sqlrite/sql/parser/

create.rs

use sqlparser::ast::{
    ColumnDef, ColumnOption, CreateTable, DataType, Expr, ObjectName, ObjectNamePart, Statement,
    UnaryOperator, Value as AstValue,
};

use crate::error::{Result, SQLRiteError};
use crate::sql::db::table::Value;

/// True when an `ObjectName` resolves to a single identifier `VECTOR`
/// (case-insensitive). Phase 7a adds the `VECTOR(N)` column type as a
/// sqlparser `DataType::Custom` — the engine recognizes it via this
/// helper so the regular DataType match arm above stays uncluttered.
fn is_vector_type(name: &ObjectName) -> bool {
    name.0.len() == 1
        && match &name.0[0] {
            ObjectNamePart::Identifier(ident) => ident.value.eq_ignore_ascii_case("VECTOR"),
            // Function-form ObjectNamePart shouldn't appear in a CREATE TABLE
            // column type position. If it ever does, treat it as not-a-vector
            // and the outer match falls through to the "Invalid" arm.
            _ => false,
        }
}

/// Parses the dimension out of the `Custom` args for `VECTOR(N)`.
/// `args` is the `Vec<String>` sqlparser hands back for parenthesized
/// type arguments — for `VECTOR(384)` that's `["384"]`. Validates that
/// exactly one positive-integer argument was supplied.
fn parse_vector_dim(args: &[String]) -> std::result::Result<usize, String> {
    match args {
        [] => Err("VECTOR requires a dimension, e.g. `VECTOR(384)`".to_string()),
        [single] => {
            let trimmed = single.trim();
            match trimmed.parse::<usize>() {
                Ok(d) if d > 0 => Ok(d),
                Ok(_) => Err(format!("VECTOR dimension must be ≥ 1 (got `{trimmed}`)")),
                Err(_) => Err(format!(
                    "VECTOR dimension must be a positive integer (got `{trimmed}`)"
                )),
            }
        }
        many => Err(format!(
            "VECTOR takes exactly one dimension argument (got {})",
            many.len()
        )),
    }
}

/// The schema for each SQL column in every table is represented by
/// the following structure after parsed and tokenized
#[derive(PartialEq, Debug, Clone)]
pub struct ParsedColumn {
    /// Name of the column
    pub name: String,
    /// Datatype of the column in String format
    pub datatype: String,
    /// Value representing if column is PRIMARY KEY
    pub is_pk: bool,
    /// Value representing if column was declared with the NOT NULL Constraint
    pub not_null: bool,
    /// Value representing if column was declared with the UNIQUE Constraint
    pub is_unique: bool,
    /// Literal value to use when this column is omitted from an INSERT.
    /// Restricted to literal expressions (integer, real, text, bool, NULL);
    /// non-literal `DEFAULT` expressions are rejected at CREATE TABLE time.
    pub default: Option<Value>,
}

/// The following structure represents a CREATE TABLE query already parsed
/// and broken down into name and a Vector of `ParsedColumn` metadata
///
#[derive(Debug)]
pub struct CreateQuery {
    /// name of table after parking and tokenizing of query
    pub table_name: String,
    /// Vector of `ParsedColumn` type with column metadata information
    pub columns: Vec<ParsedColumn>,
}

/// Parses a single sqlparser `ColumnDef` into our internal `ParsedColumn`
/// representation. Extracted from `CreateQuery::new` so `ALTER TABLE ADD
/// COLUMN` can reuse the same column-shape parsing without re-implementing
/// the type / constraint / default plumbing.
///
/// Caller-side responsibilities not handled here:
/// - duplicate column name detection (a multi-column invariant)
/// - "more than one PRIMARY KEY" detection (a multi-column invariant)
pub fn parse_one_column(col: &ColumnDef) -> Result<ParsedColumn> {
    let name = col.name.to_string();

