Since combinediff doesn't have the advantage of being able to look at the files that are to be modified, it has stricter requirements on the input format than patch(1) does. The output of GNU diff will be okay, even with extensions, but if you intend to use a hand-edited patch it might be wise to clean up the offsets and counts using recountdiff(1) first.
Note, however, that the two patches must be in strict incremental order. In other words, the second patch must be relative to the state of the original set of files after the first patch was applied.
The diffs may be in context format. The output, however, will be in unified format.
-p n, --strip-match=n
-U n, --unified=n
-d pattern, --drop-context=PATTERN
Note that the interpretation of the shell wildcard pattern does not count slash characters or periods as special (in other words, no flags are given to fnmatch). This is so that "*/basename"-type patterns can be given without limiting the number of pathname components.
The -U option is a bit erratic: it can control the amount of context displayed for files that are modified in both patches, but not for files that only appear in one patch (which appear with the same amount of context in the output as in the input).
Tim Waugh <email@example.com>