summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorTom Sepez <tsepez@chromium.org>2018-04-06 23:48:24 +0000
committerChromium commit bot <commit-bot@chromium.org>2018-04-06 23:48:24 +0000
commit6058efdbdc186e120e7e2121c290ac4d820ffbf8 (patch)
tree8944872de2d924d3efd76266cc929c6f25199134
parent2aa01f5ccbf1464b43527c1ffa6b42bafed9ebeb (diff)
downloadpdfium-6058efdbdc186e120e7e2121c290ac4d820ffbf8.tar.xz
Add span.h from chromium base.chromium/3392
Allows indexing with better bounds-checking to occur. Some small modifications are required to deal with PDFium being intentionally held at C++11 compliance, not C++14. Use in one place as check on correctness. Change-Id: Id2875cf0a93980112bc536a93c4f9ec5306c0dac Reviewed-on: https://pdfium-review.googlesource.com/29671 Commit-Queue: Tom Sepez <tsepez@chromium.org> Reviewed-by: Chris Palmer <palmer@chromium.org> Reviewed-by: dsinclair <dsinclair@chromium.org>
-rw-r--r--core/fpdfapi/page/cpdf_streamcontentparser.cpp4
-rw-r--r--core/fpdfapi/page/cpdf_streamparser.cpp23
-rw-r--r--core/fpdfapi/page/cpdf_streamparser.h15
-rw-r--r--core/fpdfapi/page/cpdf_streamparser_unittest.cpp11
-rw-r--r--core/fxcrt/unowned_ptr.h3
-rw-r--r--testing/libfuzzer/pdf_streamparser_fuzzer.cc3
-rw-r--r--third_party/BUILD.gn1
-rw-r--r--third_party/base/span.h336
8 files changed, 368 insertions, 28 deletions
diff --git a/core/fpdfapi/page/cpdf_streamcontentparser.cpp b/core/fpdfapi/page/cpdf_streamcontentparser.cpp
index 697349e987..c251c5ec66 100644
--- a/core/fpdfapi/page/cpdf_streamcontentparser.cpp
+++ b/core/fpdfapi/page/cpdf_streamcontentparser.cpp
@@ -37,6 +37,7 @@
#include "core/fxge/cfx_graphstatedata.h"
#include "third_party/base/logging.h"
#include "third_party/base/ptr_util.h"
+#include "third_party/base/span.h"
#include "third_party/base/stl_util.h"
namespace {
@@ -1515,7 +1516,8 @@ uint32_t CPDF_StreamContentParser::Parse(const uint8_t* pData,
pData);
uint32_t InitObjCount = m_pObjectHolder->GetPageObjectList()->size();
- CPDF_StreamParser syntax(pData, dwSize, m_pDocument->GetByteStringPool());
+ CPDF_StreamParser syntax(pdfium::make_span(pData, dwSize),
+ m_pDocument->GetByteStringPool());
CPDF_StreamParserAutoClearer auto_clearer(&m_pSyntax, &syntax);
while (1) {
uint32_t cost = m_pObjectHolder->GetPageObjectList()->size() - InitObjCount;
diff --git a/core/fpdfapi/page/cpdf_streamparser.cpp b/core/fpdfapi/page/cpdf_streamparser.cpp
index 3b6d12038f..c6660b799b 100644
--- a/core/fpdfapi/page/cpdf_streamparser.cpp
+++ b/core/fpdfapi/page/cpdf_streamparser.cpp
@@ -101,13 +101,12 @@ uint32_t DecodeInlineStream(const uint8_t* src_buf,
} // namespace
-CPDF_StreamParser::CPDF_StreamParser(const uint8_t* pData, uint32_t dwSize)
- : m_Size(dwSize), m_Pos(0), m_WordSize(0), m_pBuf(pData) {}
+CPDF_StreamParser::CPDF_StreamParser(pdfium::span<const uint8_t> span)
+ : m_pBuf(span) {}
-CPDF_StreamParser::CPDF_StreamParser(const uint8_t* pData,
- uint32_t dwSize,
+CPDF_StreamParser::CPDF_StreamParser(pdfium::span<const uint8_t> span,