    // Parsing each column for it data type
    // For now only accepting basic data types
    let datatype: String = match &col.data_type {
        DataType::TinyInt(_)
        | DataType::SmallInt(_)
        | DataType::Int2(_)
        | DataType::Int(_)
        | DataType::Int4(_)
        | DataType::Int8(_)
        | DataType::Integer(_)
        | DataType::BigInt(_) => "Integer".to_string(),
        DataType::Boolean => "Bool".to_string(),
        DataType::Text => "Text".to_string(),
        DataType::Varchar(_bytes) => "Text".to_string(),
        DataType::Real => "Real".to_string(),
        DataType::Float(_precision) => "Real".to_string(),
        DataType::Double(_) => "Real".to_string(),
        DataType::Decimal(_) => "Real".to_string(),
        // Phase 7e — `JSON` parses as a unit variant in
        // sqlparser's DataType enum. JSONB is treated as
        // an alias (matches PostgreSQL's permissive
        // behaviour); both store as text under the hood.
        DataType::JSON | DataType::JSONB => "Json".to_string(),
        // Phase 7a — `VECTOR(N)` parses as Custom("VECTOR", ["N"]).
        // sqlparser's SQLite dialect doesn't have a built-in
        // Vector variant; Custom is what unrecognized type
        // names + their parenthesized args fall through to.
        DataType::Custom(name, args) if is_vector_type(name) => match parse_vector_dim(args) {
            Ok(dim) => format!("vector({dim})"),
            Err(e) => {
                return Err(SQLRiteError::General(format!(
                    "Invalid VECTOR column '{}': {e}",
                    col.name
                )));
            }
        },
        other => {
            eprintln!("not matched on custom type: {other:?}");
            "Invalid".to_string()
        }
    };

    let mut is_pk: bool = false;
    let mut is_unique: bool = false;
    let mut not_null: bool = false;
    let mut default: Option<Value> = None;
    for column_option in &col.options {
        match &column_option.option {
            ColumnOption::PrimaryKey(_) => {
                // For now, only Integer and Text types can be PRIMARY KEY and Unique
                // Therefore Indexed.
                if datatype != "Real" && datatype != "Bool" {
                    is_pk = true;
                    is_unique = true;
                    not_null = true;
                }
            }
            ColumnOption::Unique(_) => {
                // For now, only Integer and Text types can be UNIQUE
                // Therefore Indexed.
                if datatype != "Real" && datatype != "Bool" {
                    is_unique = true;
                }
            }
            ColumnOption::NotNull => {
                not_null = true;
            }
            ColumnOption::Default(expr) => {
                default = Some(eval_literal_default(expr, &datatype, &name)?);
            }
            _ => (),
        };
    }

    Ok(ParsedColumn {
        name,
        datatype,
        is_pk,
        not_null,
        is_unique,
        default,
    })
}

/// Evaluates a `DEFAULT <expr>` clause to a runtime `Value`. Restricted to
/// literal expressions — anything else (function calls, column references,
/// arithmetic on non-literals, `CURRENT_TIMESTAMP`, …) is rejected with a
/// typed error so users see the limit at `CREATE TABLE` time rather than
/// silently accepting a `DEFAULT` we can't honour at INSERT time.
///
/// Negative numeric literals come through sqlparser as `UnaryOp { Minus, Value(N) }`;
/// we unwrap one level of leading `+`/`-` to support `DEFAULT -1` / `DEFAULT +3.14`.
///
/// Type-checks the literal against the column's declared datatype and
/// rejects mismatches (e.g. `INTEGER ... DEFAULT 'foo'`).
fn eval_literal_default(expr: &Expr, datatype: &str, col_name: &str) -> Result<Value> {
    let value = match expr {
        Expr::Value(v) => &v.value,
        Expr::UnaryOp {
            op: UnaryOperator::Minus,
            expr: inner,
        } => {
            return match inner.as_ref() {
                Expr::Value(v) => match &v.value {
                    AstValue::Number(n, _) => {
                        let neg = format!("-{n}");
                        coerce_number_default(&neg, datatype, col_name)
                    }
                    _ => Err(SQLRiteError::General(format!(
                        "DEFAULT for column '{col_name}' must be a literal value"
                    ))),
                },
                _ => Err(SQLRiteError::General(format!(
                    "DEFAULT for column '{col_name}' must be a literal value"
                ))),
            };
        }
        Expr::UnaryOp {
            op: UnaryOperator::Plus,
            expr: inner,
        } => {
            return eval_literal_default(inner, datatype, col_name);
        }
        _ => {
            return Err(SQLRiteError::General(format!(
                "DEFAULT for column '{col_name}' must be a literal value"
            )));
        }
    };

