Git (which you can get at https://github.com/git/git) on Documentation/**/*.txt
.
I built Git, then used the system Git (at 2.21, slightly behind this version which is 2.22.0.545.g9c9b961d7e
) to do this:
$ rm Documentation/*/*.txt
$ git status --short | head
D Documentation/RelNotes/1.5.0.1.txt
D Documentation/RelNotes/1.5.0.2.txt
D Documentation/RelNotes/1.5.0.3.txt
D Documentation/RelNotes/1.5.0.4.txt
D Documentation/RelNotes/1.5.0.5.txt
D Documentation/RelNotes/1.5.0.6.txt
D Documentation/RelNotes/1.5.0.7.txt
D Documentation/RelNotes/1.5.0.txt
D Documentation/RelNotes/1.5.1.1.txt
D Documentation/RelNotes/1.5.1.2.txt
$ git checkout -- 'Documentation/**/*.txt'
$ git status --short | head
$
Now, however:
$ ./git checkout HEAD -- 'Documentation/**/*.txt'
error: pathspec 'Documentation/**/*.txt' did not match any file(s) known to git
It seems that pathspecs work when used with the index, but not when used with a commit hash.
As a workaround—note that this workaround is somewhat defective, though you can patch around it further—you can first read the desired commit into a temporary index, then do the git checkout
from the temporary index. The following script is completely untested.
#! /bin/sh
# git-co-c-ps: git checkout <commit> <pathspec>
# work around a bug in which combining <commit> and <pathspec> does not work
. git-sh-setup
case $# in
2) ;;
*) die "usage: $0 <commit> <pathspec>";;
esac
hash=$(git rev-parse "$1^{commit}") || exit 1
export GIT_INDEX_FILE=$(mktemp) || exit 1
rm -f $GIT_INDEX_FILE
trap "rm -f $GIT_INDEX_FILE" 0 1 2 3 15
git read-tree "$hash" || exit 1
git checkout -- "$2"
The (untested) idea here is to read the commit specified by "$1" into a temporary index. (We could relax the ^{commit}
to be just ^{tree}
as any tree-ish should suffice.) Then, having read that into a temporary index, we use the mode of "git checkout" with a pathspec ($2) that works, using the temporary index. This writes to the work-tree.
The defect is that the now-updated work-tree files are not also updated in the index. A real git checkout
would have copied these same files to the index. It should be possible to unset $GIT_INDEX_FILE
and git add
the added files to the real index. This is left as an exercise. Note that it is not as trivial as just git add -- "$2"
since some work-tree files may be in the work-tree not because they were read into the temporary index by the git read-tree
operation, but rather because they were already in the work-tree. (Some may be tracked and modified-but-not-yet-staged/added, and others may be untracked.)
Another, perhaps better, workaround is to use git ls-files
on the temporary index, to find the files of interest. Use xargs
or similar to pass these file names to a git checkout
that is run without the temporary index:
files_file=$(mktemp)
trap "rm -f $files_file" 0 1 2 3 15
(
export GIT_INDEX_FILE=$(mktemp) || exit 125
rm -f $GIT_INDEX_FILE
trap "rm -f $GIT_INDEX_FILE" 0 1 2 3 15
git read-tree $hash
git ls-files -z -- "$2"
) > $files_file
# if the inner shell exited with an error, exit now
case $? in 125) exit 1;; esac
xargs -0 ...
(Filling in the rest of this to make a usable Git shell command is also left as an exercise.)
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With