Skip to content

πŸŽ› πŸ”Š A Python library for audio.

Fix Python 3.10 crash when reading bytes from Python.

Peter Sobote4c8b663dacd
diff --git a/pedalboard/io/PythonInputStream.h b/pedalboard/io/PythonInputStream.h
index 5d77dbc..1c9a787 100644
--- a/pedalboard/io/PythonInputStream.h
+++ b/pedalboard/io/PythonInputStream.h
@@ -102,49 +102,57 @@ public:
if (PythonException::isPending())
return 0;
- py::bytes resultBytes;
- {
- py::gil_scoped_acquire acquire;
- try {
- auto readResult = fileLike.attr("read")(bytesToRead);
-
- if (!py::isinstance<py::bytes>(readResult)) {
- std::string message =
- "File-like object passed to AudioFile was expected to return "
- "bytes from its read(...) method, but "
- "returned " +
- py::str(readResult.get_type().attr("__name__"))
- .cast<std::string>() +
- ".";
-
- if (py::hasattr(fileLike, "mode") &&
- fileLike.attr("mode").cast<std::string>() == "r") {
- message += " (Try opening the stream in \"rb\" mode instead of "
- "\"r\" mode if possible.)";
- }
-
- throw py::type_error(message);
- return 0;
+ py::gil_scoped_acquire acquire;
+ try {
+ auto readResult = fileLike.attr("read")(bytesToRead);
+
+ if (!py::isinstance<py::bytes>(readResult)) {
+ std::string message =
+ "File-like object passed to AudioFile was expected to return "
+ "bytes from its read(...) method, but "
+ "returned " +
+ py::str(readResult.get_type().attr("__name__"))
+ .cast<std::string>() +
+ ".";
+
+ if (py::hasattr(fileLike, "mode") &&
+ py::str(fileLike.attr("mode")).cast<std::string>() == "r") {
+ message += " (Try opening the stream in \"rb\" mode instead of "
+ "\"r\" mode if possible.)";
}
- resultBytes = readResult.cast<py::bytes>();
- } catch (py::error_already_set e) {
- e.restore();
- return 0;
- } catch (const py::builtin_exception &e) {
- e.set_error();
+ throw py::type_error(message);
return 0;
}
- }
- std::string resultAsString = resultBytes;
- std::memcpy((char *)buffer, (char *)resultAsString.data(),
- resultAsString.size());
+ py::bytes bytesObject = readResult.cast<py::bytes>();
+ char *pythonBuffer = nullptr;
+ ssize_t pythonLength = 0;
+
+ if (PYBIND11_BYTES_AS_STRING_AND_SIZE(bytesObject.ptr(), &pythonBuffer,
+ &pythonLength)) {
+ throw py::buffer_error(
+ "Internal error: failed to read bytes from bytes object!");
+ }
+
+ if (!buffer && pythonLength > 0) {
+ throw py::buffer_error("Internal error: bytes pointer is null, but a "
+ "non-zero number of bytes were returned!");
+ }
- int bytesCopied = resultAsString.size();
+ if (buffer && pythonLength) {
+ std::memcpy(buffer, pythonBuffer, pythonLength);
+ }
- lastReadWasSmallerThanExpected = bytesToRead > bytesCopied;
- return bytesCopied;
+ lastReadWasSmallerThanExpected = bytesToRead > pythonLength;
+ return pythonLength;
+ } catch (py::error_already_set e) {
+ e.restore();
+ return 0;
+ } catch (const py::builtin_exception &e) {
+ e.set_error();
+ return 0;
+ }
}
bool isExhausted() noexcept {
diff --git a/pedalboard/io/PythonOutputStream.h b/pedalboard/io/PythonOutputStream.h
index 2316490..035f8ff 100644
--- a/pedalboard/io/PythonOutputStream.h
+++ b/pedalboard/io/PythonOutputStream.h
@@ -95,7 +95,7 @@ public:
py::gil_scoped_acquire acquire;
try {
- const size_t maxEffectiveSize = std::min(numTimesToRepeat, 8192UL);
+ const size_t maxEffectiveSize = std::min(numTimesToRepeat, (size_t)8192);
std::vector<char> buffer(maxEffectiveSize, byte);
for (size_t i = 0; i < numTimesToRepeat; i += buffer.size()) {
diff --git a/pedalboard/io/ReadableAudioFile.h b/pedalboard/io/ReadableAudioFile.h
index 35d7e4e..fe9fe53 100644
--- a/pedalboard/io/ReadableAudioFile.h
+++ b/pedalboard/io/ReadableAudioFile.h
@@ -572,7 +572,7 @@ inline void init_readable_audio_file(py::module &m) {
.def_property_readonly("samplerate", &ReadableAudioFile::getSampleRate,
"The sample rate of this file in samples "
"(per channel) per second (Hz).")