const WeakPtr<ByteStringPool>& pPool)
- : m_Size(dwSize), m_Pos(0), m_WordSize(0), m_pBuf(pData), m_pPool(pPool) {}
+ : m_pPool(pPool), m_pBuf(span) {}
CPDF_StreamParser::~CPDF_StreamParser() {}
@@ -115,7 +114,7 @@ std::unique_ptr<CPDF_Stream> CPDF_StreamParser::ReadInlineStream(
CPDF_Document* pDoc,
std::unique_ptr<CPDF_Dictionary> pDict,
CPDF_Object* pCSObj) {
- if (m_Pos == m_Size)
+ if (m_Pos == m_pBuf.size())
return nullptr;
if (PDFCharIsWhitespace(m_pBuf[m_Pos]))
@@ -176,17 +175,17 @@ std::unique_ptr<CPDF_Stream> CPDF_StreamParser::ReadInlineStream(
std::unique_ptr<uint8_t, FxFreeDeleter> pData;
uint32_t dwStreamSize;
if (Decoder.IsEmpty()) {
- if (OrigSize > m_Size - m_Pos)
- OrigSize = m_Size - m_Pos;
+ if (OrigSize > m_pBuf.size() - m_Pos)
+ OrigSize = m_pBuf.size() - m_Pos;
pData.reset(FX_Alloc(uint8_t, OrigSize));
- memcpy(pData.get(), m_pBuf + m_Pos, OrigSize);
+ memcpy(pData.get(), &m_pBuf[m_Pos], OrigSize);
dwStreamSize = OrigSize;
m_Pos += OrigSize;
} else {
uint8_t* pIgnore = nullptr;
uint32_t dwDestSize = OrigSize;
dwStreamSize =
- DecodeInlineStream(m_pBuf + m_Pos, m_Size - m_Pos, width, height,
+ DecodeInlineStream(&m_pBuf[m_Pos], m_pBuf.size() - m_Pos, width, height,
Decoder, pParam, &pIgnore, &dwDestSize);
FX_Free(pIgnore);
if (static_cast<int>(dwStreamSize) < 0)
@@ -212,7 +211,7 @@ std::unique_ptr<CPDF_Stream> CPDF_StreamParser::ReadInlineStream(
}
m_Pos = dwSavePos;
pData.reset(FX_Alloc(uint8_t, dwStreamSize));
- memcpy(pData.get(), m_pBuf + m_Pos, dwStreamSize);
+ memcpy(pData.get(), &m_pBuf[m_Pos], dwStreamSize);
m_Pos += dwStreamSize;
}
pDict->SetNewFor<CPDF_Number>("Length", static_cast<int>(dwStreamSize));
@@ -603,5 +602,5 @@ ByteString CPDF_StreamParser::ReadHexString() {
}
bool CPDF_StreamParser::PositionIsInBounds() const {
- return m_Pos < m_Size;
+ return m_Pos < m_pBuf.size();
}
diff --git a/core/fpdfapi/page/cpdf_streamparser.h b/core/fpdfapi/page/cpdf_streamparser.h
index bdd07643ce..78727da481 100644
--- a/core/fpdfapi/page/cpdf_streamparser.h
+++ b/core/fpdfapi/page/cpdf_streamparser.h
@@ -16,14 +16,14 @@
#include "core/fpdfapi/parser/cpdf_stream.h"
#include "core/fxcrt/string_pool_template.h"
#include "core/fxcrt/weak_ptr.h"
+#include "third_party/base/span.h"
class CPDF_StreamParser {
public:
enum SyntaxType { EndOfData, Number, Keyword, Name, Others };
- CPDF_StreamParser(const uint8_t* pData, uint32_t dwSize);
- CPDF_StreamParser(const uint8_t* pData,
- uint32_t dwSize,
+ explicit CPDF_StreamParser(pdfium::span<const uint8_t> span);
+ CPDF_StreamParser(pdfium::span<const uint8_t> span,
const WeakPtr<ByteStringPool>& pPool);
~CPDF_StreamParser();
@@ -51,12 +51,11 @@ class CPDF_StreamParser {
ByteString ReadHexString();
bool PositionIsInBounds() const;
- uint32_t m_Size; // Length in bytes of m_pBuf.
- uint32_t m_Pos; // Current byte position within m_pBuf.
- uint32_t m_WordSize; // Current byte position within m_WordBuffer.
- const uint8_t* m_pBuf;
- std::unique_ptr<CPDF_Object> m_pLastObj;
+ uint32_t m_Pos = 0; // Current byte position within m_pBuf.
