Merge branch 'interpreter' into 'main'

Interpreter

See merge request x433485/minisql!2
This commit is contained in:
Yuriy Dupyn 2024-01-09 19:19:10 +01:00
commit 71d4bd76f0
13 changed files with 1277 additions and 256 deletions

9
Cargo.lock generated
View file

@ -55,6 +55,12 @@ dependencies = [
"rustc-demangle",
]
[[package]]
name = "bimap"
version = "0.6.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "230c5f1ca6a325a32553f8640d31ac9b49f2411e901e427570154868b46da4f7"
[[package]]
name = "bincode"
version = "2.0.0-rc.3"
@ -147,6 +153,9 @@ checksum = "f665ee40bc4a3c5590afb1e9677db74a508659dfd71e126420da8274909a0167"
[[package]]
name = "minisql"
version = "0.1.0"
dependencies = [
"bimap",
]
[[package]]
name = "miniz_oxide"

View file

@ -6,3 +6,4 @@ edition = "2021"
# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
[dependencies]
bimap = "0.6.3"

17
minisql/src/error.rs Normal file
View file

@ -0,0 +1,17 @@
use crate::internals::row::ColumnPosition;
use crate::internals::schema::{ColumnName, TableName};
use crate::operation::InsertionValues;
use crate::type_system::{DbType, Uuid, Value};
#[derive(Debug)]
pub enum Error {
TableDoesNotExist(TableName),
ColumnDoesNotExist(TableName, ColumnName),
ColumnPositionDoesNotExist(TableName, ColumnPosition),
ValueDoesNotMatchExpectedType(TableName, ColumnName, DbType, Value),
AttemptingToInsertAlreadyPresentId(TableName, Uuid),
MissingTypeAnnotationOfColumn(TableName, ColumnPosition),
MissingColumnInInsertValues(TableName, ColumnName, InsertionValues),
MismatchBetweenInsertValuesAndColumns(TableName, InsertionValues),
AttemptToIndexNonIndexableColumn(TableName, ColumnName),
}

View file

@ -0,0 +1,38 @@
use crate::type_system::{IndexableValue, Uuid};
use std::collections::{BTreeMap, HashSet};
#[derive(Debug)]
pub struct ColumnIndex {
index: BTreeMap<IndexableValue, HashSet<Uuid>>,
}
impl ColumnIndex {
pub fn new() -> Self {
let index = BTreeMap::new();
Self { index }
}
pub fn get(&self, value: &IndexableValue) -> Option<&HashSet<Uuid>> {
self.index.get(value)
}
pub fn add(&mut self, value: IndexableValue, id: Uuid) {
match self.index.get_mut(&value) {
Some(ids) => {
ids.insert(id);
}
None => {
self.index.insert(value, HashSet::from([id]));
}
}
}
pub fn remove(&mut self, value: &IndexableValue, id_to_be_removed: Uuid) -> bool {
match self.index.get_mut(value) {
Some(ids) => {
ids.remove(&id_to_be_removed) // true if was present
}
None => false,
}
}
}

View file

@ -0,0 +1,4 @@
pub mod column_index;
pub mod row;
pub mod schema;
pub mod table;

View file

@ -0,0 +1,71 @@
use crate::type_system::Value;
use std::ops::{Index, IndexMut};
use std::slice::SliceIndex;
pub type ColumnPosition = usize;
#[derive(Debug, Clone)]
pub struct Row(Vec<Value>);
impl<Idx> Index<Idx> for Row
where
Idx: SliceIndex<[Value]>,
{
type Output = Idx::Output;
fn index(&self, index: Idx) -> &Self::Output {
&self.0[index]
}
}
impl<Idx> IndexMut<Idx> for Row
where
Idx: SliceIndex<[Value]>,
{
fn index_mut(&mut self, index: Idx) -> &mut Self::Output {
&mut self.0[index]
}
}
impl FromIterator<Value> for Row {
fn from_iter<I: IntoIterator<Item = Value>>(iter: I) -> Self {
let mut v = vec![];
for x in iter {
v.push(x)
}
Row(v)
}
}
impl Row {
pub fn new() -> Self {
Row(vec![])
}
pub fn with_number_of_columns(number_of_columns: usize) -> Self {
Row(Vec::with_capacity(number_of_columns))
}
pub fn push(&mut self, value: Value) {
self.0.push(value)
}
pub fn len(&self) -> usize {
self.0.len()
}
pub fn get(&self, column_position: ColumnPosition) -> Option<&Value> {
self.0.get(column_position)
}
pub fn restrict_columns(&self, columns: &Vec<ColumnPosition>) -> Row {
// If the index from `columns` is non-existant in `row`, it will just ignore it.
let mut subrow: Row = Row::new();
for column_position in columns {
if let Some(value) = self.get(*column_position) {
subrow.0.push(value.clone())
}
}
subrow
}
}

