Simply speaking, I need a way to fast convert files, so files from VSS can be committed to a Git repo (Why? I like it. http://wiki.freebsd.org/LocalGit and http://wiki.freebsd.org/LocalMercurial .)
Files containing trailing blanks can only be committed with "git commit -n", but I don't like that. Since the VSS=>Git operation is not done frequently, I was using dos2unix then sed to trim the blanks, but sed cannot recognize a file is binary or text, and my existing method to probe that is too slow.
Suggest me to use Hg or bzr please. Any speed benchmark on win32? Repo size is 13k files, 250M (single snapshot).
Thanks!
----
update: I got it with git's own pre-commit hook (use diff to find text file or not). Commit first, then trim trailing blanks, then commit again: diff will be faster, and object can be reused I think (so add can be faster). Code like this, how to use xargs?
#!/bin/sh
function do_trim_text_files_and_commit()
{
if [ $# == 0 ]; then
return
fi
chmod +w "$@"
dos2unix "$@"
sed -i -e "s,\( \|\t\)\+$,," "$@"
git add -f "$@"
}
# from git "pre-commit", use "diff" to find text files
function trim_text_files_and_commit_again()
{
git-diff-index -p -M --cached HEAD -- |
perl -e '
my $filename;
my $reported_filename = "";
while(<>){
if (m|^diff --git a/(.*) b/\1$|) {
$filename = $1;
next;
}
if (/^@@ -\S+ \+(\d+)/) {
next;
}
if (/^ /) {
next;
}
if (s/^\+//) {
$lineno++;
chomp;
if (/\s$/) {
if ($reported_filename ne $filename){
print "$filename\n";
$reported_filename = $filename;
}
}
}
}
' |
tr '\n' '\0' | xargs -0 trim-text-files
}
function add_files()
{
if [ $# == 0 ]; then
echo "no files specified" > /dev/stderr
return
fi
git add -f "$@" &&
# trim-text-files can only be run in repo TOP folder.
while :; do
if [ -d .git ]; then
trim_text_files_and_commit_again
break
fi
cd ..
done
}
if [ "$0" == "/usr/local/bin/git-add-project" ]; then
git ls-files "$@" |
tr '\n' '\0' | xargs -0 ~/git-add.sh
elif [ "$0" == "/usr/local/bin/trim-text-files" ]; then
do_trim_text_files_and_commit "$@"
else
add_files "$@"
fi
