diff --git a/bsp_tool/extensions/diff/__init__.py b/bsp_tool/extensions/diff/__init__.py
index 671ca487..d1e9e924 100644
--- a/bsp_tool/extensions/diff/__init__.py
+++ b/bsp_tool/extensions/diff/__init__.py
@@ -1,191 +1,130 @@
-"""Run with 64-bit python! Respawn .bsp files are large!"""
 import difflib
-import io
-import itertools
-import re
-from typing import Dict, Iterable, List
-
-
-r1_dir = "E:/Mod/Titanfall/maps"
-r1o_dir = "E:/Mod/TitanfallOnline/maps"
-r2_dir = "E:/Mod/Titanfall2/maps"
-
-shared_maps = [("mp_angel_city", "mp_angel_city"),
-               ("mp_colony", "mp_colony02"),
-               ("mp_relic", "mp_relic02"),
-               ("mp_rise", "mp_rise"),
-               ("mp_wargames", "mp_wargames")]
-# ^ r1 map name, r2 map name
-
-
-def diff_bsps(bsp1, bsp2, full=False) -> str:
-    """WARNING: full diffs can be incredibly large!"""
-    out = []
-    if bsp1.folder == bsp2.folder:
-        out.append(f"Comparing {bsp1} -> {bsp2}...")
-    else:
-        out.append(f"Comparing {bsp1.folder}/{bsp1} -> {bsp2.folder}/{bsp2}...")
-    # NOTE: comparing lumps by index, same number of lumps expected
-    for lump1, lump2 in zip(bsp1.branch.LUMP, bsp2.branch.LUMP):
-        lump1 = lump1.name
-        lump2 = lump2.name
-        # diff headers
-        if lump1 not in bsp1.headers or lump2 not in bsp2.headers:
-            continue  # lazy fix for rbsp externals
-            # TODO: note absent headers (not just for respawn.ExternalLumpManager!)
-        bsp1_header = bsp1.headers[lump1]
-        bsp2_header = bsp2.headers[lump2]
-        lump_name = lump1 if lump1 == lump2 else f"{lump1} -> {lump2}"
-        # NOTE: fourCC (decompressed size) vs length is not calculated
-        # -- in fact, no check to check opposing compressed state (one compressed, one uncompressed)
-        # -- however, LZMA compressed lump contents are always decompressed before comparison
-        header_diff = "".join(["Y" if bsp1_header.offset == bsp2_header.offset else "N",
-                               "Y" if bsp1_header.length == bsp2_header.length else "N",
-                               "Y" if bsp1_header.version == bsp2_header.version else "N",
-                               "Y" if bsp1_header.fourCC == bsp2_header.fourCC else "N"])
-        # diff lump contents
-        try:
-            lump_1_contents = bsp1.lump_as_bytes(lump1)
-            lump_2_contents = bsp2.lump_as_bytes(lump2)
-        except Exception as exc:
-            out.append(f"{lump_name}  {header_diff} ???? {exc}")
-            continue  # skip this lump
-        lumps_match = bool(lump_1_contents == lump_2_contents)
-        contents_diff = "YES!" if lumps_match else "NOPE"
-        out.append(f"{lump_name}  {header_diff} {contents_diff}")
-        # was a lump removed / added?
