import re, operator, io, sys

rem_range = re.compile(b'@@ -([0-9]+)(?:,([0-9]+))? \\+([0-9]+)(?:,([0-9]+))? @@(?: .*)?').fullmatch
rem_diffgit = re.compile('diff --git a/([^ ]*) b/([^ ]*)').fullmatch

INFINITY = 2**64

class PatchError(Exception):
    __slots__ = ('a', 'b')
    def __init__(self, a, b, /, fn=None):
        super().__init__(fn)
        self.a = a
        self.b = b

    def __str__(self, /):
        fn = self.args[0]
        r = [f' in {fn!r}:' if fn is not None else ':']

        def tabmangle(p):
            o = io.BytesIO()
            p.write(o.write)
            p = o.getvalue().decode('utf-8', 'replace').splitlines()
            return ['    ' + l for l in p]

        r.append('')
        r += tabmangle(self.a)

        r.append('')
        r += tabmangle(self.b)

        return '\n'.join(r)

class ConflictError(PatchError):
    __slots__ = ()
    
    def __str__(self, /):
        return 'Conflict' + PatchError.__str__(self)

class MismatchError(PatchError):
    __slots__ = ()
    
    def __str__(self, /):
        return 'Mismatch' + PatchError.__str__(self)

class Hunk:
    __slots__ = ('lhs_line', 'rhs_line', 'pre_context', 'lhs', 'rhs', 'post_context')

    def __init__(self, /, *, lhs_line, rhs_line, pre_context, lhs, rhs, post_context):
        assert lhs_line > 0
        assert rhs_line > 0

        self.lhs_line = lhs_line
        self.rhs_line = rhs_line
        self.pre_context = pre_context
        self.lhs = lhs
        self.rhs = rhs
        self.post_context = post_context

    @property
    def lhs_size(self, /):
        return len(self.pre_context) + len(self.lhs) + len(self.post_context)

    @property
    def rhs_size(self, /):
        return len(self.pre_context) + len(self.rhs) + len(self.post_context)

    def write(self, w, /, munge=False, *, lhs_line=None, rhs_line=None, lhs_nl=True, rhs_nl=True):
        if lhs_line is None:
            lhs_line = self.lhs_line
        if rhs_line is None:
            rhs_line = self.rhs_line
        pre_context = self.pre_context
        lhs = self.lhs
        rhs = self.rhs
        post_context = self.post_context
        clen = len(pre_context) + len(post_context)
        llen = len(lhs) + clen
        rlen = len(rhs) + clen

        if munge:
            w(f'@@ -xxx,{llen} +yyy,{rlen} @@'.encode('utf-8'))
        else:
            w(f'@@ -{lhs_line},{llen} +{rhs_line},{rlen} @@'.encode('utf-8'))

        for l in pre_context:
            w(b'\n ')
            w(l)

        for l in lhs:
            w(b'\n-')
            w(l)

        for l in rhs:
            w(b'\n+')
            w(l)

        for l in post_context:
            w(b'\n ')
            w(l)

        w(b'\n')

    @classmethod
    def parse_l(tp, l, line, /):
        m = rem_range(l)
        if m is None:
            raise ValueError(f'Invalid patch: bad range info {l!r}')
        lhs_line,n2,rhs_line,n4 = m.groups()
        lhs_line = 1 if lhs_line == b'0' else int(lhs_line)
        rhs_line = 1 if rhs_line == b'0' else int(rhs_line)
        n2 = int(n2) if n2 else 1
        n4 = int(n4) if n4 else 1

