恢复corrupted gitsvn repository的过程记录

问题

由于主机异常宕机导致git中的文件变成了0字节的，执行git相关的命令报错。

fatal: object 016660b7605cfc2da85f631bbe809f7cb7962608 is corrupted

恢复过程

# 从镜像clone中恢复一个git repo，幸好有一个mirror clone。

git clone --mirror billing@10.164.129.25:/billing/git/cmcc.git /srv/git/cmcc1.git

# 把损坏的git repo中与SVN相关的内容复制到新的git repo中。

cp -R cmcc.bad.git/config cmcc1.git/
cp -R cmcc.bad.git/svn cmcc1.git/
cp -R cmcc.bad.git/logs cmcc1.git/
...
cp -R  cmcc.bad.git/info/refs cmcc1.git/info/
cp -R cmcc.bad.git/refs/heads/master cmcc1.git/refs/heads/

# 根据git log中的信息对新git repo中的相关文件中的hash做了些调整

# 尝试执行/usr/bin/git --git-dir=/srv/git/cmcc1.git svn fetch 仍然报错

唯一没法调整的文件是svn/refs/heads/master/.rev_map.21a24a4a-c245-7e45-a0a4-218b2805dfce，这个不是文本的。

幸运的是从网上搜到了一个生成rev_map文件的脚本 https://gist.github.com/1953425

#! /usr/bin/env gawk -f
# git --no-pager log | gawk -f this-script > revmap.txt
# xxd -c 24 -r revmap.txt \
#    .git/svn/refs/remotes/trunk/.rev_map.cfe28804-0f27-0410-a406-dd0f0b0b656f
# 0000000: 0000 0001 cce6 285e 48e1 e3cc 0d7d 0617 beb0 4e88 a126 8634
# 0000018: 0000 0006 6e4f ada4 bed4 6660 c987 dc74 1243 8973 16cc c810
BEGIN {
    FS=" "
    # this number comes from ``git --no-pager log | grep ^commit | wc -l``
    # and is ZERO-indexed because the last xxd will begin with 0000
    # you might be tempted to think it is the highest r-number, but no,
    # because evidently not every subversion commit made it to git
    cnt=30554
}
/^commit / {
    cmit=$2
}
/git-svn/ {
    # remove everything except the r-number
    sub(/[^@]+@/,"")
    # remove the repository-id
    sub(/ .*$/,"")
    addr=cnt * 24
    # thankfully, xxd is tolerant of not having
    # any spaces, so we can just dump the commitish
    # right into the file
    printf "%08x: %08x%s  \n", addr, $0, cmit
    cnt = cnt - 1
}

删除现有的rev_map文件，按照上面的说明重新生成了一个。

执行 /usr/bin/git --git-dir=/srv/git/cmcc1.git svn fetch 成功。

操作过程经过很多失败的尝试，记录不是很完整，仅作参考。