-        if (len(lump_1_contents) == 0 or len(lump_2_contents) == 0) and not lumps_match:
-            out.append(" ".join(["+" if hasattr(bsp1, lump1) else "-", f"{bsp1.filename}.{lump1}"]))
-            out.append(" ".join(["+" if hasattr(bsp2, lump2) else "-", f"{bsp2.filename}.{lump2}"]))
-        # detailed comparisons
-        elif full:
-            if not lumps_match:
-                # TODO: measure the scale of the differences
-                if lump1 in bsp1.branch.LUMP_CLASSES and lump2 in bsp2.branch.LUMP_CLASSES:
-                    diff = difflib.unified_diff([lc.__repr__() for lc in getattr(bsp1, lump1)],
-                                                [lc.__repr__() for lc in getattr(bsp2, lump2)],
-                                                f"{bsp1.filename}.{lump1}",
-                                                f"{bsp1.filename}.{lump1}")
-                    out.extend(diff)
-                # SPECIAL_LUMP_CLASSES
-                elif all([ln == "ENTITIES" for ln in (lump1, lump2)]):
-                    out.append(diff_entities(bsp1.ENTITIES, bsp2.ENTITIES))
-                elif all([ln == "PAKFILE" for ln in (lump1, lump2)]):
-                    # NOTE: this will fail on nexon.cso2 bsps, as their pakfiles are unmapped
-                    out.append(diff_pakfiles(bsp1, bsp2))
-                # TODO: GAME_LUMP diff model_names
-                else:  # BASIC_LUMP_CLASSES / general raw bytes
-                    # NOTE: xxd line numbers prevent accurately tracing insertions
-                    # TODO: set xxd width to cover LumpClass._format, 1 entry per line
-                    # -- if formats don't match, give an option to skip
-                    diff = difflib.context_diff(xxd(io.BytesIO(lump_1_contents)),
-                                                xxd(io.BytesIO(lump_2_contents)),
-                                                f"{bsp1.filename}.{lump1}",
-                                                f"{bsp2.filename}.{lump2}")
-                    # TODO: run xxd without creating line numbers
-                    # -- then, generate line numbers from diff & update diff with these line numbers
-                    out.extend(diff)
+from typing import Any, Dict, Generator, List
+
+from . import base
+from . import shared
+
+from bsp_tool import branches
+from bsp_tool.base import Bsp
+from bsp_tool.lumps import BasicBspLump, RawBspLump, ExternalRawBspLump
+
+
+def diff_lumps(old_lump: Any, new_lump: Any) -> base.Diff:
+    LumpClasses = set()
+    for lump in (old_lump, new_lump):
+        if issubclass(lump.__class__, BasicBspLump):
+            LumpClasses.add(lump.LumpClass)
+        else:  # SpecialLumpClass / RawBspLump
+            LumpClasses.add(lump.__class__)
+    # match LumpClasses to a base.Diff subclass
+    # TODO: mismatched lump type diffs (substitute defaults for alternate versions?)
+    # -- should only be used for extremely similar lumps
+    if len(LumpClasses) > 1:
+        # AbridgedDiff?
+        raise NotImplementedError("Cannot diff lumps of differring LumpClass")
+    if LumpClasses == {branches.shared.Entities}:
+        DiffClass = shared.EntitiesDiff
+    elif LumpClasses == {branches.shared.PakFile}:
+        DiffClass = shared.PakFileDiff
+    elif RawBspLump in LumpClasses or ExternalRawBspLump in LumpClasses:
+        # TODO: core.xxd diff
+        raise NotImplementedError("Cannot diff raw lumps")
+    # if all([issubclass(lc, branches.base.BitField) for lc in LumpClasses]):
+    #     DiffClass = base.BitFieldDiff
+    # if all([issubclass(lc, branches.base.MappedArray) for lc in LumpClasses]):
+    #     DiffClass = base.MappedArrayDiff
+    # if all([issubclass(lc, branches.base.Struct) for lc in LumpClasses]):
+    #     DiffClass = base.StructDiff
+    else:  # default
+        DiffClass = base.Diff
+    return DiffClass(old_lump, new_lump)
+
+
+class BspDiff:
+    """deferred diffs of lumps & headers etc."""
+    old: Bsp
+    new: Bsp
+
+    def __init__(self, old: Bsp, new: Bsp):
+        if old.branch != new.branch:
+            raise NotImplementedError("Cannot diff bsps from different branches")
+        self.old = old
+        self.new = new
+        self.headers = HeadersDiff(old.headers, new.headers)
+        # NOTE: a change in header offsets does not imply a change in lump data
+        # TODO: other metadata (file magic, version, revision, signature etc.)