View file

@ -0,0 +1,162 @@
use crate::error::Error;
use crate::internals::row::{ColumnPosition, Row};
use crate::operation::{ColumnSelection, InsertionValues};
use crate::result::DbResult;
use crate::type_system::{DbType, IndexableValue, Uuid, Value};
use bimap::BiMap;
use std::collections::HashMap;
// Note that it is nice to split metadata from the data because
// then you can give the metadata to the parser without giving it the data.
#[derive(Debug)]
pub struct TableSchema {
table_name: TableName, // used for descriptive errors
pub primary_key: ColumnPosition,
pub column_name_position_mapping: BiMap<ColumnName, ColumnPosition>,
pub types: Vec<DbType>,
}
pub type TableName = String;
pub type ColumnName = String;
impl TableSchema {
pub fn new(table_name: TableName, primary_key: ColumnPosition, column_name_position_map: Vec<(ColumnName, ColumnPosition)>, types: Vec<DbType>) -> Self {
let mut column_name_position_mapping: BiMap<ColumnName, ColumnPosition> = BiMap::new();
for (column_name, column_position) in column_name_position_map {
column_name_position_mapping.insert(column_name, column_position);
}
Self { table_name, primary_key, column_name_position_mapping, types }
}
pub fn table_name(&self) -> &TableName {
&self.table_name
}
fn get_column(&self, column_name: &ColumnName) -> DbResult<(DbType, ColumnPosition)> {
match self.column_name_position_mapping.get_by_left(column_name) {
Some(column_position) => match self.types.get(*column_position) {
Some(type_) => Ok((*type_, *column_position)),
None => Err(Error::MissingTypeAnnotationOfColumn(
self.table_name.clone(),
*column_position,
)),
},
None => Err(Error::ColumnDoesNotExist(
self.table_name.clone(),
column_name.clone(),
)),
}
}
pub fn column_position_from_column_name(
&self,
column_name: &ColumnName,
) -> DbResult<ColumnPosition> {
self.get_column(column_name)
.map(|(_, column_position)| column_position)
}
pub fn is_primary(&self, column_position: ColumnPosition) -> bool {
self.primary_key == column_position
}
fn column_positions_from_column_names(
&self,
column_names: &[ColumnName],
) -> DbResult<Vec<ColumnPosition>> {
let mut positions: Vec<ColumnPosition> = Vec::with_capacity(column_names.len());
for column_name in column_names {
let column_position = self.column_position_from_column_name(column_name)?;
positions.push(column_position)
}
Ok(positions)
}
pub fn column_name_from_column_position(
&self,
column_position: ColumnPosition,
) -> DbResult<ColumnName> {
match self
.column_name_position_mapping
.get_by_right(&column_position)
{
Some(column_name) => Ok(column_name.clone()),
None => Err(Error::ColumnPositionDoesNotExist(
self.table_name.clone(),
column_position,
)),
}
}
pub fn column_positions_from_column_selection(
&self,
column_selection: &ColumnSelection,
) -> DbResult<Vec<ColumnPosition>> {
match column_selection {
ColumnSelection::All => {
let mut column_positions: Vec<ColumnPosition> = self
.column_name_position_mapping
.iter()
.map(|(_, column_position)| *column_position)
.collect();
column_positions.sort();
Ok(column_positions)
}
ColumnSelection::Columns(column_names) => {
self.column_positions_from_column_names(column_names)
}
}
}
fn number_of_columns(&self) -> usize {
self.column_name_position_mapping.len()
}
pub fn row_from_insertion_values(
&self,
insertion_values: InsertionValues,
) -> DbResult<(Uuid, Row)> {
// TODO: There should be proper validation of the insertion_values.
// And it shouldn't really be done here.
//
// In the below we don't check for duplicate column names
//
let number_of_columns = self.number_of_columns();
if number_of_columns != insertion_values.len() {
return Err(Error::MismatchBetweenInsertValuesAndColumns(
self.table_name.clone(),
insertion_values,
));
}
let mut row: Row = Row::with_number_of_columns(number_of_columns);
let mut values: HashMap<ColumnName, Value> = HashMap::new();
for (column_name, db_value) in &insertion_values {
values.insert(column_name.clone(), db_value.clone());
}
for column_position in 0..number_of_columns {
let column_name: ColumnName = self.column_name_from_column_position(column_position)?;
match values.get(&column_name) {
Some(db_value) => row.push(db_value.clone()),
None => {
return Err(Error::MissingColumnInInsertValues(
self.table_name.clone(),
column_name,
insertion_values,
))
}
}
}
let id: Uuid = match row.get(self.primary_key) {
Some(Value::Indexable(IndexableValue::Uuid(id))) => *id,
Some(_) => unreachable!(),
None => unreachable!(),
};
Ok((id, row))
}
}

