String interning de-duplicates strings in memory. It is useful when dealing with strings that you expect to recur often. The price for de-duplication is typically an O(N) hashing operation on the string to convert it to some stable, string-specific ID.

A common use-case for interned strings is to represent syntactic identifiers (variable names) in compilers and interpreters. Each variable name is stored as a single interned string. Symbol lookups in symbol tables, field name queries on records/structs, and import resolution all rely on interned strings to work more efficiently.

Another use-case for interned strings is as a concise identifier for objects in game-engines. Names can be interned to turn them into integers. Subsequent queries using these names can supply the interned string, reducing the cost of hashing and comparing the strings.

This implementation uses a hash-map to associate every interned string with a unique ID. When an intstr instance is constructed, we lock the hash-map and check if the string has been interned already. If so, we return the previously returned ID. Otherwise, we mint a new ID, update the global hash-map, and return the new ID.

This means that each intstr points to exactly one string. It also allows you to recover the original string representation from an ID by performing a reverse-lookup on the aforementioned map using a supplementary lookup table.

The global hash-map is synchronized by a mutex, making it thread-safe. We acquire shared locks to the global mutex when checking if a string is already interned, only taking exclusive locks when we need to update the global state with an unencountered string. Since string interning is employed when the strings are expected to occur repeatedly, we expect to rarely take the global exclusive lock.

Finally, we expose the underlying integer handle and ABI as part of the API. We guarantee the integer will increment monotonically, that the 0 ID will always refer to an empty string, and that the specified integer type (uint32_t by default) is all each intstr instance will ever contain. This makes our class suitable for use in tagged-object interpreters where the integer must be packed and unpacked reliably.

The implementation is simple, thread-safe, and is contained in a single header. It is easy to modify and patch, e.g. changing to 16-bit or 64-bit integers.

Happy hacking!