+
+    def __getattr__(self, lump_name: str) -> Any:
+        old_lump = getattr(self.old, lump_name, None)
+        new_lump = getattr(self.new, lump_name, None)
+        no_old_lump = old_lump is None
+        no_new_lump = new_lump is None
+        if no_old_lump and no_new_lump:
+            raise AttributeError(f"Neither bsp has {lump_name} lump to be diffed")
+        elif no_old_lump or no_new_lump:
+            return NoneDiff(old_lump, new_lump)
         else:
-            out.extend([str(bsp1_header), str(bsp2_header)])
-    return "\n".join(out)
-
-
-def diff_rbsps(rbsp1, rbsp2, external=True, full=False) -> str:
-    """compare internal to external lumps with diff_rbsps(bsp, bsp.external, external=False)"""
-    out = ["*** .bsp files ***", diff_bsps(rbsp1, rbsp2, full)]
-    # NOTE: could confirm ent_types against ENTITY_PARTITION lump
-    # -- however respawn seems to always use every .ent, leaving the script file empty if unused
-    # -- this makes ENTITY_PARTITION practically useless, as it never changes
-    out.append("*** .ent files ***")
-    for ent_type in ("env", "fx", "script", "snd", "spawn"):
-        ent_lump = f"ENTITIES_{ent_type}"
-        lump1 = getattr(rbsp1, ent_lump, list())
-        lump2 = getattr(rbsp2, ent_lump, list())
-        ents_match = "YES!" if lump1 == lump2 else "NOPE"
-        out.append(f"{ent_lump}  {ents_match}")
-        if full and ents_match == "NOPE":
-            out.append(diff_entities(lump1, lump2))
-    if external:
-        out.append("*** .bsp_lump files ***")
-        out.append(diff_bsps(rbsp1.external, rbsp2.external, full))
-        # TODO: close each lump after reading to save memory & avoid the "Too many open files" OSError
-    return "\n".join(out)
-
-
-EntityLump = List[Dict[str, str]]
-# ^ [{"key": "value"}]
-
-
-def diff_entities(lump1: EntityLump, lump2: EntityLump) -> str:
-    out = []
-    for i, e1, e2 in zip(itertools.count(), lump1, lump2):
-        if e1 != e2:
-            out.extend([f"Entity #{i}", "  {"])
-            # TODO: be a little dynamic to make sure keys align
-            # -- otherwise many false negatives might appear in a relatively simple diff
-            for k1, k2, v1, v2 in zip(e1.keys(), e2.keys(), e1.values(), e2.values()):
-                if v1 != v2:
-                    out.extend([f'-   "{k1}" "{v1}"',
-                                f'+   "{k2}" "{v2}"'])
-                else:
-                    out.append(f'    "{k1}" "{v1}"')
-            out.append("  }")
-    return "\n".join(out)
-
-
-def diff_pakfiles(bsp1, bsp2) -> str:
-    """Works on any ValveBsp based .bsp (except CS:O2)"""
-    out = []
-    pak1_files = bsp1.PAKFILE.namelist()
-    pak2_files = bsp2.PAKFILE.namelist()
-    for filename in pak1_files:
-        absent = filename not in pak2_files
-        out.append(f"- {filename}" if absent else f"  {filename}")
-        if not absent:
-            file1 = bsp1.PAKFILE.read(filename)
-            file2 = bsp2.PAKFILE.read(filename)
-            if file1 == file2:  # skip matches
-                continue
-            out[-1] = f"~ {filename}"