View file

@ -0,0 +1,262 @@
use std::collections::{BTreeMap, HashMap, HashSet};
use crate::error::Error;
use crate::internals::column_index::ColumnIndex;
use crate::internals::row::{ColumnPosition, Row};
use crate::internals::schema::{ColumnName, TableSchema, TableName};
use crate::result::DbResult;
use crate::type_system::{IndexableValue, Uuid, Value};
#[derive(Debug)]
pub struct Table {
schema: TableSchema,
rows: Rows, // TODO: Consider wrapping this in a lock. Also consider if we need to have the
// same lock for both rows and indexes
indexes: HashMap<ColumnPosition, ColumnIndex>,
}
pub type Rows = BTreeMap<Uuid, Row>;
impl Table {
pub fn new(table_schema: TableSchema) -> Self {
Self {
schema: table_schema,
rows: BTreeMap::new(),
indexes: HashMap::new(),
}
}
pub fn schema(&self) -> &TableSchema {
&self.schema
}
pub fn rows(&self) -> &Rows {
&self.rows
}
pub fn indexes(&self) -> &HashMap<ColumnPosition, ColumnIndex> {
&self.indexes
}
pub fn table_name(&self) -> &TableName {
&self.schema.table_name()
}
// ======Selection======
fn get_row_by_id(&self, id: Uuid) -> Option<Row> {
self.rows.get(&id).cloned()
}
fn get_rows_by_ids(&self, ids: HashSet<Uuid>) -> Vec<Row> {
ids.into_iter()
.filter_map(|id| self.get_row_by_id(id))
.collect()
}
fn get_rows_by_value(&self, column_position: ColumnPosition, value: &Value) -> Vec<Row> {
// brute-force search
self.rows
.values()
.filter_map(|row| {
if row.get(column_position) == Some(value) {
Some(row.clone())
} else {
None
}
})
.collect()
}
pub fn select_all_rows(&self, selected_column_positions: &Vec<ColumnPosition>) -> Vec<Row> {
self.rows
.values()
.map(|row| row.restrict_columns(selected_column_positions))
.collect()
}
pub fn select_rows_where_eq(
&self,
selected_column_positions: &Vec<ColumnPosition>,
column_position: ColumnPosition,
value: Value,
) -> DbResult<Vec<Row>> {
match value {
Value::Indexable(value) => match self.fetch_ids_from_index(column_position, &value)? {
Some(ids) => Ok(self
.get_rows_by_ids(ids)
.iter()
.map(|row| row.restrict_columns(selected_column_positions))
.collect()),
None => Ok(self
.get_rows_by_value(column_position, &Value::Indexable(value))
.iter()
.map(|row| row.restrict_columns(selected_column_positions))
.collect()),
},
_ => Ok(self
.get_rows_by_value(column_position, &value)
.iter()
.map(|row| row.restrict_columns(selected_column_positions))
.collect()),
}
}
// ======Insertion======
pub fn insert_row_at(&mut self, id: Uuid, row: Row) -> DbResult<()> {
if self.rows.get(&id).is_some() {
return Err(Error::AttemptingToInsertAlreadyPresentId(
self.table_name().clone(),
id,
));
}
for (column_position, column_index) in &mut self.indexes {
match row.get(*column_position) {
Some(Value::Indexable(val)) => column_index.add(val.clone(), id),
Some(_) => {}
None => {
return Err(Error::ColumnPositionDoesNotExist(
self.schema.table_name().clone(), // Note that I can't simply use self.table_name() here because of rust borrowing rules.
*column_position,
))
}
}
}
let _ = self.rows.insert(id, row);
Ok(())
}
// ======Deletion======
fn delete_row_by_id(&mut self, id: Uuid) -> usize {
match self.rows.remove(&id) {
Some(row) => {
for (column_position, column_index) in &mut self.indexes {
if let Value::Indexable(value) = &row[*column_position] {
let _ = column_index.remove(value, id);
};
}
1
}
None => 0,
}
}
fn delete_rows_by_ids(&mut self, ids: HashSet<Uuid>) -> usize {
let mut total_count = 0;
for id in ids {
total_count += self.delete_row_by_id(id)
}
total_count
}
fn delete_rows_by_value(&mut self, column_position: ColumnPosition, value: &Value) -> usize {
let matched_ids: HashSet<Uuid> = self
.rows
.iter()
.filter_map(|(id, row)| {
if row.get(column_position) == Some(value) {
Some(*id)
} else {
None
}
})
.collect();
self.delete_rows_by_ids(matched_ids)
}
pub fn delete_all_rows(&mut self) -> usize {
let number_of_rows = self.rows.len();
self.rows = BTreeMap::new();
self.indexes = HashMap::new();
number_of_rows
}
pub fn delete_rows_where_eq(
&mut self,
column_position: ColumnPosition,
value: Value,
) -> DbResult<usize> {
match value {
Value::Indexable(value) => match self.fetch_ids_from_index(column_position, &value)? {
Some(ids) => Ok(self.delete_rows_by_ids(ids)),
None => Ok(self.delete_rows_by_value(column_position, &Value::Indexable(value))),
},
_ => Ok(self.delete_rows_by_value(column_position, &value)),
}
}
// ======Indexing======
pub fn attach_index(&mut self, column_position: ColumnPosition) -> DbResult<()> {
let mut column_index: ColumnIndex = ColumnIndex::new();
update_index_from_table(&mut column_index, self, column_position)?;
self.indexes.insert(column_position, column_index);
Ok(())
}
fn fetch_ids_from_index(
&self,
column_position: ColumnPosition,
value: &IndexableValue,
) -> DbResult<Option<HashSet<Uuid>>> {
if self.schema.is_primary(column_position) {
match value {
IndexableValue::Uuid(id) => Ok(Some(HashSet::from([*id]))),
_ => {
let column_name: ColumnName = self
.schema
.column_name_from_column_position(column_position)?;
let type_ = self.schema.types[column_position];
Err(Error::ValueDoesNotMatchExpectedType(
self.table_name().clone(),
column_name,
type_,
Value::Indexable(value.clone()),
))
}
}
} else {
match self.indexes.get(&column_position) {
Some(index) => {
// Note that we are cloning the ids here! This can be very wasteful in some cases.
// It would be possible to just return a reference,
// but this seems fairly non-trivial.
let ids = index.get(value).cloned();
Ok(ids)
}
None => Ok(None),
}
}
}
}
// Should be used in the case when an index is created after the table has existed for a
// while. In such a case you need to build the index from the already existing rows.
fn update_index_from_table(
column_index: &mut ColumnIndex,
table: &Table,
column_position: ColumnPosition,
) -> DbResult<()> {
for (id, row) in &table.rows {
let value = match row.get(column_position) {
Some(Value::Indexable(value)) => value.clone(),
Some(_) => {
let column_name: ColumnName = table
.schema
.column_name_from_column_position(column_position)?;
return Err(Error::AttemptToIndexNonIndexableColumn(
table.table_name().to_string(),
column_name,
));
}
None => {
return Err(Error::ColumnPositionDoesNotExist(
table.table_name().to_string(),
column_position,
))
}
};
column_index.add(value, *id)
}
Ok(())
}

