mirror of
https://github.com/mollyim/webrtc.git
synced 2025-05-14 06:10:40 +01:00

This reverts commit01aa210fad
. Reason for revert: downstream project adjusted Original change's description: > Revert "Enable and fix chromium clang warnings in rtp_rtcp test targets" > > This reverts commit9486b117da
. > > Reason for revert: Breaks downstream project > > Original change's description: > > Enable and fix chromium clang warnings in rtp_rtcp test targets > > > > Bug: webrtc:163 > > Change-Id: I4ed3e63296d8bf06536a83196d597c7a906ba11c > > Reviewed-on: https://webrtc-review.googlesource.com/60802 > > Reviewed-by: Björn Terelius <terelius@webrtc.org> > > Reviewed-by: Patrik Höglund <phoglund@webrtc.org> > > Commit-Queue: Danil Chapovalov <danilchap@webrtc.org> > > Cr-Commit-Position: refs/heads/master@{#22357} > > TBR=danilchap@webrtc.org,phoglund@webrtc.org,terelius@webrtc.org > > Change-Id: I2c3777ea9f26813bdb395e7fd68f6b49443586ea > No-Presubmit: true > No-Tree-Checks: true > No-Try: true > Bug: webrtc:163 > Reviewed-on: https://webrtc-review.googlesource.com/61060 > Reviewed-by: Oleh Prypin <oprypin@webrtc.org> > Commit-Queue: Oleh Prypin <oprypin@webrtc.org> > Cr-Commit-Position: refs/heads/master@{#22365} TBR=danilchap@webrtc.org,phoglund@webrtc.org,oprypin@webrtc.org,terelius@webrtc.org Change-Id: I0b4cb6d05b37caeb52cca9abf95417ad3ad6f76b No-Presubmit: true No-Tree-Checks: true No-Try: true Bug: webrtc:163 Reviewed-on: https://webrtc-review.googlesource.com/61080 Reviewed-by: Danil Chapovalov <danilchap@webrtc.org> Commit-Queue: Danil Chapovalov <danilchap@webrtc.org> Cr-Commit-Position: refs/heads/master@{#22368}
272 lines
9.9 KiB
C++
272 lines
9.9 KiB
C++
/*
|
|
* Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
|
|
*
|
|
* Use of this source code is governed by a BSD-style license
|
|
* that can be found in the LICENSE file in the root of the source
|
|
* tree. An additional intellectual property rights grant can be found
|
|
* in the file PATENTS. All contributing project authors may
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
*/
|
|
|
|
#include <limits>
|
|
|
|
#include "modules/rtp_rtcp/source/byte_io.h"
|
|
#include "test/gtest.h"
|
|
|
|
namespace webrtc {
|
|
namespace {
|
|
|
|
class ByteIoTest : public ::testing::Test {
|
|
protected:
|
|
ByteIoTest() = default;
|
|
~ByteIoTest() override = default;
|
|
|
|
enum { kAlignments = sizeof(uint64_t) - 1 };
|
|
|
|
// Method to create a test value that is not the same when byte reversed.
|
|
template <typename T>
|
|
T CreateTestValue(bool negative, uint8_t num_bytes) {
|
|
// Examples of output:
|
|
// T = int32_t, negative = false, num_bytes = 4: 0x00010203
|
|
// T = int32_t, negative = true, num_bytes = 4: 0xFFFEFDFC
|
|
// T = int32_t, negative = false, num_bytes = 3: 0x000102
|
|
// * T = int32_t, negative = true, num_bytes = 3: 0xFFFEFD
|
|
|
|
T val = 0;
|
|
for (uint8_t i = 0; i != num_bytes; ++i) {
|
|
val = (val << 8) + (negative ? (0xFF - i) : (i + 1));
|
|
}
|
|
|
|
// This loop will create a sign extend mask if num_bytes if necessary.
|
|
// For the last example (marked * above), the number needs to be sign
|
|
// extended to be a valid int32_t. The sign extend mask is 0xFF000000.
|
|
// Comments for each step with this example below.
|
|
if (std::numeric_limits<T>::is_signed && negative &&
|
|
num_bytes < sizeof(T)) {
|
|
// Start with mask = 0xFFFFFFFF.
|
|
T mask = static_cast<T>(-1);
|
|
// Create a temporary for the lowest byte (0x000000FF).
|
|
const T neg_byte = static_cast<T>(0xFF);
|
|
for (int i = 0; i < num_bytes; ++i) {
|
|
// And the inverse of the temporary and the mask:
|
|
// 0xFFFFFFFF & 0xFFFFFF00 = 0xFFFFFF00.
|
|
// 0xFFFFFF00 & 0xFFFF00FF = 0xFFFF0000.
|
|
// 0xFFFF0000 & 0xFF00FFFF = 0xFF000000.
|
|
mask &= ~(neg_byte << (i * 8));
|
|
}
|
|
// Add the sign extension mask to the actual value.
|
|
val |= mask;
|
|
}
|
|
return val;
|
|
}
|
|
|
|
// Populate byte buffer with value, in big endian format.