+ uint32_t m_WordSize = 0; // Current byte position within m_WordBuffer.
WeakPtr<ByteStringPool> m_pPool;
+ std::unique_ptr<CPDF_Object> m_pLastObj;
+ pdfium::span<const uint8_t> m_pBuf;
uint8_t m_WordBuffer[kMaxWordLength + 1]; // Include space for NUL.
};
diff --git a/core/fpdfapi/page/cpdf_streamparser_unittest.cpp b/core/fpdfapi/page/cpdf_streamparser_unittest.cpp
index 40a41befe1..d83fedcb7d 100644
--- a/core/fpdfapi/page/cpdf_streamparser_unittest.cpp
+++ b/core/fpdfapi/page/cpdf_streamparser_unittest.cpp
@@ -4,12 +4,13 @@
#include "core/fpdfapi/page/cpdf_streamparser.h"
#include "testing/gtest/include/gtest/gtest.h"
+#include "third_party/base/span.h"
TEST(cpdf_streamparser, ReadHexString) {
{
// Position out of bounds.
uint8_t data[] = "12ab>";
- CPDF_StreamParser parser(data, 5);
+ CPDF_StreamParser parser(data);
parser.SetPos(6);
EXPECT_EQ("", parser.ReadHexString());
}
@@ -17,7 +18,7 @@ TEST(cpdf_streamparser, ReadHexString) {
{
// Regular conversion.
uint8_t data[] = "1A2b>abcd";
- CPDF_StreamParser parser(data, 5);
+ CPDF_StreamParser parser(data);
EXPECT_EQ("\x1a\x2b", parser.ReadHexString());
EXPECT_EQ(5u, parser.GetPos());
}
@@ -25,7 +26,7 @@ TEST(cpdf_streamparser, ReadHexString) {
{
// Missing ending >
uint8_t data[] = "1A2b";
- CPDF_StreamParser parser(data, 5);
+ CPDF_StreamParser parser(data);
EXPECT_EQ("\x1a\x2b", parser.ReadHexString());
EXPECT_EQ(5u, parser.GetPos());
}
@@ -33,14 +34,14 @@ TEST(cpdf_streamparser, ReadHexString) {
{
// Uneven number of bytes.
uint8_t data[] = "1A2>asdf";
- CPDF_StreamParser parser(data, 5);
+ CPDF_StreamParser parser(data);
EXPECT_EQ("\x1a\x20", parser.ReadHexString());
EXPECT_EQ(4u, parser.GetPos());
}
{
uint8_t data[] = ">";
- CPDF_StreamParser parser(data, 5);
+ CPDF_StreamParser parser(data);
EXPECT_EQ("", parser.ReadHexString());
EXPECT_EQ(1u, parser.GetPos());
}
diff --git a/core/fxcrt/unowned_ptr.h b/core/fxcrt/unowned_ptr.h
index f9753cde37..b1c9c66b3e 100644
--- a/core/fxcrt/unowned_ptr.h
+++ b/core/fxcrt/unowned_ptr.h
@@ -32,7 +32,8 @@
//
// The array indexing operation [] is not supported on an unowned ptr,
// because an unowned ptr expresses a one to one relationship with some
-// other heap object.
+// other heap object. Use pdfium::span<> for the cases where indexing
+// into an unowned array is desired, which performs the same checks.