        hunk = []
        prev = -1
        lhs_nl = True
        rhs_nl = True
        post_nl = True
        while True:
            l = line()
            if l.startswith(b'\\ '):
                if prev == 0:
                    lhs_nl = False
                elif prev == 1:
                    rhs_nl = False
                elif prev == -1:
                    post_nl = False
                else:
                    raise ValueError('Invalid patch: bad "No newline at end of file" flag')
                prev = -2
                continue
            if not n2 and not n4:
                break
            k = l[0]
            if k == b' '[0]:
                if not post_nl or not lhs_nl or not rhs_nl:
                    raise ValueError('Invalid patch: bad "No newline at end of file" flag')
                if not n2 or not n4:
                    raise ValueError('Invalid patch: bad line count')
                n2 -= 1
                n4 -= 1
                prev = -1
            elif k == b'-'[0]:
                if not post_nl or not lhs_nl:
                    raise ValueError('Invalid patch: bad "No newline at end of file" flag')
                if not n2:
                    raise ValueError('Invalid patch: bad line count')
                n2 -= 1
                prev = 0
            elif k == b'+'[0]:
                if not post_nl or not rhs_nl:
                    raise ValueError('Invalid patch: bad "No newline at end of file" flag')
                if not n4:
                    raise ValueError('Invalid patch: bad line count')
                n4 -= 1
                prev = 1
            else:
                raise ValueError(f'Invalid patch: bad prefix character {chr(k)!r}')
                assert False
            hunk.append(l)
        lret = l

        strip1 = operator.itemgetter(slice(1,None))

        for i,l in enumerate(hunk):
            if l[0] != b' '[0]:
                break
        else:
            raise ValueError('Invalid patch: no changed lines in hunk')
        pre_context = (*map(strip1, hunk[:i]),)
        del hunk[:i]

        j = -1
        while True:
            l = hunk[j]
            if l[0] != b' '[0]:
                break
            j -= 1
        j += len(hunk) + 1
        if not post_nl:
            hunk[j] += b'\n\\ No newline at end of file'
        post_context = (*map(strip1, hunk[j:]),)
        del hunk[j:]

        lhs = []
        rhs = []
        for l in hunk:
            l0 = l[0]
            if l0 != b'+'[0]:
                lhs.append(l)
            if l0 != b'-'[0]:
                rhs.append(l)
        if not lhs_nl:
            lhs[-1] += b'\n\\ No newline at end of file'
        if not rhs_nl:
            rhs[-1] += b'\n\\ No newline at end of file'

        return lret, tp(
            lhs_line = lhs_line,
            rhs_line = rhs_line,
            pre_context = pre_context,
            lhs = (*map(strip1, lhs),),
            rhs = (*map(strip1, rhs),),
            post_context = post_context,
        )

    def reduce_context(self, n=3):
        pre = self.pre_context
        lhs = self.lhs
        rhs = self.rhs
        post = self.post_context

        i = 0
        j = -1
        m = min(len(lhs), len(rhs))
        while i - j <= m:
            if lhs[i] == rhs[i]:
                i += 1
                if i - j > m:
                    break
                if lhs[j] == rhs[j]:
                    j -= 1
            else:
                if lhs[j] != rhs[j]:
                    break
                j -= 1
        j += 1
        jl = len(lhs) + j
        jr = len(rhs) + j

        assert lhs[:i] == rhs[:i]
        assert lhs[jl:] == rhs[jr:]
        pre = pre + lhs[:i]
        post = lhs[jl:] + post
        lhs = lhs[i:jl]
        rhs = rhs[i:jr]

        if not lhs and not rhs:
            return None

        assert lhs != rhs
        assert not lhs or not rhs or lhs[0] != rhs[0] and lhs[-1] != rhs[-1]

        lhs_line = self.lhs_line
        rhs_line = self.rhs_line

        if n is None:
            pass
        elif not n:
            d = len(pre)
            lhs_line += d
            rhs_line += d
            pre = ()
            post = ()
        else:
            if len(pre) > n:
                d = len(pre) - n
                lhs_line += d
                rhs_line += d
                pre = pre[d:]

            if len(post) > n:
                post = post[:n]

        return Hunk(
            lhs_line = lhs_line,
            rhs_line = rhs_line,
            pre_context = pre,
            lhs = lhs,
            rhs = rhs,
            post_context = post,
        )

def join_hunks(hs_a, hs_b):
    hs_a = iter(hs_a)
    hs_b = iter(hs_b)

