Optimize named params via caching and use of smallvec

This commit is contained in:
Thom Chiovoloni 2020-04-13 23:47:12 -07:00 committed by Thom Chiovoloni
parent cf9b6e9ae2
commit 552416039e
7 changed files with 213 additions and 12 deletions

View File

@ -99,6 +99,7 @@ fallible-iterator = "0.2"
fallible-streaming-iterator = "0.1"
memchr = "2.2.0"
uuid = { version = "0.8", optional = true }
smallvec = "1.3"
[dev-dependencies]
doc-comment = "0.3"

View File

@ -128,6 +128,8 @@ mod version;
#[cfg(feature = "vtab")]
pub mod vtab;
pub(crate) mod util;
// Number of cached prepared statements we'll hold on to.
const STATEMENT_CACHE_DEFAULT_CAPACITY: usize = 16;
/// To be used when your statement has no [parameter](https://sqlite.org/lang_expr.html#varparam).
@ -274,7 +276,7 @@ fn path_to_cstring(p: &Path) -> Result<CString> {
#[cfg(not(unix))]
fn path_to_cstring(p: &Path) -> Result<CString> {
let s = p.to_str().ok_or_else(|| Error::InvalidPath(p.to_owned()))?;
str_to_cstring(s)
Ok(CString::new(s)?)
}
/// Name for a database within a SQLite connection.

View File

@ -7,11 +7,11 @@ use std::ptr;
// Private newtype for raw sqlite3_stmts that finalize themselves when dropped.
#[derive(Debug)]
pub struct RawStatement(*mut ffi::sqlite3_stmt, bool);
pub struct RawStatement(*mut ffi::sqlite3_stmt, bool, crate::util::ParamIndexCache);
impl RawStatement {
pub unsafe fn new(stmt: *mut ffi::sqlite3_stmt, tail: bool) -> RawStatement {
RawStatement(stmt, tail)
RawStatement(stmt, tail, Default::default())
}
pub fn is_null(&self) -> bool {
@ -87,12 +87,14 @@ impl RawStatement {
unsafe { ffi::sqlite3_bind_parameter_count(self.0) as usize }
}
pub fn bind_parameter_index(&self, name: &CStr) -> Option<usize> {
let r = unsafe { ffi::sqlite3_bind_parameter_index(self.0, name.as_ptr()) };
match r {
0 => None,
i => Some(i as usize),
}
pub fn bind_parameter_index(&self, name: &str) -> Option<usize> {
self.2.get_or_insert_with(name, |param_cstr| {
let r = unsafe { ffi::sqlite3_bind_parameter_index(self.0, param_cstr.as_ptr()) };
match r {
0 => None,
i => Some(i as usize),
}
})
}
pub fn clear_bindings(&self) -> c_int {

View File

@ -6,7 +6,7 @@ use std::slice::from_raw_parts;
use std::{convert, fmt, mem, ptr, str};
use super::ffi;
use super::{len_as_c_int, str_for_sqlite, str_to_cstring};
use super::{len_as_c_int, str_for_sqlite};
use super::{
AndThenRows, Connection, Error, MappedRows, RawStatement, Result, Row, Rows, ValueRef,
};
@ -432,8 +432,7 @@ impl Statement<'_> {
/// Will return Err if `name` is invalid. Will return Ok(None) if the name
/// is valid but not a bound parameter of this statement.
pub fn parameter_index(&self, name: &str) -> Result<Option<usize>> {
let c_name = str_to_cstring(name)?;
Ok(self.stmt.bind_parameter_index(&c_name))
Ok(self.stmt.bind_parameter_index(name))
}
fn bind_parameters<P>(&mut self, params: P) -> Result<()>

5
src/util/mod.rs Normal file
View File

@ -0,0 +1,5 @@
// Internal utilities
pub(crate) mod param_cache;
mod small_cstr;
pub(crate) use param_cache::ParamIndexCache;
pub(crate) use small_cstr::SmallCString;

60
src/util/param_cache.rs Normal file
View File

@ -0,0 +1,60 @@
use super::SmallCString;
use std::cell::RefCell;
use std::collections::BTreeMap;
/// Maps parameter names to parameter indices.
#[derive(Default, Clone, Debug)]
// BTreeMap seems to do better here unless we want to pull in a custom hash
// function.
pub(crate) struct ParamIndexCache(RefCell<BTreeMap<SmallCString, usize>>);
impl ParamIndexCache {
pub fn get_or_insert_with<F>(&self, s: &str, func: F) -> Option<usize>
where
F: FnOnce(&std::ffi::CStr) -> Option<usize>,
{
let mut cache = self.0.borrow_mut();
// Avoid entry API, needs allocation to test membership.
if let Some(v) = cache.get(s) {
return Some(*v);
}
// If there's an internal nul in the name it couldn't have been a
// parameter, so early return here is ok.
let name = SmallCString::new(s).ok()?;
let val = func(&name)?;
cache.insert(name, val);
Some(val)
}
}
#[cfg(test)]
mod test {
use super::*;
#[test]
fn test_cache() {
let p = ParamIndexCache::default();
let v = p.get_or_insert_with("foo", |cstr| {
assert_eq!(cstr.to_str().unwrap(), "foo");
Some(3)
});
assert_eq!(v, Some(3));
let v = p.get_or_insert_with("foo", |_| {
panic!("shouldn't be called this time");
});
assert_eq!(v, Some(3));
let v = p.get_or_insert_with("gar\0bage", |_| {
panic!("shouldn't be called here either");
});
assert_eq!(v, None);
let v = p.get_or_insert_with("bar", |cstr| {
assert_eq!(cstr.to_str().unwrap(), "bar");
None
});
assert_eq!(v, None);
let v = p.get_or_insert_with("bar", |cstr| {
assert_eq!(cstr.to_str().unwrap(), "bar");
Some(30)
});
assert_eq!(v, Some(30));
}
}