621
minisql/src/interpreter.rs Normal file
View file

@ -0,0 +1,621 @@
use crate::error::Error;
use crate::internals::row::{ColumnPosition, Row};
use crate::internals::schema::{TableName, TableSchema};
use crate::internals::table::Table;
use crate::operation::{ColumnSelection, Condition, Operation};
use crate::result::DbResult;
use crate::type_system::{DbType, IndexableValue, Value};
use bimap::BiMap;
// Use `TablePosition` as index
pub type Tables = Vec<Table>;
pub type TablePosition = usize;
// ==============Interpreter================
#[derive(Debug)]
pub struct State {
table_name_position_mapping: BiMap<TableName, TablePosition>,
tables: Tables,
}
#[derive(Debug)]
pub enum Response {
Selected(Vec<Row>),
Inserted,
Deleted(usize), // how many were deleted
TableCreated,
IndexCreated,
}
impl State {
fn new() -> Self {
Self {
table_name_position_mapping: BiMap::new(),
tables: vec![],
}
}
fn table_from_name<'b: 'a, 'a>(&'b self, table_name: &TableName) -> DbResult<&'a Table> {
match self.table_name_position_mapping.get_by_left(table_name) {
Some(table_position) => {
let table = &self.tables[*table_position];
Ok(table)
}
None => Err(Error::TableDoesNotExist(table_name.clone())),
}
}
fn table_from_name_mut<'b: 'a, 'a>(
&'b mut self,
table_name: &TableName,
) -> DbResult<&'a mut Table> {
match self.table_name_position_mapping.get_by_left(table_name) {
Some(table_position) => {
let table = &mut self.tables[*table_position];
Ok(table)
}
None => Err(Error::TableDoesNotExist(table_name.clone())),
}
}
fn attach_table(&mut self, table_name: TableName, table: Table) {
let new_table_position: TablePosition = self.tables.len();
self.table_name_position_mapping
.insert(table_name, new_table_position);
self.tables.push(table);
}
// TODO: Decide if we want for this to return a response (but then you have to deal with lifetimes,
// because you'll be forced to put an iterator/slice into the Response data-structure.
// Alternative is to pass a row-consumer to the functionas that knows how to communicate with
// the client, but the details of communication are hidden behind an interface
//
// writer: impl SqlResponseConsumer
pub fn interpret(&mut self, operation: Operation) -> DbResult<Response> {
// TODO: lock stuff
use Operation::*;
match operation {
Select(table_name, column_selection, maybe_condition) => {
let table: &Table = self.table_from_name(&table_name)?;
let selected_column_positions: Vec<ColumnPosition> = table
.schema()
.column_positions_from_column_selection(&column_selection)?;
let selected_rows = match maybe_condition {
None => table.select_all_rows(&selected_column_positions),
Some(Condition::Eq(eq_column_name, value)) => {
let eq_column_position = table
.schema()
.column_position_from_column_name(&eq_column_name)?;
table.select_rows_where_eq(
&selected_column_positions,
eq_column_position,
value,
)?
}
};
Ok(Response::Selected(selected_rows))
}
Insert(table_name, values) => {
let table: &mut Table = self.table_from_name_mut(&table_name)?;
let (id, row) = table.schema().row_from_insertion_values(values)?;
table.insert_row_at(id, row)?;
Ok(Response::Inserted)
}
Delete(table_name, maybe_condition) => {
let table: &mut Table = self.table_from_name_mut(&table_name)?;
let rows_affected = match maybe_condition {
None => table.delete_all_rows(),
Some(Condition::Eq(eq_column_name, value)) => {
let eq_column_position = table
.schema()
.column_position_from_column_name(&eq_column_name)?;
table.delete_rows_where_eq(eq_column_position, value)?
}
};
Ok(Response::Deleted(rows_affected))
}
CreateTable(table_name, table_schema) => {
let table = Table::new(table_schema);
self.attach_table(table_name, table);
Ok(Response::TableCreated)
}
CreateIndex(table_name, column_name) => {
let table: &mut Table = self.table_from_name_mut(&table_name)?;
let column_position: ColumnPosition = table
.schema()
.column_position_from_column_name(&column_name)?;
table.attach_index(column_position)?;
Ok(Response::IndexCreated)
}
}
}
}
// TODO: Give a better name to something that you can respond to with rows
trait SqlResponseConsumer {
// TODO:
}
#[cfg(test)]
mod tests {
use super::*;
use std::collections::HashSet;
fn users_schema() -> TableSchema {
let id: ColumnPosition = 0;
let name: ColumnPosition = 1;
let age: ColumnPosition = 2;
TableSchema::new(
"users".to_string(),
id,
vec!(
("id".to_string(), id),
("name".to_string(), name),
("age".to_string(), age),
),
vec![DbType::Uuid, DbType::String, DbType::Int],