    need_a = True
    need_b = True
    have_cur = False

    def make_cur(n):
        if not have_cur:
            return None
        return Hunk(
            lhs_line = cur_lhs_line,
            rhs_line = cur_rhs_line,
            pre_context = (),
            lhs = tuple(cur_lhs),
            rhs = tuple(cur_rhs),
            post_context = (),
        ).reduce_context(n)

    delta = 0
    r = []

    while True:
        if need_a:
            need_a = False
            a = next(hs_a, None)

        if need_b:
            need_b = False
            b = next(hs_b, None)

        if not have_cur:
            if a is not None and (b is None or a.rhs_line <= b.lhs_line):
                cur_lhs_line = a.lhs_line
                cur_rhs_line = a.lhs_line + delta
                cur_lhs = [*a.pre_context, *a.lhs, *a.post_context]
                cur_rhs = [*a.pre_context, *a.rhs, *a.post_context]
                need_a = True
            elif b is not None:
                cur_lhs_line = b.rhs_line - delta
                cur_rhs_line = b.rhs_line
                cur_lhs = [*b.pre_context, *b.lhs, *b.post_context]
                cur_rhs = [*b.pre_context, *b.rhs, *b.post_context]
                need_b = True
            else:
                return (*r,)
            have_cur = True
            continue

        cur_lhs_end = cur_lhs_line + len(cur_lhs)
        cur_rhs_end = cur_rhs_line + len(cur_rhs)

        if a is not None and a.lhs_line <= cur_lhs_end:
            lhs = [*a.pre_context, *a.lhs, *a.post_context]
            rhs = [*a.pre_context, *a.rhs, *a.post_context]

            i = a.lhs_line - cur_lhs_line
            k = i + len(rhs)

            if len(cur_lhs) >= k:
                if cur_lhs[i:k] != rhs:
                    raise MismatchError(make_cur(None), a)
                cur_lhs[i:k] = lhs
            else:
                j = len(cur_lhs) - i
                if cur_lhs[i:] != rhs[:j]:
                    raise MismatchError(make_cur(None), a)
                cur_lhs[i:] = lhs
                cur_rhs.extend(rhs[j:])
            need_a = True
            continue

        if b is not None and b.rhs_line <= cur_rhs_end:
            lhs = [*b.pre_context, *b.lhs, *b.post_context]
            rhs = [*b.pre_context, *b.rhs, *b.post_context]
            i = b.rhs_line - cur_rhs_line
            k = i + len(lhs)

            if len(cur_rhs) >= k:
                if cur_rhs[i:k] != lhs:
                    raise MismatchError(make_cur(None), b)
                cur_rhs[i:k] = rhs
            else:
                j = len(cur_rhs) - i
                if cur_rhs[i:] != lhs[:j]:
                    raise MismatchError(make_cur(None), b)
                cur_rhs[i:] = rhs
                cur_lhs.extend(lhs[j:])
            need_b = True
            continue

        cur = make_cur(3)
        have_cur = False
        if cur is None:
            continue
        r.append(cur)
        delta += len(cur.rhs) - len(cur.lhs)

def commute_hunks(hs_a, hs_b):
    hs_a = iter(hs_a)
    hs_b = iter(hs_b)

    need_a = True
    need_b = True

    delta_a = 0
    delta_b = 0
    ra = []
    rb = []
    wa = ra.append
    wb = rb.append

    while True:
        if need_a:
            need_a = False
            try:
                a = next(hs_a)
            except StopIteration:
                if b is None and not need_b:
                    break
                a = None
                a_begin = INFINITY
                a_pre = ()
                a_post = ()
                a_end = INFINITY
            else:
                a_pre = a.pre_context
                a_post = a.post_context
                a_begin = a.rhs_line + len(a_pre)
                a_end = a_begin + len(a.rhs)

