[PATCH] #3460 svn patch is not fuzzy when applying unidiffs

Hi!

[[[
Fix #3460 - svn patch is not fuzzy when applying unidiffs.

* subversion/include/private/svn_diff_private.h
  (svn_hunk_t): Add fields context_before and context_before. They are
    used for determining if there is enough context to apply a patch
    with fuzz.

* subversion/libsvn_diff/parse-diff.c
  (parse_next_hunk): Count number of lines of context at start and end
    of hunk and save the information in hunk.

* subversion/tests/cmdline/patch_tests.py
  (patch_with_fuzz): New.
  (test_list): Add new test.

* subversion/libsvn_client/patch.c
  (hunk_info_t): Add field fuzz.
  (match_hunk): Add new parameter fuzz. Record nr of lines read and
    ignore the ones at the beginning and end of the hunk that is fuzzy.
    Use context_before and context_after to ignore the cases when there
    isn't enough context to do fuzzy patching.
  (scan_for_match): Add new parameter fuzz. Call match_hunk() with fuzz.
  (get_hunk_info): Add new parameter fuzz. Call scan_for_match() with
    fuzz. Save the used fuzz in hi to be used when the hunk should be
    copied to the target.
  (copy_hunk_text): Add nr_of_lines and fuzz parameter. Record read_lines
    and only copy those who are not within the fuzz limit. They have
    already been or will be copied from the target. If nr_of_lines is 0
    we assume that we should copy all lines and ignore fuzz.
  (apply_one_hunk): Adjust lines copied to the target to include the
    context lines at the beginning of the hunk who are fuzzy. Adjust the
    current_line of the target to point to the last line of the hunk
    minus fuzz.
  (apply_one_patch): Try to call get_hunk_info() with no fuzz. If we get
    no matching line try with fuzz 1 and if that fails try with fuzz 2.
  
Patch by: Daniel Näslund <daniel{_AT_}longitudo.com>
]]]

copy_hunk_text() is called from two places. It can either be a
modified_text or diff_text. Diff_text contains the whole hunk and is
copied if we have a rejected hunk. There is no recording of the number
of lines. So I said that if nr_of_lines is zero then we just copy all
lines. A bit of a quick hack. I think we can do it since it's just a
local helper but someone else might object.

It's easy to make off-by-one errors. I probably should extend the test
case further. But this one at least covers:

* Fuzz at beginning of file
* Fuzz at end of file
* Fuzz with value 1
* Fuzz with value 2
* Fuzz 2 with only the second context line changed

The patch test only adds new lines. There is no removal or delete.

Daniel