-            out.extend(difflib.context_diff(xxd(io.BytesIO(file1)),
-                                            xxd(io.BytesIO(file2)),
-                                            f"{bsp1.filename}.PAKFILE.{filename}",
-                                            f"{bsp2.filename}.PAKFILE.{filename}"))
-    out.extend([f"+ {f}" for f in pak2_files if f not in pak1_files])
-    return "\n".join(out)
-
-
-# binary diff helpers
-def split(iterable: Iterable, chunk_size: int) -> Iterable:
-    for i, _ in enumerate(iterable[::chunk_size]):
-        yield iterable[i * chunk_size:(i + 1) * chunk_size]
-
-
-def xxd(data: io.BytesIO, width: int = 16) -> str:
-    """view a binary file like with a certain hex editor"""
-    out = list()
-    allowed_chars = re.compile(r"[a-zA-Z0-9/\\]")
-    i, bytes_ = 0, data.read(width)
-    while bytes_ != b"":
-        address = f"0x{i * width:08X}"
-        hex_ = " ".join([f"{b:02X}" for b in bytes_])
-        if len(hex_) < 3 * width:  # pad last line of hex with spaces
-            hex_ += " " * (3 * width - len(hex_))
-        ascii_ = "".join([chr(b) if allowed_chars.match(chr(b)) else "." for b in bytes_])
-        out.append(f"{address}:  {hex_}  {ascii_}\n")
-        i, bytes_ = i + 1, data.read(width)
-    return out
-
-
-if __name__ == "__main__":
-    import os
-    import sys
-    sys.path.insert(0, r"C:\Users\Jared\Documents\GitHub\bsp_tool")
-    import bsp_tool  # run from top-level
-
-    for r1_map, r2_map in shared_maps:
-        with open(f"{r1_map}.diff", "w") as log_file:
-            print(f"Writing {r1_map}.diff ...")
-            r1_bsp = bsp_tool.load_bsp(os.path.join(r1_dir, f"{r1_map}.bsp"))
-            r2_bsp = bsp_tool.load_bsp(os.path.join(r2_dir, f"{r2_map}.bsp"))
-            log_file.write(diff_rbsps(r1_bsp, r2_bsp))
+            diff = diff_lumps(old_lump, new_lump)
+            setattr(self, lump_name, diff)  # cache
+            return diff
+
+    def save(self, base_filename: str, log_mode: base.LogMode = base.LogMode.VERBOSE):
+        """generate & save .diff files"""
+        # for each lump (match by name)
+        # filename.lump.00.ENTITIES.diff: old_goldsrc.ENTITIES (0) -> new_blue_shift.ENTITIES (1)
+        # filename.lump.01.PLANES.diff: old_goldsrc.PLANES (1) -> new_blue_shift.PLANES (0)
+        # RespawnBsp
+        # -- filename.ENTITITES.fx.diff: filename_fx.ent
+        # -- filename.lump.00XX.LUMP_NAME.diff
+        # -- filename.lump.00XX.LUMP_NAME.bsp_lump.diff
+        # filename.bsp.diff: headers & Y/N lump matches
+        raise NotImplementedError()
+
+
+class NoneDiff(base.Diff):
+    """for diffing against None"""
+    def short_stats(self) -> str:
+        brand_new = self.old is None
+        assert brand_new or self.new is None
+        if brand_new:
+            return f"{len(self.new)} insertions(+)"
+        else:
+            return f"{len(self.old)} deletions(-)"
+
+    def unified_diff(self) -> List[str]:
+        return [self.short_stats()]
+
+
+class HeadersDiff(base.Diff):
+    # TODO: support comparisons between different branches
+    # TODO: how do we communicate a change in branch order?