132
src/util/small_cstr.rs Normal file
View File

@ -0,0 +1,132 @@
use smallvec::{smallvec, SmallVec};
use std::ffi::{CStr, CString, NulError};
#[derive(Clone, PartialEq, Eq, PartialOrd, Ord)]
pub struct SmallCString(smallvec::SmallVec<[u8; 16]>);
impl SmallCString {
#[inline]
pub fn new(s: &str) -> Result<Self, NulError> {
if s.as_bytes().contains(&0u8) {
return Err(Self::fabricate_nul_error(s));
}
let mut buf = SmallVec::with_capacity(s.len() + 1);
buf.extend_from_slice(s.as_bytes());
buf.push(0);
Ok(Self(buf))
}
#[inline]
pub fn as_str(&self) -> &str {
debug_assert!(std::str::from_utf8(&self.as_bytes_without_nul()).is_ok());
// Constructor takes a &str so this is safe.
unsafe { std::str::from_utf8_unchecked(&self.as_bytes_without_nul()) }
}
#[inline]
pub fn as_bytes_without_nul(&self) -> &[u8] {
&self.0[..self.0.len() - 1]
}
#[inline]
pub fn len(&self) -> usize {
debug_assert_ne!(self.0.len(), 0);
self.0.len() - 1
}
#[inline]
pub fn is_empty(&self) -> bool {
self.len() == 0
}
#[inline]
pub fn as_cstr(&self) -> &CStr {
debug_assert!(CStr::from_bytes_with_nul(&self.0).is_ok());
unsafe { CStr::from_bytes_with_nul_unchecked(&self.0) }
}
#[cold]
fn fabricate_nul_error(b: &str) -> NulError {
CString::new(b).unwrap_err()
}
}
impl Default for SmallCString {
#[inline]
fn default() -> Self {
Self(smallvec![0])
}
}
impl std::fmt::Debug for SmallCString {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
f.debug_tuple("SmallCString").field(&self.as_str()).finish()
}
}
impl std::ops::Deref for SmallCString {
type Target = CStr;
#[inline]
fn deref(&self) -> &CStr {
self.as_cstr()
}
}
impl PartialEq<SmallCString> for str {
#[inline]
fn eq(&self, s: &SmallCString) -> bool {
s.as_bytes_without_nul() == self.as_bytes()
}
}
impl PartialEq<str> for SmallCString {
#[inline]
fn eq(&self, s: &str) -> bool {
self.as_bytes_without_nul() == s.as_bytes()
}
}
impl std::borrow::Borrow<str> for SmallCString {
#[inline]
fn borrow(&self) -> &str {
self.as_str()
}
}
#[cfg(test)]
mod test {
use super::*;
#[test]
fn test_small_cstring() {
// We don't go through the normal machinery for default, so make sure
// things work.
assert_eq!(SmallCString::default().0, SmallCString::new("").unwrap().0);
assert_eq!(SmallCString::new("foo").unwrap().len(), 3);
assert_eq!(SmallCString::new("foo").unwrap().0.as_slice(), b"foo\0");
assert_eq!(
SmallCString::new("foo").unwrap().as_bytes_without_nul(),
b"foo",
);
assert_eq!(SmallCString::new("😀").unwrap().len(), 4);
assert_eq!(
SmallCString::new("😀").unwrap().0.as_slice(),
b"\xf0\x9f\x98\x80\0",
);
assert_eq!(
SmallCString::new("😀").unwrap().as_bytes_without_nul(),
b"\xf0\x9f\x98\x80",
);
assert_eq!(SmallCString::new("").unwrap().len(), 0);
assert!(SmallCString::new("").unwrap().is_empty());
assert_eq!(SmallCString::new("").unwrap().0.as_slice(), b"\0");
assert_eq!(SmallCString::new("").unwrap().as_bytes_without_nul(), b"");
assert!(SmallCString::new("\0").is_err());
assert!(SmallCString::new("\0abc").is_err());
assert!(SmallCString::new("abc\0").is_err());
}
}