Speed up reading of index data records

src/rosbags/rosbag1/reader.py

@@ -24,7 +24,20 @@ from rosbags.typesys.msg import normalize_msgtype
 
 if TYPE_CHECKING:
     from types import TracebackType
-    from typing import BinaryIO, Callable, Generator, Iterable, Literal, Optional, Type, Union
+    from typing import (
+        BinaryIO,
+        Callable,
+        Generator,
+        Iterable,
+        Literal,
+        Optional,
+        Tuple,
+        Type,
+        Union,
+    )
+
+    Unpack = Callable[[bytes], Tuple[int]]
+    UnpackFrom = Callable[[bytes, int], Tuple[int]]
 
 
 class ReaderError(Exception):
@@ -103,9 +116,9 @@ class IndexData(NamedTuple):
         return NotImplemented  # pragma: no cover
 
 
-deserialize_uint8: Callable[[bytes], tuple[int]] = struct.Struct('<B').unpack  # type: ignore
-deserialize_uint32: Callable[[bytes], tuple[int]] = struct.Struct('<L').unpack  # type: ignore
-deserialize_uint64: Callable[[bytes], tuple[int]] = struct.Struct('<Q').unpack  # type: ignore
+deserialize_uint8: Unpack = struct.Struct('<B').unpack  # type: ignore
+deserialize_uint32: UnpackFrom = struct.Struct('<L').unpack_from  # type: ignore
+deserialize_uint64: Unpack = struct.Struct('<Q').unpack  # type: ignore
 
 
 def deserialize_time(val: bytes) -> int:
@@ -158,7 +171,7 @@ class Header(Dict[str, Any]):
 
         """
         try:
-            return deserialize_uint32(self[name])[0]
+            return deserialize_uint32(self[name], 0)[0]
         except (KeyError, struct.error) as err:
             raise ReaderError(f'Could not read uint32 field {name!r}.') from err
 
@@ -243,7 +256,7 @@ class Header(Dict[str, Any]):
         length = len(binary)
         while pos < length:
             try:
-                size = deserialize_uint32(binary[pos:pos + 4])[0]
+                size = deserialize_uint32(binary, pos)[0]
             except struct.error as err:
                 raise ReaderError('Header field size could not be read.') from err
             pos += 4
@@ -280,7 +293,7 @@ def read_uint32(src: BinaryIO) -> int:
 
     """
     try:
-        return deserialize_uint32(src.read(4))[0]
+        return deserialize_uint32(src.read(4), 0)[0]
     except struct.error as err:
         raise ReaderError('Could not read uint32.') from err
 
@@ -326,6 +339,8 @@ class Reader:
 
     """
 
+    # pylint: disable=too-many-instance-attributes
+
     def __init__(self, path: Union[str, Path]):
         """Initialize.
 
@@ -343,6 +358,7 @@ class Reader:
         self.bio: Optional[BinaryIO] = None
         self.connections: list[Connection] = []
         self.indexes: dict[int, list[IndexData]]
+        self.index_data_header_offsets: Optional[tuple[int, int]] = None
         self.chunk_infos: list[ChunkInfo] = []
         self.chunks: dict[int, Chunk] = {}
         self.current_chunk: tuple[int, BinaryIO] = (-1, BytesIO())
@@ -537,6 +553,9 @@ class Reader:
     def read_index_data(self, pos: int, indexes: dict[int, list[IndexData]]) -> None:
         """Read index data from position.
 
+        The implementation purposely avoids the generic Header class and
+        its costly string processing.
+
         Args:
             pos: Seek position.
             indexes: Accumulated index data.
@@ -546,20 +565,45 @@ class Reader:
 
         """
         assert self.bio
-        header = Header.read(self.bio, RecordType.IDXDATA)
 
-        ver = header.get_uint32('ver')
-        if ver != 1:
-            raise ReaderError(f'IDXDATA version {ver} is not supported.')
-        conn = header.get_uint32('conn')
-        count = header.get_uint32('count')
+        buf = self.bio.read(55)
+        if not self.index_data_header_offsets:
+            size, = deserialize_uint32(buf, 0)
+            assert size == 47
+            idx = 4
+            connpos = -1
+            countpos = -1
+            while idx < size:
+                char = buf[idx + 6]
+                if char == 61:  # ord(b'=')
+                    assert buf[idx + 7] == 4
+                    idx += 8
+                elif char == 114:  # ord(b'r')
+                    if (ver := buf[idx + 8]) != 1:
+                        raise ReaderError(f'IDXDATA version {ver} is not supported.')
+                    idx += 12
+                elif char == 110:  # ord(b'n')
+                    connpos = idx + 9
+                    idx += 13
+                else:
+                    assert char == 117  # ord(b'u')
+                    countpos = idx + 10
+                    idx += 14
+            self.index_data_header_offsets = (connpos, countpos)
+        connpos, countpos = self.index_data_header_offsets
 
-        self.bio.seek(4, os.SEEK_CUR)
+        conn, = deserialize_uint32(buf, connpos)
+        count, = deserialize_uint32(buf, countpos)
+        size, = deserialize_uint32(buf, 51)
+        assert size == count * 12
 
         index = indexes[conn]
-        for _ in range(count):
-            time = deserialize_time(self.bio.read(8))
-            offset = read_uint32(self.bio)
+        buf = self.bio.read(size)
+        idx = 0
+        while idx < size:
+            time = deserialize_uint32(buf, idx)[0] * 10**9 + deserialize_uint32(buf, idx + 4)[0]
+            offset, = deserialize_uint32(buf, idx + 8)
+            idx += 12
             index.append(IndexData(time, pos, offset))
 
     def messages(