- .def_property_readonly("channels", &ReadableAudioFile::getNumChannels,
+ .def_property_readonly("num_channels", &ReadableAudioFile::getNumChannels,
"The number of channels in this file.")
.def_property_readonly("frames", &ReadableAudioFile::getLengthInSamples,
"The total number of frames (samples per "
diff --git a/pedalboard/io/WriteableAudioFile.h b/pedalboard/io/WriteableAudioFile.h
index 97148e4..b678497 100644
--- a/pedalboard/io/WriteableAudioFile.h
+++ b/pedalboard/io/WriteableAudioFile.h
@@ -225,16 +225,15 @@ public:
// Use the pythonOutputStream's filename if possible, falling back to the
// provided filename string (which should contain an extension) if
// necessary.
- std::string filenameOrFormat;
if (!filename.empty()) {
- filenameOrFormat = ("." + filename);
+ extension = filename;
} else if (auto streamName = pythonOutputStream->getFilename()) {
- filenameOrFormat = streamName.value();
+ // Dummy-stream-filename added here to avoid a JUCE assertion
+ // if the stream name doesn't start with a slash.
+ juce::File file(File::getSeparatorString().toStdString() + "dummy-stream-filename-" + streamName.value());
+ extension = file.getFileExtension().toStdString();
}
- juce::File file(filenameOrFormat);
- extension = file.getFileExtension().toStdString();
-
format = formatManager.findFormatForFileExtension(extension);
if (!format) {
@@ -869,7 +868,8 @@ inline void init_writeable_audio_file(py::module &m) {
.def_property_readonly("samplerate", &WriteableAudioFile::getSampleRate,
"The sample rate of this file in samples "
"(per channel) per second (Hz).")
- .def_property_readonly("channels", &WriteableAudioFile::getNumChannels,
+ .def_property_readonly("num_channels",
+ &WriteableAudioFile::getNumChannels,
"The number of channels in this file.")
.def_property_readonly("frames", &WriteableAudioFile::getFramesWritten,
"The total number of frames (samples per "
diff --git a/tests/test_io.py b/tests/test_io.py
index 5003d2e..ade33c3 100644
--- a/tests/test_io.py
+++ b/tests/test_io.py
@@ -148,7 +148,7 @@ def test_basic_formats_available_on_all_platforms(extension: str):
def test_basic_read(audio_filename: str, samplerate: float):
af = pedalboard.io.AudioFile(audio_filename)
assert af.samplerate == samplerate
- assert af.channels == 1
+ assert af.num_channels == 1
if any(ext in audio_filename for ext in EXPECT_LENGTH_TO_BE_EXACT):
assert af.frames == int(samplerate * EXPECTED_DURATION_SECONDS)
else:
@@ -172,12 +172,14 @@ def test_basic_read(audio_filename: str, samplerate: float):
assert samples.shape == (1, int(samplerate * EXPECTED_DURATION_SECONDS / 2))
assert f"samplerate={int(af.samplerate)}" in repr(af)
- assert f"num_channels={af.channels}" in repr(af)
+ assert f"num_channels={af.num_channels}" in repr(af)
assert f"file_dtype={af.file_dtype}" in repr(af)
af.seek(0)
actual = af.read(af.frames)
- expected = generate_sine_at(af.samplerate, num_channels=af.channels, num_seconds=af.duration)
+ expected = generate_sine_at(
+ af.samplerate, num_channels=af.num_channels, num_seconds=af.duration
+ )
# Crop the ends of the file, as lossy formats sometimes don't encode the whole file:
actual = actual[:, : len(expected)]