namespace fxcrt {
diff --git a/testing/libfuzzer/pdf_streamparser_fuzzer.cc b/testing/libfuzzer/pdf_streamparser_fuzzer.cc
index 46113d42c6..4d9a368916 100644
--- a/testing/libfuzzer/pdf_streamparser_fuzzer.cc
+++ b/testing/libfuzzer/pdf_streamparser_fuzzer.cc
@@ -7,9 +7,10 @@
#include "core/fpdfapi/page/cpdf_streamparser.h"
#include "core/fpdfapi/parser/cpdf_object.h"
+#include "third_party/base/span.h"
extern "C" int LLVMFuzzerTestOneInput(const uint8_t* data, size_t size) {
- CPDF_StreamParser parser(data, size);
+ CPDF_StreamParser parser(pdfium::make_span(data, size));
while (std::unique_ptr<CPDF_Object> pObj =
parser.ReadNextObject(true, false, 0))
continue;
diff --git a/third_party/BUILD.gn b/third_party/BUILD.gn
index 1bff4347df..8b212e1622 100644
--- a/third_party/BUILD.gn
+++ b/third_party/BUILD.gn
@@ -562,6 +562,7 @@ jumbo_source_set("pdfium_base") {
"base/numerics/safe_math_impl.h",
"base/optional.h",
"base/ptr_util.h",
+ "base/span.h",
"base/stl_util.h",
"base/sys_byteorder.h",
"base/template_util.h",
diff --git a/third_party/base/span.h b/third_party/base/span.h
new file mode 100644
index 0000000000..d8d8f29e7c
--- /dev/null
+++ b/third_party/base/span.h
@@ -0,0 +1,336 @@
+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef PDFIUM_THIRD_PARTY_BASE_SPAN_H_
+#define PDFIUM_THIRD_PARTY_BASE_SPAN_H_
+
+#include <stddef.h>
+
+#include <algorithm>
+#include <array>
+#include <iterator>
+#include <type_traits>
+#include <utility>
+
+#include "third_party/base/logging.h"
+
+namespace pdfium {
+
+template <typename T>
+class span;
+
+namespace internal {
+
+template <typename T>
+struct IsSpanImpl : std::false_type {};
+
+template <typename T>
+struct IsSpanImpl<span<T>> : std::true_type {};
+
+template <typename T>
+using IsSpan = IsSpanImpl<typename std::decay<T>::type>;
+
+template <typename T>
+struct IsStdArrayImpl : std::false_type {};
+
+template <typename T, size_t N>
+struct IsStdArrayImpl<std::array<T, N>> : std::true_type {};
+
+template <typename T>
+using IsStdArray = IsStdArrayImpl<typename std::decay<T>::type>;
+
+template <typename From, typename To>
+using IsLegalSpanConversion = std::is_convertible<From*, To*>;
+
+template <typename Container, typename T>
+using ContainerHasConvertibleData =
+ IsLegalSpanConversion<typename std::remove_pointer<decltype(
+ std::declval<Container>().data())>::type,
+ T>;
+template <typename Container>
+using ContainerHasIntegralSize =
+ std::is_integral<decltype(std::declval<Container>().size())>;
+
+template <typename From, typename To>
+using EnableIfLegalSpanConversion =
+ typename std::enable_if<IsLegalSpanConversion<From, To>::value>::type;
+
+// SFINAE check if Container can be converted to a span<T>. Note that the
+// implementation details of this check differ slightly from the requirements in
+// the working group proposal: in particular, the proposal also requires that
+// the container conversion constructor participate in overload resolution only
+// if two additional conditions are true:
+//
+// 1. Container implements operator[].
+// 2. Container::value_type matches remove_const_t<element_type>.
+//
+// The requirements are relaxed slightly here: in particular, not requiring (2)
+// means that an immutable span can be easily constructed from a mutable
+// container.
+template <typename Container, typename T>
+using EnableIfSpanCompatibleContainer =
+ typename std::enable_if<!internal::IsSpan<Container>::value &&
+ !internal::IsStdArray<Container>::value &&
+ ContainerHasConvertibleData<Container, T>::value &&
+ ContainerHasIntegralSize<Container>::value>::type;
+
+template <typename Container, typename T>
+using EnableIfConstSpanCompatibleContainer =
+ typename std::enable_if<std::is_const<T>::value &&
+ !internal::IsSpan<Container>::value &&
+ !internal::IsStdArray<Container>::value &&
+ ContainerHasConvertibleData<Container, T>::value &&
+ ContainerHasIntegralSize<Container>::value>::type;
+
+} // namespace internal
+
+// A span is a value type that represents an array of elements of type T. Since
+// it only consists of a pointer to memory with an associated size, it is very
+// light-weight. It is cheap to construct, copy, move and use spans, so that
+// users are encouraged to use it as a pass-by-value parameter. A span does not
+// own the underlying memory, so care must be taken to ensure that a span does
+// not outlive the backing store.
+//
+// span is somewhat analogous to StringPiece, but with arbitrary element types,
+// allowing mutation if T is non-const.
+//
+// span is implicitly convertible from C++ arrays, as well as most [1]
+// container-like types that provide a data() and size() method (such as
+// std::vector<T>). A mutable span<T> can also be implicitly converted to an
+// immutable span<const T>.
