improved string padded (simdjson#440)

DBJDBJ · lemire · commit 85e84fc1facd · 2020-01-10T10:15:48.000-05:00
* dirent portable latest version

* improved

std::string argument passed by const reference
ctor added with std::string_view  argument
`allocate_padded_buffer()`  moved here with **optional** check on `length &lt; 1`

* allocate_padded_buffer moved to padded_string.h
diff --git a/include/simdjson/padded_string.h b/include/simdjson/padded_string.h
@@ -5,39 +5,84 @@
 #include <memory>
 #include <string>
 
+// padded buffer can be made with length < 1
+#define SIMDJSON_OK_EMPTY_PADDED_BUFFER 1
+
 namespace simdjson {
 // low-level function to allocate memory with padding so we can read passed the
 // "length" bytes safely. if you must provide a pointer to some data, create it
 // with this function: length is the max. size in bytes of the string caller is
 // responsible to free the memory (free(...))
-char *allocate_padded_buffer(size_t length);
+inline char *allocate_padded_buffer(size_t length) noexcept {
+
+#ifndef NDEBUG 
+#ifndef SIMDJSON_OK_EMPTY_PADDED_BUFFER 
+  if (length < 1) {
+    errno = EINVAL;
+    perror("simdjson::allocate_padded_buffer() length argument is less than 1");
+    return nullptr;
+  }
+  #endif // SIMDJSON_OK_EMPTY_PADDED_STRING
+#endif // NDEBUG
+
+  // we could do a simple malloc
+  // return (char *) malloc(length + SIMDJSON_PADDING);
+  // However, we might as well align to cache lines...
+  size_t totalpaddedlength = length + SIMDJSON_PADDING;
+  char *padded_buffer = aligned_malloc_char(64, totalpaddedlength);
+
+  #ifndef NDEBUG
+  if (padded_buffer == nullptr) {
+    errno = EINVAL;
+    perror("simdjson::allocate_padded_buffer() aligned_malloc_char() failed");
+    return nullptr;
+  }
+#endif // NDEBUG
+
+  memset(padded_buffer + length, 0, totalpaddedlength - length);
+  return padded_buffer;
+} // allocate_padded_buffer
 
 // Simple string with padded allocation.
 // We deliberately forbid copies, users should rely on swap or move
 // constructors.
-class padded_string {
-public:
+struct padded_string final {
+
   explicit padded_string() noexcept : viable_size(0), data_ptr(nullptr) {}
+
   explicit padded_string(size_t length) noexcept
       : viable_size(length), data_ptr(allocate_padded_buffer(length)) {
 
     if (data_ptr != nullptr)
       data_ptr[length] = '\0'; // easier when you need a c_str
   }
+
   explicit padded_string(char *data, size_t length) noexcept
       : viable_size(length), data_ptr(allocate_padded_buffer(length)) {
-    if (data_ptr != nullptr) {
+    if (data != nullptr) {
       memcpy(data_ptr, data, length);
       data_ptr[length] = '\0'; // easier when you need a c_str
     }
   }
-  padded_string(std::string s) noexcept
-      : viable_size(s.size()), data_ptr(allocate_padded_buffer(s.size())) {
+
+  // note: do not pass std::string arguments by value
+  padded_string(const std::string & str_ ) noexcept
+      : viable_size(str_.size()), data_ptr(allocate_padded_buffer(str_.size())) {
+    if (data_ptr != nullptr) {
+      memcpy(data_ptr, str_.data(), str_.size());
+      data_ptr[str_.size()] = '\0'; // easier when you need a c_str
+    }
+  }
+
+  // note: do pass std::string_view arguments by value
+  padded_string(std::string_view sv_) noexcept
+      : viable_size(sv_.size()), data_ptr(allocate_padded_buffer(sv_.size())) {
     if (data_ptr != nullptr) {
-      memcpy(data_ptr, s.data(), s.size());
-      data_ptr[s.size()] = '\0'; // easier when you need a c_str
+      memcpy(data_ptr, sv_.data(), sv_.size());
+      data_ptr[sv_.size()] = '\0'; // easier when you need a c_str
     }
   }
+
   padded_string(padded_string &&o) noexcept
       : viable_size(o.viable_size), data_ptr(o.data_ptr) {
     o.data_ptr = nullptr; // we take ownership
@@ -60,21 +105,26 @@ class padded_string {
     o.viable_size = tmp_viable_size;
   }
 
-  ~padded_string() { aligned_free_char(data_ptr); }
+  ~padded_string() {
+      aligned_free_char(data_ptr);
+      this->data_ptr = nullptr;
+  }
 
-  size_t size() const { return viable_size; }
+  size_t size() const  { return viable_size; }
 
-  size_t length() const { return viable_size; }
+  size_t length() const  { return viable_size; }
 
-  char *data() const { return data_ptr; }
+  char *data() const  { return data_ptr; }
 
 private:
   padded_string &operator=(const padded_string &o) = delete;
   padded_string(const padded_string &o) = delete;
 
-  size_t viable_size;
-  char *data_ptr;
-};
+  size_t viable_size ;
+  char *data_ptr{nullptr};
+
+}; // padded_string
+
 } // namespace simdjson
 
 #endif
diff --git a/src/jsonioutil.cpp b/src/jsonioutil.cpp
@@ -4,15 +4,6 @@
 #include <climits>
 
 namespace simdjson {
-char *allocate_padded_buffer(size_t length) {
-  // we could do a simple malloc
-  // return (char *) malloc(length + SIMDJSON_PADDING);
-  // However, we might as well align to cache lines...
-  size_t totalpaddedlength = length + SIMDJSON_PADDING;
-  char *padded_buffer = aligned_malloc_char(64, totalpaddedlength);
-  memset(padded_buffer + length, 0, totalpaddedlength - length);
-  return padded_buffer;
-}
 
 padded_string get_corpus(const std::string &filename) {
   std::FILE *fp = std::fopen(filename.c_str(), "rb");