|
|
template <typename T>
|
|
void PopulateTestData(uint8_t* data, T value, int num_bytes, bool bigendian) {
|
|
if (bigendian) {
|
|
for (int i = 0; i < num_bytes; ++i) {
|
|
data[i] = (value >> ((num_bytes - i - 1) * 8)) & 0xFF;
|
|
}
|
|
} else {
|
|
for (int i = 0; i < num_bytes; ++i) {
|
|
data[i] = (value >> (i * 8)) & 0xFF;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Test reading big endian numbers.
|
|
// Template arguments: Type T, read method RM(buffer), B bytes of data.
|
|
template <typename T, T (*RM)(const uint8_t*), int B>
|
|
void TestRead(bool big_endian) {
|
|
// Test both for values that are positive and negative (if signed)
|
|
for (int neg = 0; neg < 2; ++neg) {
|
|
bool negative = neg > 0;
|
|
|
|
// Write test value to byte buffer, in big endian format.
|
|
T test_value = CreateTestValue<T>(negative, B);
|
|
uint8_t bytes[B + kAlignments];
|
|
|
|
// Make one test for each alignment.
|
|
for (int i = 0; i < kAlignments; ++i) {
|
|
PopulateTestData(bytes + i, test_value, B, big_endian);
|
|
|
|
// Check that test value is retrieved from buffer when used read method.
|
|
EXPECT_EQ(test_value, RM(bytes + i));
|
|
}
|
|
}
|
|
}
|
|
|
|
// Test writing big endian numbers.
|
|
// Template arguments: Type T, write method WM(buffer, value), B bytes of data
|
|
template <typename T, void (*WM)(uint8_t*, T), int B>
|
|
void TestWrite(bool big_endian) {
|
|
// Test both for values that are positive and negative (if signed).
|
|
for (int neg = 0; neg < 2; ++neg) {
|
|
bool negative = neg > 0;
|
|
|
|
// Write test value to byte buffer, in big endian format.
|
|
T test_value = CreateTestValue<T>(negative, B);
|
|
uint8_t expected_bytes[B + kAlignments];
|
|
uint8_t bytes[B + kAlignments];
|
|
|
|
// Make one test for each alignment.
|
|
for (int i = 0; i < kAlignments; ++i) {
|
|
PopulateTestData(expected_bytes + i, test_value, B, big_endian);
|
|
|
|
// Zero initialize buffer and let WM populate it.
|
|
memset(bytes, 0, B + kAlignments);
|
|
WM(bytes + i, test_value);
|
|
|
|
// Check that data produced by WM is big endian as expected.
|
|
for (int j = 0; j < B; ++j) {
|
|
EXPECT_EQ(expected_bytes[i + j], bytes[i + j]);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
};
|
|
|
|
TEST_F(ByteIoTest, Test16UBitBigEndian) {
|
|
TestRead<uint16_t, ByteReader<uint16_t>::ReadBigEndian,
|
|
sizeof(uint16_t)>(true);
|
|
TestWrite<uint16_t, ByteWriter<uint16_t>::WriteBigEndian,
|
|
sizeof(uint16_t)>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test24UBitBigEndian) {
|
|
TestRead<uint32_t, ByteReader<uint32_t, 3>::ReadBigEndian, 3>(true);
|
|
TestWrite<uint32_t, ByteWriter<uint32_t, 3>::WriteBigEndian, 3>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test32UBitBigEndian) {
|
|
TestRead<uint32_t, ByteReader<uint32_t>::ReadBigEndian,
|
|
sizeof(uint32_t)>(true);
|
|
TestWrite<uint32_t, ByteWriter<uint32_t>::WriteBigEndian,
|
|
sizeof(uint32_t)>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test64UBitBigEndian) {
|
|
TestRead<uint64_t, ByteReader<uint64_t>::ReadBigEndian,
|
|
sizeof(uint64_t)>(true);
|
|
TestWrite<uint64_t, ByteWriter<uint64_t>::WriteBigEndian,
|
|
sizeof(uint64_t)>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test16SBitBigEndian) {
|
|
TestRead<int16_t, ByteReader<int16_t>::ReadBigEndian,
|
|
sizeof(int16_t)>(true);
|
|
TestWrite<int16_t, ByteWriter<int16_t>::WriteBigEndian,
|
|
sizeof(int16_t)>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test24SBitBigEndian) {
|
|
TestRead<int32_t, ByteReader<int32_t, 3>::ReadBigEndian, 3>(true);
|
|
TestWrite<int32_t, ByteWriter<int32_t, 3>::WriteBigEndian, 3>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test32SBitBigEndian) {
|
|
TestRead<int32_t, ByteReader<int32_t>::ReadBigEndian,
|
|
sizeof(int32_t)>(true);
|
|
TestWrite<int32_t, ByteWriter<int32_t>::WriteBigEndian,
|
|
sizeof(int32_t)>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test64SBitBigEndian) {
|
|
TestRead<int64_t, ByteReader<int64_t>::ReadBigEndian,
|
|
sizeof(int64_t)>(true);
|
|
TestWrite<int64_t, ByteWriter<int64_t>::WriteBigEndian,
|
|
