feat: check enum literals at compile time and fix rule helper and slot checks (#1258)

New K013 compile error: a constant string literal that fills an enum-typed column of an enforced contract must be an enum member (direct, in UNION branches, or through unfiltered CTE and derived-table passthroughs; skipped on SQL Server). Rules can read enum types with ctx.columns.enum_types. The rule test harness loads project rule helpers. Slot checks cover LATERAL subqueries and relation-qualified reads beside groups.

Author: kvlonge

crates/sqlbuild-analysis/src/contracts/_helpers/enum_literals.rs

@@ -0,0 +1,377 @@
+//! Constant string literals that reach an output column, and where the authored file writes them.
+
+use polyglot_sql::expressions::{Cte, Literal, Select, With};
+use polyglot_sql::tokens::{Token, TokenType};
+use polyglot_sql::{Dialect, DialectType, Expression};
+
+use crate::contracts::models::LiteralLocation;
+
+/// How deep the walk follows set operations and passthrough columns before it stops.
+const MAX_DEPTH: usize = 64;
+
+/// The output column a walk follows: by name, or by position in a set-operation branch.
+#[derive(Clone, Copy)]
+enum Target<'n> {
+    Name(&'n str),
+    Position(usize),
+}
+
+/// The CTEs a query can read, in definition order, and how deep the walk has gone.
+#[derive(Clone)]
+struct Scope<'a> {
+    ctes: Vec<&'a Cte>,
+    depth: usize,
+}
+
+impl<'a> Scope<'a> {
+    /// The scope one step deeper, with `with`'s CTEs visible unless they are recursive.
+    fn within(&self, with: Option<&'a With>) -> Self {
+        let mut ctes: Vec<&'a Cte> = self.ctes.clone();
+        if let Some(with) = with
+            && !with.recursive
+        {
+            ctes.extend(with.ctes.iter());
+        }
+        Self {
+            ctes,
+            depth: self.depth + 1,
+        }
+    }
+
+    /// The scope a CTE body sees: the CTEs defined before it.
+    fn before(&self, index: usize) -> Self {
+        Self {
+            ctes: self.ctes[..index].to_vec(),
+            depth: self.depth + 1,
+        }
+    }
+}
+
+/// The literals found so far.
+#[derive(Default)]
+struct LiteralWalk {
+    found: Vec<String>,
+}
+
+/// Every constant string literal that fills output column `column` of `query`, in query order.
+pub(crate) fn output_literals(query: &Expression, column: &str) -> Vec<String> {
+    let mut walk: LiteralWalk = LiteralWalk::default();
+    let scope: Scope<'_> = Scope {
+        ctes: Vec::new(),
+        depth: 0,
+    };
+    walk.query(query, Target::Name(column), &scope);
+    walk.found
+}
+
+/// The parsed statement of `sql` under `dialect`, when it is one query.
+pub(crate) fn parsed_query(sql: &str, dialect: &str) -> Option<Expression> {
+    let Ok(dialect) = dialect.parse::<DialectType>() else {
+        return None;
+    };
+    let Ok(mut statements) = polyglot_sql::parse(sql, dialect) else {
+        return None;
+    };
+    (statements.len() == 1).then(|| statements.pop()).flatten()
+}
+
+/// The first string literal with text `value` in `query`, located in `authored`, the whole file.
+pub(crate) fn literal_location(
+    (authored, query): (&str, &str),
+    value: &str,
+    dialect: &str,
+) -> Option<LiteralLocation> {
+    let dialect: DialectType = dialect
+        .parse::<DialectType>()
+        .unwrap_or(DialectType::Generic);
+    let Ok(tokens) = Dialect::get(dialect).tokenize(query) else {
+        return None;
+    };
+    let matches = |index: &usize| {
+        tokens[*index].token_type == TokenType::String && tokens[*index].text == value
+    };
+    let index: usize = (0..tokens.len())
+        .filter(matches)
+        .find(|index| select_item(&tokens, *index))
+        .or_else(|| (0..tokens.len()).find(matches))?;
+    let token = &tokens[index];
+    let query_start: usize = authored.find(query)?;
+    let offset: usize = authored[..query_start].chars().count();
+    let (line, column) = line_column(authored, offset + token.span.start);
+    let (end_line, end_column) = line_column(authored, offset + token.span.end);
+    Some(LiteralLocation {
+        line,
+        column,
+        end_line,
+        end_column,
+    })
+}
+
+/// Whether the token at `index` stands alone as a select-list item, as a projected literal does.
+fn select_item(tokens: &[Token], index: usize) -> bool {
+    let opens: bool = index.checked_sub(1).is_some_and(|previous| {
+        matches!(
+            tokens[previous].token_type,
+            TokenType::Select | TokenType::Distinct | TokenType::All | TokenType::Comma
+        )
+    });
+    let closes: bool = tokens.get(index + 1).is_none_or(|next| {
+        matches!(
+            next.token_type,
+            TokenType::As
+                | TokenType::Comma
+                | TokenType::From
+                | TokenType::Union
+                | TokenType::Semicolon
+                | TokenType::RParen
+                | TokenType::Identifier
+                | TokenType::QuotedIdentifier
+                | TokenType::Var
+        )
+    });
+    opens && closes
+}
+
+/// The one-based line and column of character `offset` in `text`.