+    # -- modern_warfare lump order & count is unique
+    # -- will probably need it's own class
+    old: Dict[str, Any]
+    new: Dict[str, Any]
+    _cache = Dict[str, List[str]]
+    # NOTE: changes on offset can be knock on affect of changes to an earlier lump
+
+    def __init__(self, old: Dict[str, Any], new: Dict[str, Any]):
+        super().__init__(old, new)
+        self._cache = dict()
+
+    def __getitem__(self, lump_name: str) -> str:
+        if lump_name not in {*self.old, *self.new}:
+            raise KeyError(f"No {lump_name} header to diff")
+        diff = self._cache.get(lump_name)
+        if diff is None:
+            old = f"{lump_name} {self.old[lump_name]!r}\n"
+            new = f"{lump_name} {self.new[lump_name]!r}\n"
+            diff = list(difflib.unified_diff([old], [new]))
+            self._cache[lump_name] = diff
+        return diff
+
+    def short_stats(self) -> str:
+        raise NotImplementedError()
+        # TODO: how to summarise?
+
+    def unified_diff(self) -> Generator[str, None, None]:
+        for lump_name in self.old:
+            for line in self[lump_name]:
+                yield line
diff --git a/bsp_tool/extensions/diff/base.py b/bsp_tool/extensions/diff/base.py
new file mode 100644
index 00000000..1270f3c5
--- /dev/null
+++ b/bsp_tool/extensions/diff/base.py
@@ -0,0 +1,53 @@
+import difflib
+import enum
+from typing import Generator, Iterable, List
+
+
+class LogMode(enum.Enum):
+    FAST = 0  # no diff
+    VERBOSE = 1  # small diff
+    VERY_VERBOSE = 2  # maximum diff
+
+
+class Diff:
+    old: Iterable[object]
+    new: Iterable[object]
+
+    def __init__(self, old: Iterable[object], new: Iterable[object]):
+        self.old = old
+        self.new = new
+
+    def has_no_changes(self) -> bool:
+        return self.old == self.new
+
+    def as_text(self, log_mode=LogMode.VERBOSE) -> Generator[str, None, None]:
+        """formatted diff text, one line at a time"""
+        if log_mode == LogMode.VERBOSE:
+            yield self.short_stats()
+        elif log_mode == LogMode.VERY_VERBOSE:  # GENERATES A LOT OF TEXT!
+            for line in self.unified_diff():
+                yield line
+        else:  # only raised if try to pull data from the generator
+            raise NotImplementedError(f"Unexpected Log Mode: {log_mode}")
+
+    def short_stats(self) -> str:
+        """mimick git diff --shortstat"""
+        old = set(self.old)
+        new = set(self.new)
+        return f"{new.difference(old)} insertions(+) {old.difference(new)} deletions(-)"
+
+    def unified_diff(self) -> List[str]:
+        """quick & dirty diff of __repr__"""
+        # NOTE: if the __repr__ is "<Classname @ 0xMEMORYADDRESS>" equality cannot be detemined
+        old = [repr(x) for x in self.old]
+        new = [repr(x) for x in self.new]
+        # TODO: metadata
+        # --- old.name
+        # +++ new.name
+        for line in difflib.unified_diff(old, new):
+            yield line
+
+
+# TODO: class BitFieldDiff(Diff):
+# TODO: class MappedArrayDiff(Diff):
+# TODO: class StructDiff(Diff):
diff --git a/bsp_tool/extensions/diff/core.py b/bsp_tool/extensions/diff/core.py
new file mode 100644
index 00000000..23eb5cef
--- /dev/null
+++ b/bsp_tool/extensions/diff/core.py
@@ -0,0 +1,19 @@
+import io
+import re
+
+
+def xxd(data: bytes, cols: int = 32, show_address: bool = False) -> str:
+    """view a binary file like with a certain hex editor"""
+    data = io.BytesIO(data)
+    allowed_chars = re.compile(r"[a-zA-Z0-9/\\]")
+    address, bytes_ = 0, data.read(cols)
+    while bytes_ != b"":
+        hex_ = " ".join([f"{b:02X}" for b in bytes_])
+        if len(hex_) < 3 * cols:  # last line needs padding
+            hex_ += " " * (3 * cols - len(hex_))
+        ascii_ = "".join([c if allowed_chars.match(c) else "." for c in map(chr, bytes_)])
+        if show_address:
+            yield f"0x{address:08X}:  {hex_}  {ascii_}\n"
+        else:
+            yield f"{hex_} {ascii_}\n"
+        address, bytes_ = address + cols, data.read(cols)
diff --git a/bsp_tool/extensions/diff/shared.py b/bsp_tool/extensions/diff/shared.py
new file mode 100644
index 00000000..787e26a0
--- /dev/null
+++ b/bsp_tool/extensions/diff/shared.py
@@ -0,0 +1,94 @@
+import difflib
+from typing import Dict, Generator, List
+
+from . import base
+from . import core
+
+
+class EntitiesDiff(base.Diff):
+    def short_stats(self) -> str:
+        # NOTE: ignoring lump order changes, but key-value order is still checked
+        old = {tuple(e.items()) for e in self.old}
+        new = {tuple(e.items()) for e in self.new}
+        added = len(new.difference(old))
+        removed = len(old.difference(new))
+        return f"{added} insertions(+) {removed} deletions(-)"
+
+    def unified_diff(self) -> Generator[str, None, None]:  # SLOW!
