Merge branch 'main' of github.com:pytorch/torchcodec into file_like_tutorial

Author: scotts

packaging/post_build_script.sh

@@ -37,22 +37,14 @@ assert_not_in_wheel $wheel_path "^doc"
 assert_not_in_wheel $wheel_path "^benchmarks"
 assert_not_in_wheel $wheel_path "^packaging"
 
-if [[ "$unamestr" == 'Linux' ]]; then
-    # TODO: Put this back with higher upper bound of version symbol.
-    # # See invoked python script below for details about this check.
-    # extracted_wheel_dir=$(mktemp -d)
-    # unzip -q $wheel_path -d $extracted_wheel_dir
-    # symbols_matches=$(find $extracted_wheel_dir | grep ".so$" | xargs objdump --syms | grep GLIBCXX_3.4.)
-    # python packaging/check_glibcxx.py "$symbols_matches"
-
-    echo "ls dist"
-    ls dist
-
-    old="linux_x86_64"
-    new="manylinux_2_17_x86_64.manylinux2014_x86_64"
-    echo "Replacing ${old} with ${new} in wheel name"
-    mv dist/*${old}*.whl $(echo dist/*${old}*.whl | sed "s/${old}/${new}/")
-fi
+# TODO: Put this back with higher upper bound of version symbol.
+#if [[ "$unamestr" == 'Linux' ]]; then
+#    # See invoked python script below for details about this check.
+#    extracted_wheel_dir=$(mktemp -d)
+#    unzip -q $wheel_path -d $extracted_wheel_dir
+#    symbols_matches=$(find $extracted_wheel_dir | grep ".so$" | xargs objdump --syms | grep GLIBCXX_3.4.)
+#    python packaging/check_glibcxx.py "$symbols_matches"
+#fi
 
 echo "ls dist"
 ls dist

test/test_ops.py

@@ -1070,6 +1070,49 @@ def seek(self, offset: int, whence: int) -> bytes:
                 "approximate",
             )
 
+    @pytest.mark.parametrize("how_much_to_read", ("half", "minus_10"))
+    def test_file_like_read_less_than_requested(self, how_much_to_read):
+        # Check that reading fewer bytes than requested still works. FFmpeg will
+        # figure out how to get the necessary bytes.
+        class FileLike:
+            def __init__(self, file):
+                self._file = file
+
+            def read(self, size: int) -> bytes:
+                if how_much_to_read == "half":
+                    size = size // 2
+                elif how_much_to_read == "minus_10":
+                    size = size - 10
+                else:
+                    raise ValueError("Check parametrization of this test!")
+
+                return self._file.read(size)
+
+            def seek(self, offset: int, whence: int) -> bytes:
+                return self._file.seek(offset, whence)
+
+        decoder_file_like = create_from_file_like(
+            FileLike(open(NASA_VIDEO.path, mode="rb", buffering=0))
+        )
+        add_video_stream(decoder_file_like)
+
+        decoder_reference = create_from_file(str(NASA_VIDEO.path))
+        add_video_stream(decoder_reference)
+
+        torch.manual_seed(0)
+        indices = torch.randint(
+            0, len(NASA_VIDEO.frames[NASA_VIDEO.default_stream_index]), size=(50,)
+        ).tolist()
+
+        frames_file_like, *_ = get_frames_at_indices(
+            decoder_file_like, frame_indices=indices
+        )
+        frames_references, *_ = get_frames_at_indices(
+            decoder_reference, frame_indices=indices
+        )
+
+        torch.testing.assert_close(frames_file_like, frames_references)
+
 
 class TestAudioEncoderOps: