(datafusion-sqlparser-rs) 01/01: Parser: fix exponential parse time on compound chains (#2344)

[gi...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

github-merge-queue[bot] pushed a commit to branch gh-readonly-queue/main/pr-2344-e999d3ddcf61bebe7b7aca825a3551b27cee27ac
in repository https://gitbox.apache.org/repos/asf/datafusion-sqlparser-rs.git

commit 2705a7ddebf92360a0bfa7eed080927064fec71e
Author: Luca Cappelletti <ca...@gmail.com>
AuthorDate: Thu May 21 13:09:56 2026 +0200

    Parser: fix exponential parse time on compound chains (#2344)
---
 sqlparser_bench/benches/sqlparser_bench.rs | 33 +++++++++++++++++++++++++++++-
 src/parser/mod.rs                          | 14 +++++++------
 tests/sqlparser_common.rs                  | 27 ++++++++++++++++++++++++
 3 files changed, 67 insertions(+), 7 deletions(-)

diff --git a/sqlparser_bench/benches/sqlparser_bench.rs b/sqlparser_bench/benches/sqlparser_bench.rs
index b52683aa..46c20154 100644
--- a/sqlparser_bench/benches/sqlparser_bench.rs
+++ b/sqlparser_bench/benches/sqlparser_bench.rs
@@ -152,5 +152,36 @@ fn parse_many_identifiers(c: &mut Criterion) {
     group.finish();
 }
 
-criterion_group!(benches, basic_queries, word_to_ident, parse_many_identifiers);
+/// Benchmark parsing pathological compound chains that previously caused 2^N
+/// work in `parse_compound_expr`. The input `IF a0.a1...aN.#` rejects at the
+/// trailing `#`, which used to force quadratic-or-worse backtracking through
+/// the chain.
+fn parse_compound_chain(c: &mut Criterion) {
+    let mut group = c.benchmark_group("parse_compound_chain");
+    let dialect = GenericDialect {};
+
+    for &n in &[10usize, 20, 30] {
+        let chain = (0..n)
+            .map(|i| format!("a{i}"))
+            .collect::<Vec<_>>()
+            .join(".");
+        let sql = format!("IF {chain}.#");
+
+        group.bench_function(format!("chain_{n}"), |b| {
+            b.iter(|| {
+                let _ = Parser::parse_sql(&dialect, std::hint::black_box(&sql));
+            });
+        });
+    }
+
+    group.finish();
+}
+
+criterion_group!(
+    benches,
+    basic_queries,
+    word_to_ident,
+    parse_many_identifiers,
+    parse_compound_chain
+);
 criterion_main!(benches);
diff --git a/src/parser/mod.rs b/src/parser/mod.rs
index 763b876d..91ac386a 100644
--- a/src/parser/mod.rs
+++ b/src/parser/mod.rs
@@ -2028,14 +2028,16 @@ impl<'a> Parser<'a> {
                         chain.push(AccessExpr::Dot(expr));
                         self.advance_token(); // The consumed placeholder
                     }
-                    // Fallback to parsing an arbitrary expression, but restrict to expression
-                    // types that are valid after the dot operator. This ensures that e.g.
-                    // `T.interval` is parsed as a compound identifier, not as an interval
-                    // expression.
+                    // Parse a single field component, restricted to expression types valid
+                    // after `.` (so e.g. `T.interval` is a compound identifier, not an
+                    // interval expression). Using `parse_prefix` here rather than
+                    // `parse_subexpr` avoids 2^N work on inputs like `IF a.b.c...x.#`:
+                    // the outer loop already consumes successive `.field` segments, so a
+                    // recursive `parse_subexpr` would re-walk the rest of the chain at
+                    // every dot.
                     _ => {
                         let expr = self.maybe_parse(|parser| {
-                            let expr = parser
-                                .parse_subexpr(parser.dialect.prec_value(Precedence::Period))?;
+                            let expr = parser.parse_prefix()?;
                             match &expr {
                                 Expr::CompoundFieldAccess { .. }
                                 | Expr::CompoundIdentifier(_)
diff --git a/tests/sqlparser_common.rs b/tests/sqlparser_common.rs
index bcf3c1d5..f470b93c 100644
--- a/tests/sqlparser_common.rs
+++ b/tests/sqlparser_common.rs
@@ -18977,3 +18977,30 @@ fn parse_non_pg_dialects_keep_xml_names_as_regular_identifiers() {
     let dialects = all_dialects_except(|d| d.supports_xml_expressions());
     dialects.verified_only_select("SELECT xml FROM t");
 }
+
+/// Regression test for the 2^N parse-time blowup in `parse_compound_expr` on
+/// inputs like `IF a0.a1...aN.#`. The parse is run on a worker thread and the
+/// main thread asserts that it reports back within a generous timeout. Post-fix
+/// the parser returns `Err` in well under a millisecond, so the timeout is a
+/// hang guard, not a perf threshold.
+#[test]
+fn parse_compound_chain_no_exponential_blowup() {
+    use std::sync::mpsc;
+    use std::thread;
+    use std::time::Duration;
+
+    let chain: String = (0..30)
+        .map(|i| format!("a{i}"))
+        .collect::<Vec<_>>()
+        .join(".");
+    let sql = format!("IF {chain}.#");
+
+    let (tx, rx) = mpsc::channel();
+    thread::spawn(move || {
+        let _ = Parser::parse_sql(&GenericDialect {}, &sql);
+        let _ = tx.send(());
+    });
+
+    rx.recv_timeout(Duration::from_secs(5))
+        .expect("parser should reject this quickly, not loop exponentially");
+}


---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@datafusion.apache.org
For additional commands, e-mail: commits-help@datafusion.apache.org