Fix #954. safe_memcmp in tinystring.

UBSan was complaining about std::memcmp being called with nullptr
arguments even when the count == 0. This adds a "safe" wrapper that will
just return 0 if the count is 0 or if either of the pointers are nullptr.

Author: ThisIsFineTM

src/tools.h

@@ -22,6 +22,8 @@
 #ifndef OPENZIM_LIBZIM_TOOLS_H
 #define OPENZIM_LIBZIM_TOOLS_H
 
+#include <cstddef> // size_t
+#include <cstring> // memcmp
 #include <string>
 #include <tuple>
 #include <map>
@@ -71,6 +73,17 @@ namespace zim {
     return count;
   }
 
+  // Work around for errors reported by too picky UB sanitizer tools on std::memcmp()
+  // being called with a nullptr pointer argument even though count==0. Explicit
+  // nullptr checks to satisfy other picky compilers.
+  template <typename T>
+  int safe_memcmp(T const *const lhs, T const *const rhs, std::size_t count) noexcept {
+      if(count == 0UL || lhs == nullptr || rhs == nullptr) {
+          return 0;
+      }
+      return std::memcmp(lhs, rhs, count);
+  }
+
   namespace writer {
     class Dirent;
     bool isFrontArticle(const Dirent* dirent, const Hints& hints);

src/writer/tinyString.h

@@ -20,6 +20,7 @@
 #ifndef ZIM_WRITER_TINYSTRING_H
 #define ZIM_WRITER_TINYSTRING_H
 
+#include "../tools.h"
 #include "../zim_types.h"
 #include <cstring>
 
@@ -66,11 +67,11 @@ namespace zim
         size_t size() const { return m_size; }
         const char* const data() const { return m_data; }
         bool operator==(const TinyString& other) const {
-          return (m_size == other.m_size) && (std::memcmp(m_data, other.m_data, m_size) == 0);
+          return (m_size == other.m_size) && (safe_memcmp(m_data, other.m_data, m_size) == 0);
         }
         bool operator<(const TinyString& other) const {
-          auto min_size = std::min(m_size, other.m_size);
-          auto ret = std::memcmp(m_data, other.m_data, min_size);
+          const auto min_size = std::min(m_size, other.m_size);
+          const auto ret = safe_memcmp(m_data, other.m_data, min_size);
           if (ret == 0) {
             return m_size < other.m_size;
           } else {