Does anyone know how to patch a file, while not changing lines matching a pattern?

I know there is an -I argument to diff, but this only works if every line in the hunk matches the regexp you supply. If you could make every hunk be only 1 line long it would be okay... but this doesn't seem to be an option.

Here's an example to clarify what I want to do:

file1 (5 lines):

line1
line2
line3
line4 retain
line5

file2 (6 lines):

line1
new line2
line3
new line 4
line5
line6

result (6 lines):

line1
new line2
line3
line4 retain
line5
line6

In the example you're giving it would be easier to just remove that bit from the patch.

Thanks for your tip. I tried this (though I suspect not exhaustively), but it seems a bit tricky to adjust the patch. I mean, if I just do a "grep -v retain" on the output that diff generates then it's not going to match the to-be-replaced hunks anymore. I realized that I could do a "diff -c" and then filter for lines beginning with ! and containing my string "retain" using awk to keep them but just remove the leading !. But then there's gonna be a corresponding line in the new hunk which is marked with a !, and again, it's not going to match unless I manually count and remove the corresponding ! in the new hunk. This could also yield hunks which don't have any changes (I suppose that might not be a problem), but this is really a serious pain to implement.

Is there a way to generate a patch file that can be filtered easily?