It's an interesting idea, though I am not sure if this is a good way to divide the output. The patch command already uses stderr for problems internal to the program itself. In other words stderr shows the status of patch itself while stdout displays information regarding the files being worked on.

If problems with the files being worked on are moved to stderr then they will mix with patch's own internal warnings while only "good" reports will print to stdout. That seems more messy rather than less, in my opinion.

This seems like a situation where output should be filtered and divided by existing UNIX utilities rather than by the program itself. Something like "patch -p2 < my.diff | tee all.log | grep FAIL > bad.log"

Follow-up thoughts? Opinions?