        if need_b:
            need_b = False
            try:
                b = next(hs_b)
            except StopIteration:
                if a is None:
                    break
                b = None
                b_begin = INFINITY
                b_pre = ()
                b_post = ()
                b_end = INFINITY
            else:
                b_pre = b.pre_context
                b_post = b.post_context
                b_begin = b.lhs_line + len(b_pre)
                b_end = b_begin + len(b.lhs)

        if a_end <= b_begin:
            gap = b_begin - a_end

            if (n := len(a_post) + len(b_pre) - gap) > 0:
                assert (a_pre + a.rhs + a_post)[-n:] == (b_pre + b.lhs + b_post)[:n]
                if (n := len(a_post) - gap) > 0:
                    a_post = a_post[:gap] + (b.rhs + b_post)[:n]
                if (n := len(b_pre) - gap) > 0:
                    b_pre = (a_pre + a.lhs)[-n:] + b_pre[n:]

            wa(Hunk(
                lhs_line = a.lhs_line + delta_b,
                rhs_line = a.rhs_line + delta_b,
                pre_context = a_pre,
                lhs = a.lhs,
                rhs = a.rhs,
                post_context = a_post,
            ))
            delta_a += len(a.rhs) - len(a.lhs)
            need_a = True
            continue

        if b_end <= a_begin:
            gap = a_begin - b_end

            if (n := len(b_post) + len(a_pre) - gap) > 0:
                assert (b_pre + b.lhs + b_post)[-n:] == (a_pre + a.rhs + a_post)[:n]
                if (n := len(b_post) - gap) > 0:
                    b_post = b_post[:gap] + (a.lhs + a_post)[:n]
                if (n := len(a_pre) - gap) > 0:
                    a_pre = (b_pre + b.rhs)[-n:] + a_pre[n:]

            wb(Hunk(
                lhs_line = b.lhs_line - delta_a,
                rhs_line = b.rhs_line - delta_a,
                pre_context = b_pre,
                lhs = b.lhs,
                rhs = b.rhs,
                post_context = b_post,
            ))
            delta_b += len(b.rhs) - len(b.lhs)
            need_b = True
            continue

        raise ConflictError(a, b)

    return (*rb,), (*ra,)

class FileDiff:
    __slots__ = ('src_name', 'dst_name', 'src_mode', 'dst_mode', 'hunks',)

    def __init__(self, /, *, src_name, dst_name, src_mode, dst_mode, hunks):
        assert (src_mode is ...) == (dst_mode is ...)
        assert (src_name is None) == (src_mode is None)
        assert (dst_name is None) == (dst_mode is None)

        self.src_name = src_name
        self.dst_name = dst_name
        if src_mode == dst_mode:
            self.src_mode = ...
            self.dst_mode = ...
        else:
            self.src_mode = src_mode
            self.dst_mode = dst_mode
        self.hunks = hunks

    @property
    def op_kind_name(self):
        if self.src_name is None:
            return 'create'
        if self.dst_name is None:
            return 'delete'
        is_chmod = self.dst_mode != self.dst_mode
        if self.dst_name != self.dst_name:
            return 'chmod+rename' if is_chmod else 'rename'
        if self.hunks:
            return 'chmod+modify' if is_chmod else 'modify'
        return 'chmod' if is_chmod else 'noop'

    def write(self, w):
        src_name = self.src_name
        dst_name = self.dst_name

        if src_name is None:
            assert dst_name is not None
            w(f'diff --git a/{dst_name} b/{dst_name}\n'.encode('utf-8'))
        elif dst_name is None:
            w(f'diff --git a/{src_name} b/{src_name}\n'.encode('utf-8'))
        else:
            w(f'diff --git a/{src_name} b/{dst_name}\n'.encode('utf-8'))

        if src_name is not None and dst_name is not None and src_name != dst_name:
            w(f'rename from {src_name}\nrename to {dst_name}\n'.encode('utf-8'))

        m = self.src_mode
        if dst_name is None:
            assert m is not None
            w(f'deleted file mode {m}\n'.encode('utf-8'))
        elif m not in (None, ...):
            w(f'old mode {m}\n'.encode('utf-8'))