+fn line_column(text: &str, offset: usize) -> (usize, usize) {
+    let mut line: usize = 1;
+    let mut column: usize = 1;
+    for character in text.chars().take(offset) {
+        if character == '\n' {
+            line += 1;
+            column = 1;
+        } else {
+            column += 1;
+        }
+    }
+    (line, column)
+}
+
+/// The position of output column `name` in the leftmost branch of `query`, before any star.
+fn output_position(query: &Expression, name: &str, depth: usize) -> Option<usize> {
+    if depth > MAX_DEPTH {
+        return None;
+    }
+    match query {
+        Expression::Select(select) => {
+            for (position, projection) in select.expressions.iter().enumerate() {
+                match projection {
+                    Expression::Star(_) => return None,
+                    projection
+                        if output_name(projection).is_some_and(|known| same(known, name)) =>
+                    {
+                        return Some(position);
+                    }
+                    _ => {}
+                }
+            }
+            None
+        }
+        Expression::Union(union) => output_position(&union.left, name, depth + 1),
+        Expression::Intersect(intersect) => output_position(&intersect.left, name, depth + 1),
+        Expression::Subquery(subquery) if subquery.column_aliases.is_empty() => {
+            output_position(&subquery.this, name, depth + 1)
+        }
+        Expression::Paren(paren) => output_position(&paren.this, name, depth + 1),
+        _ => None,
+    }
+}
+
+/// The name a projection outputs, when it names one.
+fn output_name(projection: &Expression) -> Option<&str> {
+    match projection {
+        Expression::Alias(alias) => Some(alias.alias.name.as_str()),
+        Expression::Column(column) => Some(column.name.name.as_str()),
+        _ => None,
+    }
+}
+
+fn same(left: &str, right: &str) -> bool {
+    left.eq_ignore_ascii_case(right)
+}
+
+/// The one `FROM` item a select passes every row of through, when `qualifier` names it or is absent.
+fn passthrough_source<'a>(select: &'a Select, qualifier: Option<&str>) -> Option<&'a Expression> {
+    if !keeps_every_row(select) {
+        return None;
+    }
+    let [source] = select.from.as_ref()?.expressions.as_slice() else {
+        return None;
+    };
+    match qualifier {
+        Some(qualifier) if !source_alias(source).is_some_and(|alias| same(alias, qualifier)) => {
+            None
+        }
+        _ => Some(source),
+    }
+}
+
+/// Whether a select outputs a row for every row it reads: no filter, join, limit or sampling.
+fn keeps_every_row(select: &Select) -> bool {
+    keeps_any_row(select)
+        && select.joins.is_empty()
+        && select.lateral_views.is_empty()
+        && select.prewhere.is_none()
+        && select.where_clause.is_none()
+        && select.having.is_none()
+        && select.qualify.is_none()
+        && select.distinct_on.is_none()
+        && select.connect.is_none()
+}
+
+/// Whether a select has no row limit or sampling, which could leave a literal with no rows.
+fn keeps_any_row(select: &Select) -> bool {
+    select.limit.is_none()
+        && select.offset.is_none()
+        && select.limit_by.is_none()
+        && select.fetch.is_none()
+        && select.top.is_none()
+        && select.sample.is_none()
+}
+
+/// The name a `FROM` item is read by: its alias, or a table's own name.
+fn source_alias(source: &Expression) -> Option<&str> {
+    match source {
+        Expression::Table(table) => Some(
+            table
+                .alias
+                .as_ref()
+                .map_or(table.name.name.as_str(), |alias| alias.name.as_str()),
+        ),
+        Expression::Subquery(subquery) => subquery.alias.as_ref().map(|alias| alias.name.as_str()),
+        _ => None,
+    }
+}
+
+impl LiteralWalk {
+    fn query<'a>(&mut self, query: &'a Expression, target: Target<'_>, scope: &Scope<'a>) {
+        if scope.depth > MAX_DEPTH {
+            return;
+        }
+        match query {
+            Expression::Select(select) => {
+                self.select(select, target, &scope.within(select.with.as_ref()))
+            }
+            Expression::Union(union) if union.limit.is_none() && union.offset.is_none() => {
+                let scope: Scope<'a> = scope.within(union.with.as_ref());
+                let target: Option<Target<'_>> = match target {
+                    Target::Name(_) if union.by_name => Some(target),
+                    Target::Position(_) if union.by_name => None,
+                    Target::Name(name) => {
+                        output_position(&union.left, name, 0).map(Target::Position)
+                    }
+                    Target::Position(_) => Some(target),
+                };
+                if let Some(target) = target {
+                    self.query(&union.left, target, &scope);
+                    self.query(&union.right, target, &scope);
+                }
+            }
+            Expression::Subquery(subquery)
