Struct HistoryEngine 

pub struct HistoryEngine { /* private fields */ }

SQLite-backed history engine. Replaces the in-memory histent doubly-linked list + histfile flat-file pair from Src/hist.c — same logical history but with FTS5 search, frequency tracking, and per-directory context.

Implementations

impl HistoryEngine

pub fn new() -> Result<Self>

new — see implementation.

pub fn in_memory() -> Result<Self>

in_memory — see implementation.

pub fn db_path() -> PathBuf

$ZSHRS_HOME/zshrs_history.db — sqlite index that powers FTS5 search, frequency tracking, dedup. Hidden under .db so the user-facing artifact is the flat-text mirror at zshrs_history (see text_path), zsh-compatible so muscle memory + cat / grep / external history tools all keep working.

The daemon owns its OWN history db at ~/.zshrs/history.db (different schema, daemon-only writer); shells append to it via history_append IPC. This shell-side db is the fallback path used when the daemon is absent.

pub fn text_path() -> PathBuf

$ZSHRS_HOME/zshrs_history — flat text mirror, one line per command in zsh extended-history format:

: <unix_ts>:<duration>;<command>

A newline inside a multi-line command is written as a backslash followed by the newline, so one record can span several physical lines (zsh/Src/hist.c:savehistfile). Each record is appended once, by update_last, when the command has finished and its duration is known; the file is never rewritten in place, so concurrent shells cannot clobber each other’s records. The sqlite index at zshrs_history.db is the query side; a missing or empty text file is rehydrated from it on open.

pub fn add(&self, command: &str, cwd: Option<&str>) -> Result<i64>

Add a command to history, updating frequency if it already exists

pub fn update_last( &self, id: i64, duration_ms: i64, exit_code: i32, ) -> Result<()>

Update the duration and exit code of the last command

pub fn search(&self, query: &str, limit: usize) -> Result<Vec<HistoryEntry>>

Search history with FTS5 (fuzzy/substring matching)

pub fn search_prefix( &self, prefix: &str, limit: usize, ) -> Result<Vec<HistoryEntry>>

Search history with prefix matching (for up-arrow completion)

pub fn search_prefix_cs( &self, prefix: &str, limit: usize, ) -> Result<Vec<HistoryEntry>>

Case-SENSITIVE prefix probe for the per-keystroke ZLE paths (autosuggest, up-arrow search). Uses GLOB, which is case-sensitive and therefore eligible for the SQLite prefix optimization over idx_history_command (BINARY collation). search_prefix’s LIKE is case-insensitive by default and CANNOT use that index — at 500k+ rows it full-scans, which is far too slow to run on every keystroke.

pub fn recent(&self, limit: usize) -> Result<Vec<HistoryEntry>>

Get recent history entries

pub fn for_directory( &self, cwd: &str, limit: usize, ) -> Result<Vec<HistoryEntry>>

Get history for a specific directory

pub fn delete(&self, id: i64) -> Result<()>

Delete a history entry

pub fn clear(&self) -> Result<()>

Clear all history

pub fn count(&self) -> Result<i64>

Get total history count

pub fn count_estimate(&self) -> Result<i64>

Row-count ESTIMATE that does not scan the table.

MAX(rowid) is resolved by descending to the rightmost leaf of the rowid b-tree, so it is independent of table size where COUNT(*) is linear in it (37 ms at 775k rows). It equals the true count for an append-only history and over-reports by the number of deleted rows otherwise — good enough for a log line, never used for arithmetic on history events.

pub fn get_by_offset(&self, offset: usize) -> Result<Option<HistoryEntry>>

Get entry by index from end (0 = most recent, like !-1)

pub fn get_by_number(&self, num: i64) -> Result<Option<HistoryEntry>>

Get entry by absolute history number (like !123)