        m = self.dst_mode
        if src_name is None:
            assert m is not None
            w(f'new file mode {m}\n'.encode('utf-8'))
        elif m not in (None, ...):
            w(f'new mode {m}\n'.encode('utf-8'))

        h = self.hunks
        if h:
            if src_name is None:
                src_name = '/dev/null'
                lhs_line = 0
            else:
                src_name = f'a/{src_name}'
                lhs_line = None
            if dst_name is None:
                dst_name = '/dev/null'
                rhs_line = 0
            else:
                dst_name = f'b/{dst_name}'
                rhs_line = None

            w(f'--- {src_name}\n+++ {dst_name}\n'.encode('utf-8'))

            for h in h:
                h.write(w, lhs_line=lhs_line, rhs_line=rhs_line)

    def __neg__(self, /):
        tp = type(self)

class Diff:
    __slots__ = ('files',)

    def __init__(self):
        self.files = ()

    def __bool__(self):
        return not not self.files

    @classmethod
    def _sort_files(tp, files):
        r = object.__new__(tp)
        r.files = (*sorted(files, key=lambda f: (f.src_name or '', f.dst_name or '')),)
        return r

    @classmethod
    def parse_l(tp, l, line):
        files = []

        while True:
            l = l.decode('utf-8').rstrip()
            if l == '--' or not l:
                break
            m = rem_diffgit(l)
            if m is None:
                raise ValueError(f'Invalid patch: expected \'diff --git ...\' header, got {l!r}')

            header = l

            src_name,dst_name = m.groups()
            src_exists = ...
            dst_exists = ...

            if not src_exists and not dst_exists:
                raise RuntimeError('wtf')
            if (not src_exists or not dst_exists) and src_name != dst_name:
                raise RuntimeError('wtf')

            has_diff = False
            has_chmod = False
            has_rename = False
            dst_mode = ...
            src_mode = ...

            while True:
                l = line()
                if l.startswith(b'--- a/'):
                    if src_exists is False: raise RuntimeError('wtf')
                    src_exists = True
                    if l[6:].decode('utf-8') != src_name:
                        raise RuntimeError('wtf')
                    has_diff = True
                    continue
                if l == b'--- /dev/null':
                    if src_exists is True: raise RuntimeError('wtf')
                    src_exists = False
                    has_diff = True
                    continue
                if l.startswith(b'+++ b/'):
                    if dst_exists is False: raise RuntimeError('wtf')
                    dst_exists = True
                    if l[6:].decode('utf-8') != dst_name:
                        raise RuntimeError('wtf')
                    has_diff = True
                    continue
                if l == b'+++ /dev/null':
                    if dst_exists is True: raise RuntimeError('wtf')
                    dst_exists = False
                    has_diff = True
                    continue
                if l.startswith(b'index '):
                    continue
                if l.startswith(b'new file mode '):
                    if src_exists is True: raise RuntimeError('wtf')
                    if dst_exists is False: raise RuntimeError('wtf')
                    src_exists = False
                    dst_exists=True
                    dst_mode = l[14:].decode('ascii')
                    src_mode = None
                    continue
                if l.startswith(b'new mode '):
                    dst_mode = l[9:].decode('ascii')
                    has_chmod = True
                    continue
                if l.startswith(b'deleted file mode '):
                    if src_exists is False: raise RuntimeError('wtf')
                    if dst_exists is True: raise RuntimeError('wtf')
                    src_exists = True
                    dst_exists = False
                    src_mode = l[18:].decode('ascii')
                    dst_mode = None
                    continue
                if l.startswith(b'old mode '):
                    if dst_exists is False:
                        raise RuntimeError('File deleted')
                    src_mode = l[9:].decode('ascii')
                    has_chmod = True
                    dst_exists = True
                    continue
                if l.startswith(b'similarity index '):
                    has_rename = True
                    continue
                if l.startswith(b'rename from '):
                    has_rename = True
                    rename_from = l[12:].decode('utf-8')
                    continue
                if l.startswith(b'rename to '):
                    rename_to = l[10:].decode('utf-8')
                    has_rename = True
                    continue
                break

            if has_chmod or has_rename:
                if src_exists is False: raise RuntimeError('wtf')
                if dst_exists is False: raise RuntimeError('wtf')
                src_exists = True
                dst_exists = True

            if has_chmod:
                assert src_mode is not ...
                assert dst_mode is not ...

