Cap generated tie-breaker values at i32::MAX

Keep generated ranges inside signed 32-bit pagination cursors until
downstream consumers can represent the full u64 domain.

CHANGELOG.md

@@ -6,7 +6,7 @@ Tantivy 0.27.0
 - Added the `FieldType::TieBreaker` variant; exhaustive matches on `FieldType` need a new arm.
 
 ## Features/Improvements
-- Add generated tie-breaker fast fields via `SchemaBuilder::add_tie_breaker_field`. Values are consecutive within a segment, start at a random offset, and are preserved across merges.
+- Add generated tie-breaker fast fields via `SchemaBuilder::add_tie_breaker_field`. Values are consecutive within a segment, start at a random offset within `[0, i32::MAX]`, and are preserved across merges.
 
 
 Tantivy 0.26.2

columnar/src/column/serialize.rs

@@ -30,8 +30,9 @@ pub(crate) fn serialize_generated_tie_breaker_column(
     num_docs: u32,
     output: &mut impl Write,
 ) -> io::Result<()> {
-    // TODO: Lift this temporary u32 limit once downstream consumers support the full u64 range.
-    let max_start = (u32::MAX - num_docs) as u64;
+    // Temporary i32 cap: some downstream pagination cursors still store the tie-breaker as
+    // int32. Lift this once those consumers can represent the full u64 range.
+    let max_start = (i32::MAX as u64).saturating_sub(u64::from(num_docs.saturating_sub(1)));
     let start: u64 = rand::rng().random_range(0..=max_start);
     let end = start + num_docs as u64;
     let values = start..end;

columnar/src/columnar/reader/mod.rs

@@ -247,6 +247,8 @@ mod tests {
         let handles = columnar.read_columns("tie").unwrap();
         let column = handles[0].open_u64_lenient().unwrap().unwrap();
         assert_eq!(column.index.get_cardinality(), crate::Cardinality::Full);
+        assert!(column.first(0).unwrap() <= i32::MAX as u64);
+        assert!(column.first(1_024).unwrap() <= i32::MAX as u64);
         for doc in 1..1_025 {
             assert_eq!(column.first(doc), Some(column.first(doc - 1).unwrap() + 1));
         }