Select expressions must have at least one table

[glittershark]

trafficstars

Without this, select * from; parses successfully, even though it shouldn't.

[trevyn]

SELECT 1; is valid SQL, should this only apply after a FROM token is seen? (or whatever the SQL grammar we’re using says)

[glittershark]

Ah, yes, definitely

[glittershark]

the ANSI sql standard grammar has <table-expression>, containing the from, where, group by, having, and window clauses, as its own node, and then makes that entire node optional in the select_statement. Something like this:

diff --git a/src/select.rs b/src/select.rs
index 0740838..0f4f9b0 100644
--- a/src/select.rs
+++ b/src/select.rs
@@ -77,10 +77,8 @@ impl fmt::Display for LimitClause {
 }
 
 #[derive(Clone, Debug, Default, Eq, Hash, PartialEq, Serialize, Deserialize)]
-pub struct SelectStatement {
+pub struct TableExpression {
     pub tables: Vec<Table>,
-    pub distinct: bool,
-    pub fields: Vec<FieldDefinitionExpression>,
     pub join: Vec<JoinClause>,
     pub where_clause: Option<ConditionExpression>,
     pub group_by: Option<GroupByClause>,
@@ -88,6 +86,13 @@ pub struct SelectStatement {
     pub limit: Option<LimitClause>,
 }
 
+#[derive(Clone, Debug, Default, Eq, Hash, PartialEq, Serialize, Deserialize)]
+pub struct SelectStatement {
+    pub distinct: bool,
+    pub fields: Vec<FieldDefinitionExpression>,
+    pub tables: Option<TableExpression>,
+}
+
 impl fmt::Display for SelectStatement {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
         write!(f, "SELECT ")?;

feels like the right way to go here. I got about halfway through everything necessary to make that compile but figured I'd ask here before going any further since it's a pretty big breaking change. The other option would be to keep the AST as-is and only change the parser

[glittershark]

cadac57 is the more minimal version of the change which leaves the AST untouched

[ms705]

Thank you! :+1: This looks good. I actually agree with @glittershark's more extensive proposed change above. We should make that change, so that SELECT queries without tables are more natural. Either here or in a new PR is fine by me -- @glittershark?