Why FBString Beats std::string: Three‑Tier Storage and Jemalloc Magic

FBString Overview

fbstring is a drop‑in replacement for std::string . Its main benefit is significantly increased performance on virtually all important primitives, achieved through a three‑tier storage strategy and cooperation with the memory allocator, especially jemalloc.

Storage Strategies

SSO (Small‑String Optimization) : Uses a stack‑based buffer to store up to 23 characters directly in the class’s array member.

Eager Copy : Stores strings up to 254 characters on the heap allocated by malloc .

Copy‑On‑Write (COW) : For strings longer than 254 characters, introduces reference counting to avoid unnecessary copies.

Core Implementation

fbstring_core is the implementation core providing all operation interfaces and realizing the three‑tier storage and memory allocation strategy.

The class can be in one of three states— small , medium , or large —and transitions between them during construction, assignment, expansion, or shrinkage. The category() method returns the current state.

Data members are stored in a union where small_ handles the small state and MediumLarge handles medium and large states. In the small state, the last byte stores maxSmallSize - currentSize .

RefCounted supports COW in the large state. It uses std::atomic for the reference count, and data_ points to the counted entity. The function fromData(Char* p) obtains the corresponding RefCounted pointer from the entity pointer.

Constructor

Construction follows the three‑state strategy, initializing the appropriate storage based on the string length.

Move Constructor

Swap Function

void swap(fbstring_core & rhs) { auto const t = ml_; ml_ = rhs.ml_; rhs.ml_ = t; }

Related Links

http://en.cppreference.com/w/cpp/atomic/atomic

http://en.cppreference.com/w/cpp/language/move_constructor