+//
+// Consider using a span for functions that take a data pointer and size
+// parameter: it allows the function to still act on an array-like type, while
+// allowing the caller code to be a bit more concise.
+//
+// For read-only data access pass a span<const T>: the caller can supply either
+// a span<const T> or a span<T>, while the callee will have a read-only view.
+// For read-write access a mutable span<T> is required.
+//
+// Without span:
+// Read-Only:
+// // std::string HexEncode(const uint8_t* data, size_t size);
+// std::vector<uint8_t> data_buffer = GenerateData();
+// std::string r = HexEncode(data_buffer.data(), data_buffer.size());
+//
+// Mutable:
+// // ssize_t SafeSNPrintf(char* buf, size_t N, const char* fmt, Args...);
+// char str_buffer[100];
+// SafeSNPrintf(str_buffer, sizeof(str_buffer), "Pi ~= %lf", 3.14);
+//
+// With span:
+// Read-Only:
+// // std::string HexEncode(base::span<const uint8_t> data);
+// std::vector<uint8_t> data_buffer = GenerateData();
+// std::string r = HexEncode(data_buffer);
+//
+// Mutable:
+// // ssize_t SafeSNPrintf(base::span<char>, const char* fmt, Args...);
+// char str_buffer[100];
+// SafeSNPrintf(str_buffer, "Pi ~= %lf", 3.14);
+//
+// Spans with "const" and pointers
+// -------------------------------
+//
+// Const and pointers can get confusing. Here are vectors of pointers and their
+// corresponding spans (you can always make the span "more const" too):
+//
+// const std::vector<int*> => base::span<int* const>
+// std::vector<const int*> => base::span<const int*>
+// const std::vector<const int*> => base::span<const int* const>
+//
+// Differences from the working group proposal
+// -------------------------------------------
+//
+// https://wg21.link/P0122 is the latest working group proposal, Chromium
+// currently implements R6. The biggest difference is span does not support a
+// static extent template parameter. Other differences are documented in
+// subsections below.
+//
+// Differences from [views.constants]:
+// - no dynamic_extent constant
+//
+// Differences from [span.objectrep]:
+// - no as_bytes()
+// - no as_writeable_bytes()
+//
+// Differences in constants and types:
+// - no element_type type alias
+// - no index_type type alias
+// - no different_type type alias
+// - no extent constant
+//
+// Differences from [span.cons]:
+// - no constructor from a pointer range
+// - no constructor from std::array
+//
+// Differences from [span.sub]:
+// - no templated first()
+// - no templated last()
+// - no templated subspan()
+// - using size_t instead of ptrdiff_t for indexing
+//
+// Differences from [span.obs]:
+// - no size_bytes()
+// - using size_t instead of ptrdiff_t to represent size()
+//
+// Differences from [span.elem]:
+// - no operator ()()
+// - using size_t instead of ptrdiff_t for indexing
+
+// [span], class template span
+template <typename T>
+class span {
+ public:
+ using value_type = typename std::remove_cv<T>::type;
+ using pointer = T*;
+ using reference = T&;
+ using iterator = T*;
+ using const_iterator = const T*;
+ using reverse_iterator = std::reverse_iterator<iterator>;
+ using const_reverse_iterator = std::reverse_iterator<const_iterator>;
+
+ // [span.cons], span constructors, copy, assignment, and destructor
+ constexpr span() noexcept : data_(nullptr), size_(0) {}
+ constexpr span(T* data, size_t size) noexcept : data_(data), size_(size) {}
+ // TODO(dcheng): Implement construction from a |begin| and |end| pointer.
+ template <size_t N>
+ constexpr span(T (&array)[N]) noexcept : span(array, N) {}
+ // TODO(dcheng): Implement construction from std::array.
+ // Conversion from a container that provides |T* data()| and |integral_type
+ // size()|.
+ template <typename Container,
+ typename = internal::EnableIfSpanCompatibleContainer<Container, T>>
+ constexpr span(Container& container)
+ : span(container.data(), container.size()) {}
+ template <
+ typename Container,
+ typename = internal::EnableIfConstSpanCompatibleContainer<Container, T>>
+ span(const Container& container) : span(container.data(), container.size()) {}
+ constexpr span(const span& other) noexcept = default;
+ // Conversions from spans of compatible types: this allows a span<T> to be
+ // seamlessly used as a span<const T>, but not the other way around.