)
}
#[test]
fn test_table_creation() {
let mut state = State::new();
let users_schema = users_schema();
let users = users_schema.table_name().clone();
state
.interpret(Operation::CreateTable(users.clone(), users_schema))
.unwrap();
assert!(state.tables.len() == 1);
let table = &state.tables[0];
assert!(table.rows().len() == 0);
assert!(table.table_name() == &users);
}
#[test]
fn test_select_empty() {
let mut state = State::new();
let users_schema = users_schema();
let users = users_schema.table_name().clone();
state
.interpret(Operation::CreateTable(users.clone(), users_schema))
.unwrap();
let response: Response = state
.interpret(Operation::Select(users.clone(), ColumnSelection::All, None))
.unwrap();
assert!(matches!(response, Response::Selected(_)));
let Response::Selected(rows) = response else {
todo!()
};
assert!(rows.len() == 0);
}
#[test]
fn test_select_nonexistant_table() {
let mut state = State::new();
let response: DbResult<Response> = state.interpret(Operation::Select(
"table_that_doesnt_exist".to_string(),
ColumnSelection::All,
None,
));
assert!(matches!(response, Err(Error::TableDoesNotExist(_))));
}
#[test]
fn test_insert_select_basic1() {
use IndexableValue::*;
use Value::*;
let mut state = State::new();
let users_schema = users_schema();
let users = users_schema.table_name().clone();
state
.interpret(Operation::CreateTable(users.clone(), users_schema))
.unwrap();
let (id, name, age) = (
Indexable(Uuid(0)),
Indexable(String("Plato".to_string())),
Indexable(Int(64)),
);
state
.interpret(Operation::Insert(
users.clone(),
vec![
("id".to_string(), id.clone()),
("name".to_string(), name.clone()),
("age".to_string(), age.clone()),
],
))
.unwrap();
let response: Response = state
.interpret(Operation::Select(users.clone(), ColumnSelection::All, None))
.unwrap();
assert!(matches!(response, Response::Selected(_)));
let Response::Selected(rows) = response else {
todo!()
};
assert!(rows.len() == 1);
let row = &rows[0];
assert!(row.len() == 3);
assert!(row[0] == id);
assert!(row[1] == name);
assert!(row[2] == age);
}
#[test]
fn test_insert_select_basic2() {
use ColumnSelection::*;
use Condition::*;
use IndexableValue::*;
use Operation::*;
use Value::*;
let mut state = State::new();
let users_schema = users_schema();
let users = users_schema.table_name().clone();
state
.interpret(CreateTable(users.clone(), users_schema))
.unwrap();
let (id0, name0, age0) = (
Indexable(Uuid(0)),
Indexable(String("Plato".to_string())),
Indexable(Int(64)),
);
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id0.clone()),
("name".to_string(), name0.clone()),
("age".to_string(), age0.clone()),
],
))
.unwrap();
let (id1, name1, age1) = (
Indexable(Uuid(1)),
Indexable(String("Aristotle".to_string())),
Indexable(Int(20)),
);
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id1.clone()),
("name".to_string(), name1.clone()),
("age".to_string(), age1.clone()),
],
))
.unwrap();
{
let response: Response = state.interpret(Select(users.clone(), All, None)).unwrap();
assert!(matches!(response, Response::Selected(_)));
let Response::Selected(rows) = response else {
todo!()
};
assert!(rows.len() == 2);
let row0 = &rows[0];
let row1 = &rows[1];
assert!(row0.len() == 3);
assert!(row0[0] == id0);
assert!(row0[1] == name0);
assert!(row0[2] == age0);
assert!(row1.len() == 3);
assert!(row1[0] == id1);
assert!(row1[1] == name1);
assert!(row1[2] == age1);
}
{
let response: Response = state
.interpret(Select(
users.clone(),
All,
Some(Eq("id".to_string(), id0.clone())),
))
.unwrap();
assert!(matches!(response, Response::Selected(_)));
let Response::Selected(rows) = response else {
todo!()
};
assert!(rows.len() == 1);
let row0 = &rows[0];
assert!(row0.len() == 3);
assert!(row0[0] == id0);
assert!(row0[1] == name0);
assert!(row0[2] == age0);
}
{
let response: Response = state
.interpret(Select(
users.clone(),
Columns(vec!["name".to_string(), "id".to_string()]),
Some(Eq("id".to_string(), id0.clone())),
))
.unwrap();
assert!(matches!(response, Response::Selected(_)));
let Response::Selected(rows) = response else {
todo!()
};
assert!(rows.len() == 1);
let row0 = &rows[0];
assert!(row0.len() == 2);
assert!(row0[0] == name0);
assert!(row0[1] == id0);
}
}
#[test]
fn test_delete() {
use ColumnSelection::*;
use Condition::*;
use IndexableValue::*;
use Operation::*;
use Value::*;
let mut state = State::new();
let users_schema = users_schema();
let users = users_schema.table_name().clone();
state
