Struct Dictionary 

pub struct Dictionary {
    pub prefix_dictionary: Arc<PrefixDictionary>,
    pub connection_cost_matrix: Arc<ConnectionCostMatrix>,
    pub character_definition: CharacterDefinition,
    pub unknown_dictionary: UnknownDictionary,
    pub metadata: Metadata,
}

prefix_dictionary and connection_cost_matrix are Arc-wrapped so that Dictionary::clone() is O(1) regardless of load method (embedded, mmap, or plain filesystem read) – these two components dominate a dictionary’s memory footprint (tens to hundreds of MB), and nothing in this codebase mutates them after construction.

Fields

prefix_dictionary: Arc<PrefixDictionary>

connection_cost_matrix: Arc<ConnectionCostMatrix>

character_definition: CharacterDefinition

unknown_dictionary: UnknownDictionary

metadata: Metadata

Implementations

impl Dictionary

pub fn unknown_word_details(&self, word_id: usize) -> Vec<&str>

Retrieve the detail fields (POS, etc.) for an unknown word entry.

pub fn word_details(&self, word_id: usize) -> Vec<&str>

pub fn load_from_path(dict_path: &Path) -> LinderaResult<Self>

Load dictionary from a directory containing dictionary files

pub fn load_from_path_with_options( dict_path: &Path, use_mmap: bool, ) -> LinderaResult<Self>

Load dictionary from a directory with options

use_mmap (when the mmap feature is enabled) routes connection_cost_matrix and prefix_dictionary through memory-mapped reads instead of plain file reads. This does not make either component lazily memory-resident at runtime: ConnectionCostMatrix always eagerly decodes into an owned Vec<i16>, and PrefixDictionary’s double-array trie (da) is always eagerly deserialized into owned daachorse structures (only PrefixDictionary’s vals_data/words_idx_data/words_data remain genuinely mmap-backed and are read lazily at lookup time). metadata and character_definition and unknown_dictionary are always plain-read regardless of this flag. In short, use_mmap only avoids the initial file-read syscall/allocation for the two components it covers — it does not provide OS-level lazy paging for tokenization. Separately, Dictionary::clone() is O(1) regardless of use_mmap, since prefix_dictionary/connection_cost_matrix are Arc-wrapped.

pub fn save_to_path(&self, dict_path: &Path) -> LinderaResult<()>

Save dictionary to a directory

Trait Implementations

impl Clone for Dictionary

fn clone(&self) -> Dictionary

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.