Use htree for inode_at, fix inode_at performance (#32)

* Simplify inode_at logic

* Add htree inode_at test

* Implement htree, and fix linear scan performance

* Bump version

* Fix test image generation

* Review feedback

* Fix fuzz

* Review feedback

* only mmove data that returns

* Update cachetools

* Fix hash logic

* Fix scan issue

* Fix deprecation warning

* Review feedback

* Alphabetical slots

Author: Eeems

.github/workflows/build.yaml

@@ -62,6 +62,7 @@ jobs:
             test64.ext4
             test64.ext4.tmp
             test_htree.ext4
+            test_htree_multi.ext4
           if-no-files-found: error
 
   test:

_test_image.sh

@@ -45,7 +45,6 @@ mkimage test32 "$tmp_dir" 20 -O ^64bit
 mkimage test64 "$tmp_dir" 20 -O 64bit
 
 rm -f "$tmp_dir"/test*.txt "$tmp_dir"/symlink.txt
-echo "[test] Generating files..."
 
 echo "[test] Making image test_htree..."
 chronic dd if=/dev/zero of=test_htree.ext4 count=20 bs=1048576
@@ -59,3 +58,15 @@ sudo umount "$tmp_dir"
 # shellcheck disable=SC2064
 trap "rmdir \"$tmp_dir\"" EXIT
 chronic e2fsck -Dy test_htree.ext4
+
+echo "[test] Making image test_htree_multi..."
+chronic dd if=/dev/zero of=test_htree_multi.ext4 count=20 bs=1048576
+chronic mkfs.ext4 -g 1024 -b 1024 -O 64bit,dir_index -N 13000 test_htree_multi.ext4
+sudo mount -t ext4 test_htree_multi.ext4 "$tmp_dir"
+# shellcheck disable=SC2064
+trap "sudo umount \"$tmp_dir\";rmdir \"$tmp_dir\"" EXIT
+printf '%s\n' "$tmp_dir"/{1..12000} | xargs sudo touch
+sudo umount "$tmp_dir"
+# shellcheck disable=SC2064
+trap "rmdir \"$tmp_dir\"" EXIT
+chronic e2fsck -Dy test_htree_multi.ext4

ext4/directory.py

@@ -36,12 +36,16 @@ def __init__(self, directory: "Directory", offset: int) -> None:
 
     @override
     def read_from_volume(self) -> None:
-        data = self.directory._open().read()[self.offset : self.offset + self.size]  # pyright: ignore[reportPrivateUsage]
-        _ = memmove(addressof(self), data, self.size)
+        reader = self.directory._open()  # pyright: ignore[reportPrivateUsage]
+        _ = reader.seek(self.offset)
+        data = reader.read(self.size)
+        # Do not guard against short reads, this happens for the last entry
+        _ = memmove(addressof(self), data, len(data))
 
 
 class DirectoryEntryBase(DirectoryEntryStruct):
     __slots__: tuple[str, ...] = ()
+
     @property
     def name_bytes(self) -> bytes:
         return bytes(self.name)[: self.name_len]  # pyright: ignore[reportAny]