.interpret(CreateTable(users.clone(), users_schema))
.unwrap();
let (id0, name0, age0) = (
Indexable(Uuid(0)),
Indexable(String("Plato".to_string())),
Indexable(Int(64)),
);
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id0.clone()),
("name".to_string(), name0.clone()),
("age".to_string(), age0.clone()),
],
))
.unwrap();
let (id1, name1, age1) = (
Indexable(Uuid(1)),
Indexable(String("Aristotle".to_string())),
Indexable(Int(20)),
);
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id1.clone()),
("name".to_string(), name1.clone()),
("age".to_string(), age1.clone()),
],
))
.unwrap();
let delete_response: Response = state
.interpret(Delete(
users.clone(),
Some(Eq("id".to_string(), id0.clone())),
))
.unwrap();
assert!(matches!(delete_response, Response::Deleted(1)));
let response: Response = state.interpret(Select(users.clone(), All, None)).unwrap();
assert!(matches!(response, Response::Selected(_)));
let Response::Selected(rows) = response else {
todo!()
};
assert!(rows.len() == 1);
let row = &rows[0];
assert!(row.len() == 3);
assert!(row[0] == id1);
assert!(row[1] == name1);
assert!(row[2] == age1);
}
#[test]
fn test_index() {
use IndexableValue::*;
use Operation::*;
use Value::*;
let mut state = State::new();
let users_schema = users_schema();
let users = users_schema.table_name().clone();
state
.interpret(CreateTable(users.clone(), users_schema))
.unwrap();
state
.interpret(CreateIndex(users.clone(), "name".to_string()))
.unwrap();
let (id0, name0, age0) = (
Indexable(Uuid(0)),
Indexable(String("Plato".to_string())),
Indexable(Int(64)),
);
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id0.clone()),
("name".to_string(), name0.clone()),
("age".to_string(), age0.clone()),
],
))
.unwrap();
let (id1, name1, age1) = (
Indexable(Uuid(1)),
Indexable(String("Aristotle".to_string())),
Indexable(Int(20)),
);
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id1.clone()),
("name".to_string(), name1.clone()),
("age".to_string(), age1.clone()),
],
))
.unwrap();
assert!(state.tables.len() == 1);
let table = &state.tables[0];
assert!(table.rows().len() == 2);
let user: ColumnPosition = 1;
assert!(table.indexes().contains_key(&user));
let index = table.indexes().get(&user).unwrap();
let plato_id = 0;
let aristotle_id = 1;
let plato_ids = index.get(&String("Plato".to_string())).cloned().unwrap_or(HashSet::new());
assert!(plato_ids.contains(&plato_id));
assert!(!plato_ids.contains(&aristotle_id));
assert!(plato_ids.len() == 1);
}
}
pub fn example() {
use ColumnSelection::*;
use Condition::*;
use IndexableValue::*;
use Operation::*;
use Value::*;
let users_schema: TableSchema = {
let id: ColumnPosition = 0;
let name: ColumnPosition = 1;
let age: ColumnPosition = 2;
TableSchema::new(
"users".to_string(),
id,
vec!(
("id".to_string(), id),
("name".to_string(), name),
("age".to_string(), age),
),
vec![DbType::Uuid, DbType::String, DbType::Int],
)
};
let users = users_schema.table_name().clone();
let mut state = State::new();
state
.interpret(Operation::CreateTable(users.clone(), users_schema))
.unwrap();
let (id0, name0, age0) = (
Indexable(Uuid(0)),
Indexable(String("Plato".to_string())),
Indexable(Int(64)),
);
println!("==INSERT Plato==");
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id0.clone()),
("name".to_string(), name0.clone()),
("age".to_string(), age0.clone()),
],
))
.unwrap();
let (id1, name1, age1) = (
Indexable(Uuid(1)),
Indexable(String("Aristotle".to_string())),
Indexable(Int(20)),
);
println!("==INSERT Aristotle==");
state
.interpret(Insert(
users.clone(),
vec![
("id".to_string(), id1.clone()),
("name".to_string(), name1.clone()),
("age".to_string(), age1.clone()),
],
))
.unwrap();
println!();
{
let response: Response = state
.interpret(Operation::Select(users.clone(), ColumnSelection::All, None))
.unwrap();
println!("==SELECT ALL==");
println!("{:?}", response);
println!();
}
{
let response: Response = state
.interpret(Select(
users.clone(),
All,
Some(Eq("id".to_string(), id0.clone())),
))
.unwrap();
println!("==SELECT Plato==");
println!("{:?}", response);
println!();
}
{
let _delete_response: Response = state
.interpret(Delete(
users.clone(),
Some(Eq("id".to_string(), id0.clone())),
))
.unwrap();
println!("==DELETE Plato==");
let response: Response = state
.interpret(Select(
users.clone(),
Columns(vec!["name".to_string(), "id".to_string()]),
None,
))
.unwrap();
println!("==SELECT All==");
println!("{:?}", response);
println!();
}
}