sizeof(int64_t)>(true);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test16UBitLittleEndian) {
|
|
TestRead<uint16_t, ByteReader<uint16_t>::ReadLittleEndian,
|
|
sizeof(uint16_t)>(false);
|
|
TestWrite<uint16_t, ByteWriter<uint16_t>::WriteLittleEndian,
|
|
sizeof(uint16_t)>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test24UBitLittleEndian) {
|
|
TestRead<uint32_t, ByteReader<uint32_t, 3>::ReadLittleEndian, 3>(false);
|
|
TestWrite<uint32_t, ByteWriter<uint32_t, 3>::WriteLittleEndian, 3>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test32UBitLittleEndian) {
|
|
TestRead<uint32_t, ByteReader<uint32_t>::ReadLittleEndian,
|
|
sizeof(uint32_t)>(false);
|
|
TestWrite<uint32_t, ByteWriter<uint32_t>::WriteLittleEndian,
|
|
sizeof(uint32_t)>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test64UBitLittleEndian) {
|
|
TestRead<uint64_t, ByteReader<uint64_t>::ReadLittleEndian,
|
|
sizeof(uint64_t)>(false);
|
|
TestWrite<uint64_t, ByteWriter<uint64_t>::WriteLittleEndian,
|
|
sizeof(uint64_t)>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test16SBitLittleEndian) {
|
|
TestRead<int16_t, ByteReader<int16_t>::ReadLittleEndian,
|
|
sizeof(int16_t)>(false);
|
|
TestWrite<int16_t, ByteWriter<int16_t>::WriteLittleEndian,
|
|
sizeof(int16_t)>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test24SBitLittleEndian) {
|
|
TestRead<int32_t, ByteReader<int32_t, 3>::ReadLittleEndian, 3>(false);
|
|
TestWrite<int32_t, ByteWriter<int32_t, 3>::WriteLittleEndian, 3>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test32SBitLittleEndian) {
|
|
TestRead<int32_t, ByteReader<int32_t>::ReadLittleEndian,
|
|
sizeof(int32_t)>(false);
|
|
TestWrite<int32_t, ByteWriter<int32_t>::WriteLittleEndian,
|
|
sizeof(int32_t)>(false);
|
|
}
|
|
|
|
TEST_F(ByteIoTest, Test64SBitLittleEndian) {
|
|
TestRead<int64_t, ByteReader<int64_t>::ReadLittleEndian,
|
|
sizeof(int64_t)>(false);
|
|
TestWrite<int64_t, ByteWriter<int64_t>::WriteLittleEndian,
|
|
sizeof(int64_t)>(false);
|
|
}
|
|
|
|
// Sets up a fixed byte array and converts N bytes from the array into a
|
|
// uint64_t. Verifies the value with hard-coded reference.
|
|
TEST(ByteIo, SanityCheckFixedByteArrayUnsignedReadBigEndian) {
|
|
uint8_t data[8] = {0xFF, 0xEE, 0xDD, 0xCC, 0xBB, 0xAA, 0x99, 0x88};
|
|
uint64_t value = ByteReader<uint64_t, 2>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEE), value);
|
|
value = ByteReader<uint64_t, 3>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEEDD), value);
|
|
value = ByteReader<uint64_t, 4>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEEDDCC), value);
|
|
value = ByteReader<uint64_t, 5>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEEDDCCBB), value);
|
|
value = ByteReader<uint64_t, 6>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEEDDCCBBAA), value);
|
|
value = ByteReader<uint64_t, 7>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEEDDCCBBAA99), value);
|
|
value = ByteReader<uint64_t, 8>::ReadBigEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xFFEEDDCCBBAA9988), value);
|
|
}
|
|
|
|
// Same as above, but for little-endian reading.
|
|
TEST(ByteIo, SanityCheckFixedByteArrayUnsignedReadLittleEndian) {
|
|
uint8_t data[8] = {0xFF, 0xEE, 0xDD, 0xCC, 0xBB, 0xAA, 0x99, 0x88};
|
|
uint64_t value = ByteReader<uint64_t, 2>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xEEFF), value);
|
|
value = ByteReader<uint64_t, 3>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xDDEEFF), value);
|
|
value = ByteReader<uint64_t, 4>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xCCDDEEFF), value);
|
|
value = ByteReader<uint64_t, 5>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xBBCCDDEEFF), value);
|
|
value = ByteReader<uint64_t, 6>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0xAABBCCDDEEFF), value);
|
|
value = ByteReader<uint64_t, 7>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0x99AABBCCDDEEFF), value);
|
|
value = ByteReader<uint64_t, 8>::ReadLittleEndian(data);
|
|
EXPECT_EQ(static_cast<uint64_t>(0x8899AABBCCDDEEFF), value);
|
|
}
|
|
} // namespace
|
|
} // namespace webrtc
|