I've noticed this strange behavior of diff and patch when I've used them to force one code base to be identical to another. Let's say I want to update update_me to look identical to leave_unchanged. I go to update_me. I run a diff from leave_unchanged to update_me. Then I patch the diff into update_me. If there are new files in leave_unchanged, patch asks me if my patch was reversed! If I answer yes, it deletes the new files in leave_unchanged. Then, if I simply re-run the patch, it correctly patches update_me.

Why does patch try to modify both leave_unchanged and update_me?

What's the proper way to do this? I found a hacky way which is to replace all +++ lines with nonsense paths so patch can't find leave_unchanged. Then it works fine. It's such an ugly solution though.

$ mkdir copyfrom
$ mkdir copyto
$ echo "Hello world" > copyfrom/myFile.txt
$ cd copyto
$ diff -Naur . ../copyfrom > my.diff
$ less my.diff

diff -Naur ./myFile.txt ../copyfrom/myFile.txt
--- ./myFile.txt        1969-12-31 19:00:00.000000000 -0500
+++ ../copyfrom/myFile.txt      2010-03-15 17:21:22.000000000 -0400
@@ -0,0 +1 @@
+Hello world

$ patch -p0 < my.diff

The next patch would create the file ../copyfrom/myFile.txt,
which already exists!  Assume -R? [n] yes
patching file ../copyfrom/myFile.txt

$ patch -p0 < my.diff
patching file ./myFile.txt