    match value {
        AstValue::Null => Ok(Value::Null),
        AstValue::Boolean(b) => {
            if datatype == "Bool" {
                Ok(Value::Bool(*b))
            } else {
                Err(SQLRiteError::General(format!(
                    "DEFAULT type mismatch for column '{col_name}': boolean is not a {datatype}"
                )))
            }
        }
        AstValue::SingleQuotedString(s) => {
            if datatype == "Text" {
                Ok(Value::Text(s.clone()))
            } else if datatype == "Json" {
                // JSON columns accept text literals only if they parse as
                // JSON — otherwise an ALTER TABLE ADD COLUMN ... JSON
                // DEFAULT '<garbage>' would silently backfill every row
                // with invalid JSON (insert_row's per-row JSON validation
                // is bypassed during the backfill path).
                serde_json::from_str::<serde_json::Value>(s).map_err(|e| {
                    SQLRiteError::General(format!(
                        "DEFAULT type mismatch for column '{col_name}': '{s}' is not valid JSON: {e}"
                    ))
                })?;
                Ok(Value::Text(s.clone()))
            } else {
                Err(SQLRiteError::General(format!(
                    "DEFAULT type mismatch for column '{col_name}': text is not a {datatype}"
                )))
            }
        }
        AstValue::Number(n, _) => coerce_number_default(n, datatype, col_name),
        _ => Err(SQLRiteError::General(format!(
            "DEFAULT for column '{col_name}' must be a literal value"
        ))),
    }
}

fn coerce_number_default(n: &str, datatype: &str, col_name: &str) -> Result<Value> {
    match datatype {
        "Integer" => n.parse::<i64>().map(Value::Integer).map_err(|_| {
            SQLRiteError::General(format!(
                "DEFAULT type mismatch for column '{col_name}': '{n}' is not a valid INTEGER"
            ))
        }),
        "Real" => n.parse::<f64>().map(Value::Real).map_err(|_| {
            SQLRiteError::General(format!(
                "DEFAULT type mismatch for column '{col_name}': '{n}' is not a valid REAL"
            ))
        }),
        other => Err(SQLRiteError::General(format!(
            "DEFAULT type mismatch for column '{col_name}': numeric literal is not a {other}"
        ))),
    }
}

impl CreateQuery {
    pub fn new(statement: &Statement) -> Result<CreateQuery> {
        match statement {
            // Confirming the Statement is sqlparser::ast:Statement::CreateTable
            Statement::CreateTable(CreateTable {
                name,
                columns,
                constraints,
                ..
            }) => {
                let table_name = name;
                let mut parsed_columns: Vec<ParsedColumn> = vec![];

                // Iterating over the columns returned form the Parser::parse:sql
                // in the mod sql
                for col in columns {
                    // Checks if columm already added to parsed_columns, if so, returns an error
                    let name = col.name.to_string();
                    if parsed_columns.iter().any(|c| c.name == name) {
                        return Err(SQLRiteError::Internal(format!(
                            "Duplicate column name: {}",
                            &name
                        )));
                    }

                    let parsed = parse_one_column(col)?;

                    // Multi-column invariant: only one PRIMARY KEY per table.
                    if parsed.is_pk && parsed_columns.iter().any(|c| c.is_pk) {
                        return Err(SQLRiteError::Internal(format!(
                            "Table '{}' has more than one primary key",
                            &table_name
                        )));
                    }

                    parsed_columns.push(parsed);
                }
                // TODO: handle constraints + check constraints + ON DELETE /
                // ON UPDATE referential actions properly. They're currently
                // parsed by `sqlparser` and dropped on the floor here.
                // (Previously we `println!`-ed them to stdout as a debug
                // aid — removed in the engine-stdout-pollution cleanup;
                // flip to a `tracing` span if we ever want them visible in
                // dev builds.)
                let _ = constraints;
                Ok(CreateQuery {
                    table_name: table_name.to_string(),
                    columns: parsed_columns,
                })
            }

            _ => Err(SQLRiteError::Internal("Error parsing query".to_string())),
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::sql::*;

    #[test]
    fn create_table_validate_tablename_test() {
        let sql_input = String::from(
            "CREATE TABLE contacts (
            id INTEGER PRIMARY KEY,
            first_name TEXT NOT NULL,
            last_name TEXT NOT NULl,
            email TEXT NOT NULL UNIQUE
        );",
        );
        let expected_table_name = String::from("contacts");

        let dialect = SqlriteDialect::new();
        let mut ast = Parser::parse_sql(&dialect, &sql_input).unwrap();

        assert!(ast.len() == 1, "ast has more then one Statement");

        let query = ast.pop().unwrap();

        // Initialy only implementing some basic SQL Statements
        if let Statement::CreateTable(_) = query {
            let result = CreateQuery::new(&query);
            match result {
                Ok(payload) => {
                    assert_eq!(payload.table_name, expected_table_name);
                }
                Err(_) => panic!("an error occured during parsing CREATE TABLE Statement"),
            }
        }
    }
}