View file

@ -1,259 +1,10 @@
use std::collections::{BTreeMap, HashMap};
// ==============SQL operations================
// TODO: Note that every operation has a table name.
// Perhaps consider factoring the table name out
// and think of the operations as operating on a unique table.
enum Operation {
Select(TableName, ColumnSelection, Option<Condition>),
Insert(TableName, InsertionValues),
Delete(TableName, Option<Condition>),
// Update(...),
CreateTable(TableName, TableSchema),
CreateIndex(TableName, ColumnName), // TODO: Is this sufficient?
// DropTable(TableName),
}
type InsertionValues = Vec<(ColumnName, DbValue)>;
enum ColumnSelection {
All,
Columns(Vec<ColumnName>),
}
enum Condition {
// And(Box<Condition>, Box<Condition>),
// Or(Box<Condition>, Box<Condition>),
// Not(Box<Condition>),
Eq(ColumnName, DbValue),
// LessOrEqual(ColumnName, DbValue),
// Less(ColumnName, DbValue),
// StringCondition(StringCondition),
}
// enum StringCondition {
// Prefix(ColumnName, String),
// Substring(ColumnName, String),
// }
// ==============Values and Types================
type UUID = u64;
// TODO: What about nulls? I would rather not have that as in SQL, it sucks.
// I would rather have non-nullable values by default,
// and something like an explicit Option type for nulls.
enum DbValue {
String(String),
Int(u64),
Number(f64),
UUID(UUID),
// TODO: what bout null?
}
// TODO: Can this be autogenerated from the values?
enum DbType {
String,
Int,
Number,
UUID,
}
impl DbValue {
// TODO: Can this be autogenerated?
fn to_type(self) -> DbType {
match self {
Self::String(_) => DbType::String,
Self::Int(_) => DbType::Int,
Self::Number(_) => DbType::Number,
Self::UUID(_) => DbType::UUID,
}
}
}
// ==============Tables================
// table-metadata and data
type TableName = String;
type TablePosition = u32;
struct Table {
schema: TableSchema,
rows: Rows, // TODO: Consider wrapping this in a lock. Also consider if we need to have the
// same lock for both rows and indexes
indexes:
HashMap<ColumnPosition, ColumnIndex> // TODO: Consider generalizing `ColumnPosition` to something that would also apply to a pair of `ColumnNames` etc
}
// TODO: Is this really indexed by DbValues?
// Maybe we should have a separate index type for each type of value we're indexing over
struct ColumnIndex {
index: BTreeMap<DbValue, UUID>
}
// Note that it is nice to split metadata from the data because
// then you can give the metadata to the parser without giving it the data.
struct TableSchema {
columns: HashMap<ColumnName, (DbType, ColumnPosition)>
}
// TODO
fn column_position(table_meta: TableSchema, column_name: ColumnName) -> Option<ColumnPosition> {
todo!()
}
// Use `TablePosition` as index
type Tables = Vec<Table>;
type ColumnName = String;
type ColumnPosition = u32;
// Use `ColumnPosition` as index
type Row = Vec<DbValue>;
type Rows =
// TODO: This should be some sort of an interface to a dictionary
// s.t. in the background it may modify stuff in memory or talk to the disk
BTreeMap<UUID, Row>;
// interface
// insert(id, value)
// ==============Interpreter================
struct State {
table_positions: HashMap<TableName, TablePosition>,
tables: Vec<Table>,
}
impl State {
fn table_from_name<'b: 'a, 'a>(&'b self, table_name: TableName) -> Option<&'a Table> {
todo!()
}
fn attach_table(&mut self, table: Table) {
todo!()
}
}
// TODO: Give a better name to something that you can respond to with rows
trait SqlConsumer {
// TODO:
}
// TODO: This should return a reference to the table
// 'tables_life contains 'table_life
fn get_table<'tables_life: 'table_life, 'table_life>(tables: &'tables_life Tables, table_name: &TableName) -> &'table_life Table {
// let table_position:
todo!()
}
// TODO: Decide if we want for this to return a response (but then you have to deal with lifetimes,
// because you'll be forced to put an iterator/slice into the Response data-structure.
// Alternative is to pass a row-consumer to the functionas that knows how to communicate with
// the client, but the details of communication are hidden behind an interface
fn interpret(table_name: TableName, operation: Operation, state: &mut State, consumer: impl SqlConsumer) -> () {
// TODO: lock stuff
use Operation::*;
match operation {
Select(table_name, column_selection, maybe_condition) => {
let table: &Table = todo!();
table.select_where(column_selection, maybe_condition, consumer)
},
Insert(table_name, values) => {
let table: &mut Table = todo!();
table.insert(values, consumer)
},
Delete(table_name, maybe_condition) => {
let table: &mut Table = todo!();
table.delete_where(maybe_condition, consumer)
},
CreateTable(table_name, table_schema) => {
let table = Table::new(table_name, table_schema);
state.attach_table(table);
todo!()
},
CreateIndex(table_name, column_name) => {
let table: &mut Table = todo!();
let index: ColumnIndex = ColumnIndex::new(table, column_name);
table.attach_index(index);
}, // TODO: Is this sufficient?
//
}
}
impl ColumnIndex {
fn new(table: &Table, column_name: ColumnName) -> ColumnIndex {
todo!()
}
}
impl Table {
fn new(table_name: TableName, table_schema: TableSchema) -> Table {
todo!()
}
fn attach_index(&mut self, column_index: ColumnIndex) {
todo!()
}
fn select_where(&self, column_selection: ColumnSelection, maybe_condition: Option<Condition>, consumer: impl SqlConsumer) {
match maybe_condition {
None => {
// .iter() will give us an iterator over all the rows
// two choices
// 1. optimized version
// self.iter_with_columns(column_selection).for_each(|row| {
// consumer.send(row)
// });
// 2.
// self.iter()
// .map(|row| row.select_columns(column_selection))
// .for_each(|reduced_row| {
// consumer.send(row)
// });
todo!()
},
Some(Condition::Eq(column_name, value)) => {
// is column_name primary key? then it is easy
// self.get(id)
// is column_name indexed? Then get the index, and then it is not easy, because you
// may get a set of ids.
// what if it is not primary nor indexed? then you need to brute force your way
// through the whole table?
todo!()
}
}
}
fn insert(&mut self, values: InsertionValues, consumer: impl SqlConsumer) {
// 1. You need to update indices
// 2. you simply insert the data
todo!()
}
fn delete_where(&mut self, maybe_condition: Option<Condition>, consumer: impl SqlConsumer) {
// kinda similar to select with respect to the conditions
// update index
todo!()
}
}
// enum Response {
// Selected(impl Iter<???>), // TODO: How to do this? Some reference to an iterator somehow... slice..?
// Inserted(???),
// Deleted(usize), // how many were deleted
// }
mod error;
mod internals;
mod interpreter;
mod operation;
mod result;
mod type_system;
fn main() {
println!("Hello, world!");
interpreter::example();
}

