[breaking] Change internal model serialization to UBJSON. (#7556)

* Use typed array for models.
* Change the memory snapshot format.
* Add new C API for saving to raw format.

src/common/config.h

@@ -111,7 +111,7 @@ class ConfigParser {
     const auto last_char = str.find_last_not_of(" \t\n\r");
     if (first_char == std::string::npos) {
       // Every character in str is a whitespace
-      return std::string();
+      return {};
     }
     CHECK_NE(last_char, std::string::npos);
     const auto substr_len = last_char + 1 - first_char;

src/common/io.cc

@@ -1,5 +1,5 @@
 /*!
- * Copyright (c) by XGBoost Contributors 2019
+ * Copyright (c) by XGBoost Contributors 2019-2022
  */
 #if defined(__unix__)
 #include <sys/stat.h>
@@ -52,7 +52,7 @@ size_t PeekableInStream::PeekRead(void* dptr, size_t size) {
 
 FixedSizeStream::FixedSizeStream(PeekableInStream* stream) : PeekableInStream(stream), pointer_{0} {
   size_t constexpr kInitialSize = 4096;
-  size_t size {kInitialSize}, total {0};
+  size_t size{kInitialSize}, total{0};
   buffer_.clear();
   while (true) {
     buffer_.resize(size);
@@ -142,5 +142,18 @@ std::string LoadSequentialFile(std::string uri, bool stream) {
   buffer.resize(total);
   return buffer;
 }
+
+std::string FileExtension(std::string fname, bool lower) {
+  if (lower) {
+    std::transform(fname.begin(), fname.end(), fname.begin(),
+                   [](char c) { return std::tolower(c); });
+  }
+  auto splited = Split(fname, '.');
+  if (splited.size() > 1) {
+    return splited.back();
+  } else {
+    return "";
+  }
+}
 }  // namespace common
 }  // namespace xgboost

src/common/io.h

@@ -1,5 +1,5 @@
 /*!
- * Copyright 2014 by Contributors
+ * Copyright by XGBoost Contributors 2014-2022
  * \file io.h
  * \brief general stream interface for serialization, I/O
  * \author Tianqi Chen
@@ -86,15 +86,31 @@ class FixedSizeStream : public PeekableInStream {
  */
 std::string LoadSequentialFile(std::string uri, bool stream = false);
 
-inline std::string FileExtension(std::string const& fname) {
-  auto splited = Split(fname, '.');
-  if (splited.size() > 1) {
-    return splited.back();
-  } else {
-    return "";
-  }
-}
+/**
+ * \brief Get file extension from file name.
+ *
+ * \param  lower Return in lower case.
+ *
+ * \return File extension without the `.`
+ */
+std::string FileExtension(std::string fname, bool lower = true);
 
+/**
+ * \brief Read the whole buffer from dmlc stream.
+ */
+inline std::string ReadAll(dmlc::Stream* fi, PeekableInStream* fp) {
+  std::string buffer;
+  if (auto fixed_size = dynamic_cast<common::MemoryFixSizeBuffer*>(fi)) {
+    fixed_size->Seek(common::MemoryFixSizeBuffer::kSeekEnd);
+    size_t size = fixed_size->Tell();
+    buffer.resize(size);
+    fixed_size->Seek(0);
+    CHECK_EQ(fixed_size->Read(&buffer[0], size), size);
+  } else {
+    FixedSizeStream{fp}.Take(&buffer);
+  }
+  return buffer;
+}
 }  // namespace common
 }  // namespace xgboost
 #endif  // XGBOOST_COMMON_IO_H_