            assert src_exists is not ...
            if not src_exists:
                assert dst_mode is not ...

            assert dst_exists is not ...
            if not dst_exists:
                assert src_mode is not ...

            if has_rename:
                if src_name == dst_name or not src_exists or not dst_exists:
                    raise RuntimeError(f'Bad rename: {src_name!r} -> {dst_name!r}')
                assert rename_from == src_name
                assert rename_to == dst_name

            if not src_exists: src_name = None
            if not dst_exists: dst_name = None

            assert (src_mode is ...) == (dst_mode is ...)
            assert has_diff or has_chmod or has_rename or not dst_exists

            #want_header = fmt_diff_git(src_name, dst_name)
            #if header != want_header:
            #    raise ValueError(f'Invalid patch: expected {want_header!r} header, got {header!r}')

            if has_diff:
                hunks = []

                while True:
                    l,hunk = Hunk.parse_l(l, line)
                    hunk = hunk.reduce_context()
                    if hunk is not None:
                        hunks.append(hunk)
                    if not l.startswith(b'@@ '):
                        break

                if src_name is None:
                    assert len(hunks) == 1
                    assert len(hunks[0].pre_context) == 0
                    assert len(hunks[0].post_context) == 0
                    assert len(hunks[0].lhs) == 0
                    assert hunks[0].lhs_line == 1
                    hunks[0].lhs_line = 1

                if dst_name is None:
                    assert len(hunks) == 1
                    assert len(hunks[0].pre_context) == 0
                    assert len(hunks[0].post_context) == 0
                    assert len(hunks[0].rhs) == 0
                    assert hunks[0].rhs_line == 1

                hunks = (*hunks,)

            else:
                hunks = ()

            files.append(FileDiff(
                src_name = src_name,
                dst_name = dst_name,
                src_mode = src_mode,
                dst_mode = dst_mode,
                hunks = hunks,
            ))

        try:
            l = line().rstrip()
        except StopIteration:
            pass
        else:
            if l[0] not in b'0123456789':
                l = l.decode('utf-8')
                raise ValueError('Invalid patch: expected git version, got {l!r}')


            while True:
                try:
                    l = line()
                except StopIteration:
                    break
                if l.strip():
                    l = l.decode('utf-8')
                    raise ValueError('Invalid patch: got trailing garbage {l!r}')

        return tp._sort_files(files)

    @classmethod
    def parse(tp, body):
        line = iter(body.split(b'\n')).__next__
        return tp.parse_l(line(), line)

    def write(self, w, /):
        for f in self.files:
            f.write(w)

    def write_munged(self, w, /):
        for f in self.files:
            w((f.src_name or '').encode('utf-8'))
            w(b'\n')
            w((f.dst_name or '').encode('utf-8'))
            w(b'\n')
            for h in f.hunks:
                h.write(w, True)
            w(b'\n')

    def __neg__(self):
        return Diff._sort_files((*(FileDiff(
            src_name = f.dst_name,
            dst_name = f.src_name,
            src_mode = f.dst_mode,
            dst_mode = f.src_mode,
            hunks = (*(Hunk(
                lhs_line = h.rhs_line,
                rhs_line = h.lhs_line,
                pre_context = h.pre_context,
                lhs = h.rhs,
                rhs = h.lhs,
                post_context = h.post_context,
            ) for h in f.hunks),),
        ) for f in self.files),))

    def __add__(a, b):
        tp = type(a)
        if type(b) is not tp:
            raise TypeError

