我有两个目录条目，a 和 b。
之前，a和b指向不同的inode。
之后，我希望 b 指向与 a 相同的 inode。
我希望这是安全的 - 我的意思是如果我在某个地方失败，b 要么指向其原始 inode，要么指向 a inode。最特别的是我不想最终 b 消失。
mv 覆盖时是原子的。
当目标已经存在时 ln 似乎不起作用。
所以看起来我可以说：
ln a tmp
mv tmp b
如果发生故障，将会留下一个“tmp”文件，这是不可取的，但也不是灾难。
有更好的方法吗？
（我实际上想做的是将具有相同内容的文件替换为包含该内容的单个 inode，在所有目录条目之间共享）
I have two directory entries, a and b.
Before, a and b point to different inodes.
Afterwards, I want b to point to the same inode as a does.
I want this to be safe - by which I mean if I fail somewhere, b either points to its original inode or the a inode. most especially I don't want to end up with b disappearing.
mv is atomic when overwriting.
ln appears to not work when the destination already exists.
so it looks like i can say:
ln a tmp
mv tmp b
which in case of failure will leave a 'tmp' file around, which is undesirable but not a disaster.
Is there a better way to do this?
(what I'm actually trying to do is replace files that have identical content with a single inode containing that content, shared between all directory entries)
发布评论
评论（2）
ln a tmp && mv tmp b || rm tmp似乎更好，因为如果 ln 失败，则
mv将不会被执行（并且在失败时使 stderr 变得混乱）。ln a tmp && mv tmp b || rm tmpseems better, as then if
lnfails, themvwill not get executed (and clutter up stderr when it fails).事实上，正如您在问题中所述，这是原子地执行此操作的最快方法。
（挑剔者角落：更快地将两个系统调用放在一个程序中）
is in fact the fastest way to do it atomically, as you stated in your question.
(Nitpickers corner: faster to place both system calls in one program)