40
minisql/src/operation.rs Normal file
View file

@ -0,0 +1,40 @@
use crate::internals::schema::{ColumnName, TableName, TableSchema};
use crate::type_system::Value;
// ==============SQL operations================
// TODO: Note that every operation has a table name.
// Perhaps consider factoring the table name out
// and think of the operations as operating on a unique table.
// TODO: `TableName` should be replaced by `TablePosition`
pub enum Operation {
Select(TableName, ColumnSelection, Option<Condition>),
Insert(TableName, InsertionValues),
Delete(TableName, Option<Condition>),
// Update(...),
CreateTable(TableName, TableSchema),
CreateIndex(TableName, ColumnName),
// DropTable(TableName),
}
pub type InsertionValues = Vec<(ColumnName, Value)>;
pub enum ColumnSelection {
All,
Columns(Vec<ColumnName>),
}
pub enum Condition {
// And(Box<Condition>, Box<Condition>),
// Or(Box<Condition>, Box<Condition>),
// Not(Box<Condition>),
Eq(ColumnName, Value),
// LessOrEqual(ColumnName, DbValue),
// Less(ColumnName, DbValue),
// StringCondition(StringCondition),
}
// enum StringCondition {
// Prefix(ColumnName, String),
// Substring(ColumnName, String),
// }

3
minisql/src/result.rs Normal file
View file

@ -0,0 +1,3 @@
use crate::error::Error;
pub type DbResult<A> = Result<A, Error>;

View file

@ -0,0 +1,42 @@
// ==============Types================
#[derive(Debug, Clone, Copy)]
pub enum DbType {
String,
Int,
Number,
Uuid,
}
// ==============Values================
pub type Uuid = u64;
// TODO: What about nulls? I would rather not have that in SQL, it sucks.
// I would rather have non-nullable values by default,
// and something like an explicit Option type for nulls.
#[derive(Debug, Clone, PartialEq)]
pub enum Value {
Number(f64), // TODO: Can't put floats as keys in maps, since they don't implement Eq. What to
// do?
Indexable(IndexableValue),
}
#[derive(Debug, Ord, Eq, Clone, PartialOrd, PartialEq)]
pub enum IndexableValue {
String(String),
Int(u64),
Uuid(Uuid),
// TODO: what about null?
}
impl Value {
pub fn to_type(&self) -> DbType {
match self {
Self::Number(_) => DbType::Number,
Self::Indexable(val) => match val {
IndexableValue::String(_) => DbType::String,
IndexableValue::Int(_) => DbType::Int,
IndexableValue::Uuid(_) => DbType::Uuid,
},
}
}
}