        r = []
        w = r.append

        fdiff = FileDiff
        def join(fa, fb):
            src_name = fa.src_name
            dst_name = fb.dst_name
            src_mode = fa.src_mode
            dst_mode = fb.dst_mode
            if src_mode is ...:
                if dst_mode is not ...:
                    src_mode = fb.src_mode
            elif dst_mode is ...:
                dst_mode = fa.dst_mode
            elif fa.dst_mode != fb.src_mode:
                raise ValueError('mismatch')

            hunks = (*join_hunks(fa.hunks, fb.hunks),)

            if src_name is None and dst_name is None:
                if hunks:
                    raise ValueError('mismatch')
                return

            if src_name == dst_name and src_mode == dst_mode and not hunks:
                return

            w(fdiff(
                src_name = src_name,
                dst_name = dst_name,
                src_mode = src_mode,
                dst_mode = dst_mode,
                hunks = hunks,
            ))

        _pair_diffs(a, b, on_lhs=w, on_rhs=w, on_pair=join, on_re=join)
        return tp._sort_files(r)

    @classmethod
    def commute(tp, a, b):
        if type(a) is not tp or type(b) is not tp:
            raise TypeError

        ra = []
        rb = []
        wa = ra.append
        wb = rb.append
        fdiff = FileDiff

        def join(fa, fb):
            name1 = fa.src_name
            name2 = fa.dst_name
            name3 = fb.dst_name
            if name1 == name2:
                name2 = name3
            elif name2 == name3:
                name2 = name1
            else:
                raise ValueError('cannot commute {name2!r}: {fa.op_kind_name} / {fb.op_kind_name}')

            mode1 = fa.src_mode
            mode3 = fb.dst_mode
            if mode1 is ...:
                if mode3 is ...:
                    mode2 = ...
                else:
                    mode1 = fb.src_mode
                    mode2 = mode3

            elif mode3 is ...:
                mode3 = fa.dst_mode
                mode2 = mode1

            else:
                mode2 = fb.dst_mode
                if fa.src_mode != mode2:
                    raise ValueError('mismatch')
                if mode1 == mode2:
                    mode2 = mode3
                elif mode2 == mode3:
                    mode2 = mode1
                else:
                    raise ValueError('cannot commute {name2!r}: chmod/chmod conflict')

            try:
                hb,ha = commute_hunks(fa.hunks, fb.hunks)
            except PatchError as err:
                err.args = (name1, name2, name3)
                raise

            wa(fdiff(
                src_name = name1,
                dst_name = name2,
                src_mode = mode1,
                dst_mode = mode2,
                hunks = ha,
            ))
            wb(fdiff(
                src_name = name2,
                dst_name = name3,
                src_mode = mode2,
                dst_mode = mode3,
                hunks = hb,
            ))

        _pair_diffs(a, b, on_lhs=wa, on_rhs=wb, on_re=join, on_pair=join)
        return tp._sort_files(rb), tp._sort_files(ra)

def _pair_diffs(a, b, *, on_lhs, on_rhs, on_pair, on_re):
    a_del = {}
    b_new = {}

    a_dst = {}
    b_dst = set()
    for f in a.files:
        dst = f.dst_name
        if dst is None:
            a_del[f.src_name] = f
            continue
        a_dst[dst] = f

    a = a_dst.pop
    for f in b.files:
        dst = f.dst_name
        if dst is not None:
            b_dst.add(dst)

        src = f.src_name
        if src is None:
            b_new[dst] = f
            continue

        try:
            f2 = a(src)
        except KeyError:
            pass
        else:
            on_pair(f2, f)
            continue
        on_rhs(f)

    for dst in a_dst.keys() & b_dst:
        raise ValueError(f'mismatch for {dst!r}')

    b = b_new.pop
    for src,f in a_del.items():
        try:
            f2 = b(src)
        except KeyError:
            pass
        else:
            on_re(f, f2)
            continue
        on_lhs(f)

    for f in b_new.values():
        on_rhs(f)

    for f in a_dst.values():
        on_lhs(f)