+ template <typename U, typename = internal::EnableIfLegalSpanConversion<U, T>>
+ constexpr span(const span<U>& other) : span(other.data(), other.size()) {}
+ span& operator=(const span& other) noexcept = default;
+ ~span() noexcept = default;
+
+ // [span.sub], span subviews
+ const span first(size_t count) const {
+ CHECK(count <= size_);
+ return span(data_, count);
+ }
+
+ const span last(size_t count) const {
+ CHECK(count <= size_);
+ return span(data_ + (size_ - count), count);
+ }
+
+ const span subspan(size_t pos, size_t count = -1) const {
+ const auto npos = static_cast<size_t>(-1);
+ CHECK(pos <= size_);
+ CHECK(count == npos || count <= size_ - pos);
+ return span(data_ + pos, count == npos ? size_ - pos : count);
+ }
+
+ // [span.obs], span observers
+ constexpr size_t size() const noexcept { return size_; }
+ constexpr bool empty() const noexcept { return size_ == 0; }
+
+ // [span.elem], span element access
+ const T& operator[](size_t index) const noexcept {
+ CHECK(index < size_);
+ return data_[index];
+ }
+ constexpr T* data() const noexcept { return data_; }
+
+ // [span.iter], span iterator support
+ constexpr iterator begin() const noexcept { return data_; }
+ constexpr iterator end() const noexcept { return data_ + size_; }
+
+ constexpr const_iterator cbegin() const noexcept { return begin(); }
+ constexpr const_iterator cend() const noexcept { return end(); }
+
+ constexpr reverse_iterator rbegin() const noexcept {
+ return reverse_iterator(end());
+ }
+ constexpr reverse_iterator rend() const noexcept {
+ return reverse_iterator(begin());
+ }
+
+ constexpr const_reverse_iterator crbegin() const noexcept {
+ return const_reverse_iterator(cend());
+ }
+ constexpr const_reverse_iterator crend() const noexcept {
+ return const_reverse_iterator(cbegin());
+ }
+
+ private:
+ T* data_;
+ size_t size_;
+};
+
+// [span.comparison], span comparison operators
+// Relational operators. Equality is a element-wise comparison.
+template <typename T>
+constexpr bool operator==(span<T> lhs, span<T> rhs) noexcept {
+ return lhs.size() == rhs.size() &&
+ std::equal(lhs.cbegin(), lhs.cend(), rhs.cbegin());
+}
+
+template <typename T>
+constexpr bool operator!=(span<T> lhs, span<T> rhs) noexcept {
+ return !(lhs == rhs);
+}
+
+template <typename T>
+constexpr bool operator<(span<T> lhs, span<T> rhs) noexcept {
+ return std::lexicographical_compare(lhs.cbegin(), lhs.cend(), rhs.cbegin(),
+ rhs.cend());
+}
+
+template <typename T>
+constexpr bool operator<=(span<T> lhs, span<T> rhs) noexcept {
+ return !(rhs < lhs);
+}
+
+template <typename T>
+constexpr bool operator>(span<T> lhs, span<T> rhs) noexcept {
+ return rhs < lhs;
+}
+
+template <typename T>
+constexpr bool operator>=(span<T> lhs, span<T> rhs) noexcept {
+ return !(lhs < rhs);
+}
+
+// Type-deducing helpers for constructing a span.
+template <typename T>
+constexpr span<T> make_span(T* data, size_t size) noexcept {
+ return span<T>(data, size);
+}
+
+template <typename T, size_t N>
+constexpr span<T> make_span(T (&array)[N]) noexcept {
+ return span<T>(array);
+}
+
+template <typename Container,
+ typename T = typename Container::value_type,
+ typename = internal::EnableIfSpanCompatibleContainer<Container, T>>
+constexpr span<T> make_span(Container& container) {
+ return span<T>(container);
+}
+
+template <
+ typename Container,
+ typename T = typename std::add_const<typename Container::value_type>::type,
+ typename = internal::EnableIfConstSpanCompatibleContainer<Container, T>>
+constexpr span<T> make_span(const Container& container) {
+ return span<T>(container);
+}
+
+} // namespace pdfium
+
+#endif // PDFIUM_THIRD_PARTY_BASE_SPAN_H_