+                if subquery.column_aliases.is_empty()
+                    && subquery.limit.is_none()
+                    && subquery.offset.is_none() =>
+            {
+                self.query(&subquery.this, target, &scope.within(None));
+            }
+            Expression::Paren(paren) => self.query(&paren.this, target, &scope.within(None)),
+            _ => {}
+        }
+    }
+
+    fn select<'a>(&mut self, select: &'a Select, target: Target<'_>, scope: &Scope<'a>) {
+        if !keeps_any_row(select) {
+            return;
+        }
+        match target {
+            Target::Position(position) => {
+                let starred: bool = select
+                    .expressions
+                    .iter()
+                    .take(position + 1)
+                    .any(|projection| matches!(projection, Expression::Star(_)));
+                if let Some(projection) = select.expressions.get(position)
+                    && !starred
+                {
+                    self.expression(projection, select, scope);
+                }
+            }
+            Target::Name(name) => {
+                let named: Vec<&Expression> = select
+                    .expressions
+                    .iter()
+                    .filter(|projection| {
+                        output_name(projection).is_some_and(|known| same(known, name))
+                    })
+                    .collect();
+                if let [projection] = named.as_slice() {
+                    self.expression(projection, select, scope);
+                    return;
+                }
+                if !named.is_empty() {
+                    return;
+                }
+                let [Expression::Star(star)] = select.expressions.as_slice() else {
+                    return;
+                };
+                if star.except.is_some() || star.replace.is_some() || star.rename.is_some() {
+                    return;
+                }
+                let qualifier: Option<&str> = star.table.as_ref().map(|table| table.name.as_str());
+                if let Some(source) = passthrough_source(select, qualifier) {
+                    self.source(source, name, scope);
+                }
+            }
+        }
+    }
+
+    fn expression<'a>(
+        &mut self,
+        expression: &'a Expression,
+        select: &'a Select,
+        scope: &Scope<'a>,
+    ) {
+        match expression {
+            Expression::Alias(alias) => self.expression(&alias.this, select, scope),
+            Expression::Paren(paren) => self.expression(&paren.this, select, scope),
+            Expression::Literal(literal) => {
+                if let Literal::String(value) = literal.as_ref() {
+                    self.found.push(value.clone());
+                }
+            }
+            Expression::Column(column) => {
+                let qualifier: Option<&str> =
+                    column.table.as_ref().map(|table| table.name.as_str());
+                if let Some(source) = passthrough_source(select, qualifier) {
+                    self.source(source, &column.name.name, scope);
+                }
+            }
+            _ => {}
+        }
+    }
+
+    /// The literals filling column `name` of a CTE or derived table that a `FROM` item reads.
+    fn source<'a>(&mut self, source: &'a Expression, name: &str, scope: &Scope<'a>) {
+        match source {
+            Expression::Table(table) if table.schema.is_none() && table.catalog.is_none() => {
+                let Some(index) = scope
+                    .ctes
+                    .iter()
+                    .rposition(|cte| same(&cte.alias.name, &table.name.name))
+                else {
+                    return;
+                };
+                let cte: &Cte = scope.ctes[index];
+                if table.column_aliases.is_empty() && cte.columns.is_empty() {
+                    self.query(&cte.this, Target::Name(name), &scope.before(index));
+                }
+            }
+            Expression::Subquery(subquery)
+                if subquery.column_aliases.is_empty() && !subquery.lateral =>
+            {
+                self.query(&subquery.this, Target::Name(name), &scope.within(None));
+            }
+            _ => {}
+        }
+    }
+}

crates/sqlbuild-analysis/src/contracts/_helpers/mod.rs

@@ -1,5 +1,6 @@
 pub(crate) mod columns;
 pub(crate) mod dynamic;
+pub(crate) mod enum_literals;
 pub(crate) mod promotion;
 pub(crate) mod setting_help;
 pub(crate) mod type_comparison;

crates/sqlbuild-analysis/src/contracts/constants.rs

@@ -8,6 +8,9 @@ pub(crate) const EXTRA_COLUMN_CODE: &str = "K005";
 pub(crate) const MISSING_DECLARATIONS_CODE: &str = "K006";
 pub(crate) const DYNAMIC_OUTPUT_NOT_PROVEN_CODE: &str = "K011";
 pub(crate) const PROMOTION_CONFLICT_CODE: &str = "K011";
+pub(crate) const ENUM_LITERAL_CODE: &str = "K013";
+/// Dialects whose default string comparison ignores case or trailing spaces, so K013 is skipped.
+pub(crate) const CASE_INSENSITIVE_DIALECTS: [&str; 1] = ["tsql"];
 
 pub const CONTRACT_ENFORCED: &str = "enforced";
 pub(crate) const CONTRACT_NONE: &str = "none";