Format Version 1: Adds crc32c checks. (#8)

* feat: FORMAT_VERSION=1, adds crc32 check value to format

Implementation is backwards compatible with format version 0 though
the format is different.

* docs: describe format version 1

* test: show validationerror is raised if bits corrupted

* feat: allow disabling crc checks

README.md

@@ -71,6 +71,11 @@ The byte string format consists of a 16 byte header, an index, and a series of (
 HEADER|INDEX|DATA_REGION
 ```
 
+| Format Version | description                            |
+|----------------|----------------------------------------|
+| 0              | Initial Release                        |
+| 1              | Adds crc32c check values to each item. |
+
 ### Header 
 
 ```
@@ -93,7 +98,7 @@ The index can be consulted by conducting an Eytzinger binary search over the lab
 
 ### Data Region
 
-The data objects are serialized to bytes and compressed individually if the header indicates they should be. They are then concatenated in the same order the index specifies.
+The data objects are serialized to bytes and compressed individually if the header indicates they should be. They are then concatenated in the same order the index specifies. The last four bytes are a crc32c check value that was added in format version 1.
 
 ## Versus Flexbuffers
 

automated_test.py

@@ -6,7 +6,7 @@
 
 import numpy as np
 
-from mapbuffer import MapBuffer, HEADER_LENGTH
+from mapbuffer import ValidationError, MapBuffer, HEADER_LENGTH
 
 @pytest.mark.parametrize("compress", (None, "gzip", "br", "zstd", "lzma"))
 def test_empty(compress):
@@ -56,6 +56,25 @@ def test_full(compress):
 
   assert len(mbuf.buffer) > HEADER_LENGTH
 
+@pytest.mark.parametrize("compress", (None, "gzip", "br", "zstd"))
+def test_crc32c(compress):
+  data = { 
+    1: b"hello",
+    2: b"world",
+  }
+  mbuf = MapBuffer(data, compress=compress)
+
+  idx = mbuf.buffer.index(b"hello")
+  buf = list(mbuf.buffer)
+  buf[idx] = ord(b'H')
+  mbuf.buffer = bytes(buf)
+
+  try:
+    mbuf[1]
+    assert False
+  except ValidationError:
+    pass
+
 @pytest.mark.parametrize("compress", (None, "gzip", "br", "zstd"))
 def test_mmap_access(compress):
   data = { 

mapbuffer/mapbuffer.py

@@ -5,24 +5,26 @@
 from .lib import nvl
 from . import compression
 
+import crc32c
 import numpy as np
 
 import mapbufferaccel
 
-FORMAT_VERSION = 0
+FORMAT_VERSION = 1
 MAGIC_NUMBERS = b"mapbufr"
 HEADER_LENGTH = 16
 
 class MapBuffer:
   """Represents a usable int->bytes dictionary as a byte string."""
   __slots__ = (
     "data", "tobytesfn", "frombytesfn", 
-    "dtype", "buffer", "_header", 
-    "_index", "_compress"
+    "dtype", "buffer", "check_crc", 
+    "_header", "_index", "_compress"
   )
   def __init__(
     self, data=None, compress=None,
-    tobytesfn=None, frombytesfn=None
+    tobytesfn=None, frombytesfn=None,
+    check_crc=True
   ):
     """
     data: dict (int->byte serializable object) or bytes 
@@ -41,6 +43,7 @@ def __init__(
     self.frombytesfn = frombytesfn
     self.dtype = np.uint64
     self.buffer = None
+    self.check_crc = check_crc
 
     self._header = None
     self._index = None
@@ -133,6 +136,16 @@ def getindex(self, i):
     else:
       value = self.buffer[offset:]
 
+    if self.format_version == 1:
+      stored_check_value = int.from_bytes(value[-4:], byteorder='little')
+      value = value[:-4]
+      if self.check_crc:
+        retrieved_check_value = crc32c.crc32c(value)
+        if retrieved_check_value != stored_check_value:
+          raise ValidationError(
+            f"Label {i} failed its crc32c check. Stored: {stored_check_value} Computed: {retrieved_check_value}"
+          )
+
     encoding = self.compress
     if encoding:
       value = compression.decompress(value, encoding, str(index[i,0]))
@@ -213,6 +226,8 @@ def dict2buf(self, data, compress=None, tobytesfn=None):
       label: compression.compress(tobytesfn(val), method=compress) 
       for label, val in data.items()
     }
+    for label in bytes_data:
+      bytes_data[label] += crc32c.crc32c(bytes_data[label]).to_bytes(4, byteorder='little')
 
     data_region = b"".join(
       ( bytes_data[label] for label in labels )
@@ -244,7 +259,7 @@ def validate_buffer(buf):
     if magic != MAGIC_NUMBERS:
       raise ValidationError(f"Magic number mismatch. Expected: {MAGIC_NUMBERS} Got: {magic}")
 
-    if mapbuf.format_version not in (0,):
+    if mapbuf.format_version not in (0,1):
       raise ValidationError(f"Unsupported format version. Got: {mapbuf.format_version}")
 
     if mapbuf.compress not in compression.COMPRESSION_TYPES:

requirements.txt

@@ -1,4 +1,5 @@
 brotli
+crc32c
 deflate>=0.2.0
 numpy
 tqdm