tolerance = get_tolerance_for_format_and_bit_depth(
@@ -188,7 +190,7 @@ def test_basic_read(audio_filename: str, samplerate: float):
af.close()
with pytest.raises(RuntimeError):
- af.channels
+ af.num_channels
with pytest.raises(RuntimeError):
af.read(1)
@@ -207,7 +209,7 @@ def test_read_raw(audio_filename: str, samplerate: float):
def test_use_reader_as_context_manager(audio_filename: str, samplerate: float):
with pedalboard.io.AudioFile(audio_filename) as af:
assert af.samplerate == samplerate
- assert af.channels == 1
+ assert af.num_channels == 1
if any(ext in audio_filename for ext in EXPECT_LENGTH_TO_BE_EXACT):
assert af.frames == int(samplerate * EXPECTED_DURATION_SECONDS)
else:
@@ -231,10 +233,10 @@ def test_use_reader_as_context_manager(audio_filename: str, samplerate: float):
assert samples.shape == (1, int(samplerate * EXPECTED_DURATION_SECONDS / 2))
assert f"samplerate={int(af.samplerate)}" in repr(af)
- assert f"num_channels={af.channels}" in repr(af)
+ assert f"num_channels={af.num_channels}" in repr(af)
with pytest.raises(RuntimeError):
- af.channels
+ af.num_channels
with pytest.raises(RuntimeError):
af.read(1)
@@ -257,7 +259,7 @@ def test_read_okay_without_extension(
try:
with pedalboard.io.AudioFile(dest_path) as af:
assert af.samplerate == samplerate
- assert af.channels == 1
+ assert af.num_channels == 1
except Exception:
if ".mp3" in audio_filename:
# Skip this test - due to a bug in JUCE's MP3 reader on Linux/Windows,
@@ -286,7 +288,7 @@ def test_read_from_seekable_stream(audio_filename: str, samplerate: float):
with af:
assert af.samplerate == samplerate
- assert af.channels == 1
+ assert af.num_channels == 1
if any(ext in audio_filename for ext in EXPECT_LENGTH_TO_BE_EXACT):
assert af.frames == int(samplerate * EXPECTED_DURATION_SECONDS)
else:
@@ -310,11 +312,11 @@ def test_read_from_seekable_stream(audio_filename: str, samplerate: float):
assert samples.shape == (1, int(samplerate * EXPECTED_DURATION_SECONDS / 2))
assert f"samplerate={int(af.samplerate)}" in repr(af)
- assert f"num_channels={af.channels}" in repr(af)
+ assert f"num_channels={af.num_channels}" in repr(af)
assert repr(stream) in repr(af)
with pytest.raises(RuntimeError):
- af.channels
+ af.num_channels
with pytest.raises(RuntimeError):
af.read(1)
@@ -521,7 +523,7 @@ def test_basic_write(
assert os.path.getsize(filename) > 0
with pedalboard.io.ReadableAudioFile(filename) as af:
assert af.samplerate == samplerate
- assert af.channels == num_channels
+ assert af.num_channels == num_channels
tolerance = get_tolerance_for_format_and_bit_depth(extension, input_format, af.file_dtype)
as_written = af.read(num_samples)
np.testing.assert_allclose(original_audio, np.squeeze(as_written), atol=tolerance)
@@ -642,7 +644,7 @@ def test_write_to_seekable_stream(
with pedalboard.io.AudioFile(stream) as af:
assert af.samplerate == samplerate
- assert af.channels == num_channels
+ assert af.num_channels == num_channels
tolerance = get_tolerance_for_format_and_bit_depth(extension, input_format, af.file_dtype)
as_written = af.read(num_samples)
np.testing.assert_allclose(original_audio, np.squeeze(as_written), atol=tolerance)