+        # TODO: meta-diff to align indices given insertions & deletions
+        # -- then come back in & abridge where helpful
+        # 3 reprs (inline: meta-diff, abridged: short diff line, multi-line: detailed diff)
+        for old_entity, new_entity in zip(self.old, self.new):
+            if not old_entity["classname"] == new_entity["classname"]:  # abridged
+                yield f"- {self.short_repr(old_entity)}\n"
+                yield f"+ {self.short_repr(new_entity)}\n"
+            elif old_entity != new_entity:  # likely to be a little similar
+                old = [f"{x}\n" for x in self.long_repr(old_entity)]
+                new = [f"{x}\n" for x in self.long_repr(new_entity)]
+                for line in difflib.Differ().compare(old, new):
+                    yield line
+            else:  # skip matching entities
+                continue
+        # quick leftovers
+        if len(self.old) > len(self.new):
+            for deleted in self.old[len(self.new):]:
+                yield f"- {self.short_repr(deleted)}\n"
+        elif len(self.new) > len(self.old):
+            for added in self.new[len(self.old):]:
+                yield f"+ {self.short_repr(added)}\n"
+
+    @staticmethod
+    def short_repr(entity: Dict[str, str]) -> str:
+        classname = entity["classname"]
+        origin = entity.get("origin", "")
+        model = entity.get("model", "")
+        if model.startswith("*"):  # brush entity
+            if origin != "":
+                return f"<{classname} (model {model}) @ {origin}>"
+            else:  # worldspawn?
+                return f"<{classname} (model {model})>"
+        else:  # point entity
+            if origin != "":
+                return f"<{classname} @ {origin}>"
+            else:
+                return f"<{classname}>"
+
+    @staticmethod
+    def long_repr(entity: Dict[str, str]) -> List[str]:
+        out = list()
+        for key, value in entity.items():
+            if not isinstance(value, list):  # duplicate key (Source Input/Output)
+                out.extend([f'"{key}" "{v}"' for v in value])
+            else:
+                out.append(f'"{key}" "{value}"')
+        return ["{", *out, "}"]
+
+
+class PakFileDiff:
+    """Works on any ValveBsp based .bsp (except CS:O2)"""
+
+    def shortstat(self) -> str:
+        """quick & dirty namelist check"""
+        old = set(self.old.namelist())
+        new = set(self.new.namelist())
+        return f"{new.difference(old)} insertions(+) {old.difference(new)} deletions(-)"
+
+    def unified_diff(self) -> Generator[str, None, None]:
+        old_filelist = [f"{fn}\n" for fn in self.old.namelist()]
+        new_filelist = [f"{fn}\n" for fn in self.new.namelist()]
+        meta_diff = difflib.ndiff(old_filelist, new_filelist)
+        for meta_line in meta_diff:
+            if meta_line.startswith(" "):  # maybe a match
+                filename = meta_line.lstrip()
+                old_file = self.old.read(filename)
+                new_file = self.new.read(filename)
+                if old_file == new_file:
+                    yield meta_line
+                else:
+                    yield f"# BEGIN {filename}"
+                    # NOTE: we could grab date_time from ZipInfo & format it, but lazy
+                    for line in difflib.unified_diff(core.xxd(old_file), core.xxd(new_file), [filename] * 2):
+                        yield line
+                    yield f"# END {filename}"
+            else:
+                yield meta_line
diff --git a/tests/extensions/diff/__init__.py b/tests/extensions/diff/__init__.py
new file mode 100644
index 00000000..e69de29b
diff --git a/tests/extensions/diff/test_BspDiff.py b/tests/extensions/diff/test_BspDiff.py
new file mode 100644
index 00000000..9b9ec07b
--- /dev/null
+++ b/tests/extensions/diff/test_BspDiff.py
@@ -0,0 +1,41 @@
+from bsp_tool import ValveBsp
+from bsp_tool.branches.valve import orange_box
+from bsp_tool.extensions import diff
+
+import pytest
+
+
+old_bsp = ValveBsp(orange_box, "tests/maps/Team Fortress 2/test_displacement_decompile.bsp")
+new_bsp = ValveBsp(orange_box, "tests/maps/Team Fortress 2/test_physcollide.bsp")
+# >>> old_lumps = {L for L, h in old_bsp.headers.items() if h.length > 0}
+# >>> new_lumps = {L for L, h in new_bsp.headers.items() if h.length > 0}
+# >>> old_lumps.difference(new_lumps)
+# {'DISPLACEMENT_TRIS', 'DISPLACEMENT_LIGHTMAP_SAMPLE_POSITIONS', 'DISPLACEMENT_VERTICES', 'DISPLACEMENT_INFO'}
+# >>> new_lumps.difference(old_lumps)
+# {'VISIBILITY'}
+
+
+class TestBspDiff:
+    diff = diff.BspDiff(old_bsp, new_bsp)
+    # TODO: diff.BspDiff(old_bsp, new_bsp) & verify no changes
+
+    def test_invalid_lump(self):
+        with pytest.raises(AttributeError):
+            print(self.diff.NONEXISTANT_LUMP)
+
+    def test_lump_added(self):
+        lump_diff = self.diff.VISIBILITY
+        assert isinstance(lump_diff, diff.NoneDiff)
+
+    def test_lump_removed(self):
+        lump_diff = self.diff.DISPLACEMENT_INFO
+        assert isinstance(lump_diff, diff.NoneDiff)
+
+
+# TODO: TestNoneDiff  (short_stats only)
+# TODO: TestDiffLumps  (assigning diff class)
+# -- branches.shared.Entities -> diff.shared.EntitiesDiff
+# -- branches.shared.PakFile -> diff.shared.PakFileDiff
+# -- branches.base.* -> diff.base.Diff
+# -- RawBspLump -> NotImplementedError
+# -- * -> diff.base.Diff
diff --git a/tests/extensions/diff/test_base.py b/tests/extensions/diff/test_base.py
new file mode 100644
index 00000000..af42467c
--- /dev/null
+++ b/tests/extensions/diff/test_base.py
@@ -0,0 +1,35 @@
+from bsp_tool.extensions import diff
+
+import pytest
+
+
+old = ["a few lines",
+       "i threw together",
+       "for a quick test"]
+new = ["A few lines I threw together",
+       "for a quick test"]
+sample = diff.base.Diff(old, new)
+
+
+class TestDiff:
+    def test_has_no_changes(self):
+        assert diff.base.Diff([0, 1], [0, 1]).has_no_changes() is True
+        assert diff.base.Diff([2, 3], [3, 4]).has_no_changes() is False
+
+    def test_as_text_FAST(self):
+        with pytest.raises(NotImplementedError):
+            for line in sample.as_text(diff.base.LogMode.FAST):
+                print(line)
+
+    def test_as_text_VERBOSE(self):
+        for i, line in enumerate(sample.as_text(diff.base.LogMode.VERBOSE)):
