Index: third_party/zlib/google/zip_reader.cc |
diff --git a/third_party/zlib/google/zip_reader.cc b/third_party/zlib/google/zip_reader.cc |
index 83086a3467fb0e21201edf36aaa80a8908ab839b..441a09835286c7813c0e93998a78390bba39a4ee 100644 |
--- a/third_party/zlib/google/zip_reader.cc |
+++ b/third_party/zlib/google/zip_reader.cc |
@@ -7,10 +7,12 @@ |
#include "base/bind.h" |
#include "base/files/file.h" |
#include "base/logging.h" |
+#include "base/macros.h" |
#include "base/message_loop/message_loop.h" |
#include "base/strings/string_util.h" |
#include "base/strings/utf_string_conversions.h" |
#include "base/thread_task_runner_handle.h" |
+#include "build/build_config.h" |
#include "third_party/zlib/google/zip_internal.h" |
#if defined(USE_SYSTEM_MINIZIP) |
@@ -438,9 +440,9 @@ bool ZipReader::ExtractCurrentEntryToString(size_t max_read_bytes, |
// correct. However, we need to assume that the uncompressed size could be |
// incorrect therefore this function needs to read as much data as possible. |
std::string contents; |
- contents.reserve(static_cast<size_t>(std::min( |
- static_cast<int64>(max_read_bytes), |
- current_entry_info()->original_size()))); |
+ contents.reserve( |
+ static_cast<size_t>(std::min(static_cast<int64_t>(max_read_bytes), |
+ current_entry_info()->original_size()))); |
StringWriterDelegate writer(max_read_bytes, &contents); |
if (!ExtractCurrentEntry(&writer)) |
@@ -476,7 +478,7 @@ void ZipReader::ExtractChunk(base::File output_file, |
const SuccessCallback& success_callback, |
const FailureCallback& failure_callback, |
const ProgressCallback& progress_callback, |
- const int64 offset) { |
+ const int64_t offset) { |
char buffer[internal::kZipBufSize]; |
const int num_bytes_read = unzReadCurrentFile(zip_file_, |
@@ -497,7 +499,7 @@ void ZipReader::ExtractChunk(base::File output_file, |
return; |
} |
- int64 current_progress = offset + num_bytes_read; |
+ int64_t current_progress = offset + num_bytes_read; |
progress_callback.Run(current_progress); |