diff --git a/patch2vimedit/hunk_changes.py b/patch2vimedit/hunk_changes.py
new file mode 100644
index 0000000..c61ea3e
--- /dev/null
+++ b/patch2vimedit/hunk_changes.py
@@ -0,0 +1,180 @@
+""" Tools to make smart edition to apply a hunk """
+
+from collections import OrderedDict
+
+
+class Levenshtein:
+    cache = {}
+    moves = {
+        "D": (-1, 0),
+        "S": (-1, -1),
+        "I": (0, -1),
+    }
+
+    def __init__(self, pre, post):
+        self.pre = pre
+        self.post = post
+        self.work_matrix = None
+        self.op_matrix = None
+
+    @classmethod
+    def get_full_cache(cls):
+        return cls.cache
+
+    def cache_key(self):
+        if isinstance(self.pre, list):
+            return (tuple(self.pre), tuple(self.post))
+        return (self.pre, self.post)
+
+    def getcache(self):
+        return self.get_full_cache().get(self.cache_key(), None)
+
+    @classmethod
+    def clean_cache(cls):
+        cls.cache = {}
+
+    def insert_cost(self, to_insert, pre_idx, post_idx):
+        return 1
+
+    def del_cost(self, to_delete, pre_idx, post_idx):
+        return 1
+
+    def subst_cost(self, subst_from, subst_to, pre_idx, post_idx):
+        return int(subst_from != subst_to)
+
+    @staticmethod
+    def _argmin(**kwargs):
+        """ Returns the first encountered pair (k, v) with lowest v wrt. `<`. """
+        if not kwargs:
+            raise Exception("No arguments")
+        argmin = next(iter(kwargs))
+        valmin = kwargs[argmin]
+
+        for key in kwargs:
+            val = kwargs[key]
+            if val < valmin:
+                argmin = key
+                valmin = val
+        return (argmin, valmin)
+
+    def compute(self):
+        cached = self.getcache()
+        if cached:
+            return cached
+
+        self.op_matrix = [[None] + ["I"] * len(self.post)] + [
+            ["D"] + [None] * (len(self.post)) for _ in range(len(self.pre))
+        ]
+        self.work_matrix = [
+            [0] * (len(self.post) + 1) for _ in range(len(self.pre) + 1)
+        ]
+
+        for post_idx in range(len(self.post)):
+            self.work_matrix[0][post_idx + 1] = self.work_matrix[0][
+                post_idx
+            ] + self.insert_cost(self.post[post_idx], 0, post_idx + 1)
+        for pre_idx in range(len(self.pre)):
+            self.work_matrix[pre_idx + 1][0] = self.work_matrix[pre_idx][
+                0
+            ] + self.del_cost(self.pre[pre_idx], pre_idx + 1, 0)
+
+        for pre_idx in range(len(self.pre)):
+            for post_idx in range(len(self.post)):
+                c_insert_cost = self.insert_cost(
+                    self.post[post_idx], pre_idx + 1, post_idx + 1
+                )
+                c_del_cost = self.del_cost(self.pre[pre_idx], pre_idx + 1, post_idx + 1)
+                c_subst_cost = self.subst_cost(
+                    self.pre[pre_idx], self.post[post_idx], pre_idx + 1, post_idx + 1
+                )
+
+                opmin, costmin = self._argmin(
+                    **OrderedDict(
+                        [
+                            ("D", self.work_matrix[pre_idx][post_idx + 1] + c_del_cost),
+                            (
+                                "I",
+                                self.work_matrix[pre_idx + 1][post_idx] + c_insert_cost,
+                            ),
+                            ("S", self.work_matrix[pre_idx][post_idx] + c_subst_cost),
+                        ]
+                    )
+                )
+
+                self.work_matrix[pre_idx + 1][post_idx + 1] = costmin
+                self.op_matrix[pre_idx + 1][post_idx + 1] = opmin
+                self.cur_mode = opmin
+
+        ops = []
+        pre_pos = len(self.pre)
+        post_pos = len(self.post)
+        while pre_pos > 0 or post_pos > 0:
+            cur_op = self.op_matrix[pre_pos][post_pos]
+            cost_pre = pre_pos
+            cost_post = post_pos
+            row_m, col_m = self.moves[cur_op]
+            pre_pos += row_m
+            post_pos += col_m
+            if cur_op == "S":
+                pre_val = self.pre[pre_pos]
+                post_val = self.post[post_pos]
+                ops.append(
+                    (
+                        cur_op,
+                        (pre_val, post_val),
+                        self.subst_cost(pre_val, post_val, cost_pre, cost_post),
+                    )
+                )
+            if cur_op == "I":
+                post_val = self.post[post_pos]
+                ops.append(
+                    (cur_op, post_val, self.insert_cost(post_val, cost_pre, cost_post))
+                )
+            if cur_op == "D":
+                pre_val = self.pre[pre_pos]
+                ops.append(
+                    (cur_op, pre_val, self.del_cost(pre_val, cost_pre, cost_post))
+                )
+
+        cached_result = {
+            "count": self.work_matrix[-1][-1],
+            "ops": ops[::-1],
+        }
+
+        self.get_full_cache()[self.cache_key()] = cached_result
+
+        return cached_result
+
+
+class InlineLevenshtein(Levenshtein):
+    """ Levenshtein distance for edition of a single line """
+
+    def insert_cost(self, to_insert, pre_idx, post_idx):
+        return 1
+
+    def del_cost(self, to_delete, pre_idx, post_idx):
+        return 1  # 'x'
+
+    def subst_cost(self, subst_from, subst_to, pre_idx, post_idx):
+        if subst_from == subst_to:
+            return 0
+        return 1
+
+
+class HunkLevenshtein(Levenshtein):
+    """ Levenshtein distance for hunk edition, ie. should we delete a whole line,
+    insert a whole line or substitute one line with another """
+
+    def insert_cost(self, to_insert, pre_idx, post_idx):
+        indent = 0
+        while indent < len(to_insert) and to_insert[indent] == " ":
+            indent += 1
+        indent_cost = 3 if indent else 0
+        return indent_cost + len(to_insert.strip()) + 1
+
+    def del_cost(self, to_delete, pre_idx, post_idx):
+        return 2
+
+    def subst_cost(self, subst_from, subst_to, pre_idx, post_idx):
+        res = InlineLevenshtein(subst_from, subst_to).compute()
+        return res["count"]