+            assert line == sample.short_stats()
+        assert i == 0, "more than one line emitted"
+
+    def test_as_text_VERY_VERBOSE(self):
+        diff_txt = sample.as_text(diff.base.LogMode.VERY_VERBOSE)
+        for line, expected in zip(diff_txt, sample.unified_diff()):
+            assert line == expected
+
+    # TODO: test_short_stats
+    # TODO: test_unified_diff
diff --git a/tests/extensions/diff/test_core.py b/tests/extensions/diff/test_core.py
new file mode 100644
index 00000000..bb01a9f3
--- /dev/null
+++ b/tests/extensions/diff/test_core.py
@@ -0,0 +1,36 @@
+import re
+
+from bsp_tool.extensions import diff
+
+import pytest
+
+
+class TestXXD:
+    # TODO: test show_address=False
+    @staticmethod
+    def expected_line(line_number: int, bytes_: bytes, cols: int = None):
+        cols = len(bytes_) if cols is None else cols
+        address = f"0x{line_number * cols:08X}"
+        hex_ = " ".join([f"{b:02X}" for b in bytes_])
+        if len(hex_) < 3 * cols:  # padding
+            hex_ += " " * (3 * cols - len(hex_))
+        allowed_chars = re.compile(r"[a-zA-Z0-9/\\]")
+        ascii_ = "".join([c if allowed_chars.match(c) else "." for c in map(chr, bytes_)])
+        return f"{address}:  {hex_}  {ascii_}\n"
+
+    @staticmethod
+    def split(bytes_: bytes, cols: int):
+        for i in range(0, len(bytes_), cols):
+            yield bytes_[i:i + cols]
+
+    samples = {"no padding needed": (b"Hello World!", 6),
+               "padding required": (b"Communist Manifesto", 10),
+               "every byte": (bytes(range(0xFF + 1)), 16)}
+
+    @pytest.mark.parametrize("sample,cols", samples.values(), ids=samples.keys())
+    def test_input(self, sample: bytes, cols: int):
+        xxd_line = diff.core.xxd(sample, cols=cols, show_address=True)
+        for i, bytes_ in enumerate(self.split(sample, cols)):
+            assert next(xxd_line) == self.expected_line(i, bytes_, cols)
+        with pytest.raises(StopIteration):
+            next(xxd_line)
diff --git a/tests/extensions/diff/test_shared.py b/tests/extensions/diff/test_shared.py
new file mode 100644
index 00000000..55c7c724
--- /dev/null
+++ b/tests/extensions/diff/test_shared.py
@@ -0,0 +1,30 @@
+import itertools
+
+from bsp_tool.extensions import diff
+
+
+old = [dict(classname="worldspawn"),
+       dict(classname="light", origin="0 0 0")]
+new = [dict(classname="worldspawn"),
+       dict(classname="info_player_start", origin="0 0 0"),
+       dict(classname="light", origin="0 0 64")]
+sample = diff.shared.EntitiesDiff(old, new)
+
+
+class TestEntitiesDiff:
+    def test_short_stats(self):
+        stats = sample.short_stats()
+        assert stats == "2 insertions(+) 1 deletions(-)"
+
+    def test_unified_diff(self):
+        # TODO: test a multiline diff
+        expected = ["- <light @ 0 0 0>\n",
+                    "+ <info_player_start @ 0 0 0>\n",
+                    "+ <light @ 0 0 64>\n"]
+        for line, expected_line in itertools.zip_longest(sample.unified_diff(), expected):
+            assert line == expected_line
+
+
+class TestPakfileDiff:
+    # TODO: generate dummy pakfiles for comparison
+    ...