feat: add LIKE prefix page skip for string columns

Extracts the fixed prefix from LIKE patterns (before the first wildcard)
and skips pages where the string range doesn't overlap. Patterns
starting with % or _ fall through to row-level evaluation.

Author: teamchong

docs/src/content/docs/performance.mdx

@@ -109,7 +109,7 @@ This is automatic — no configuration needed. Supported filter ops:
 | `is_null` | Page has zero null values (nullCount = 0) |
 | `is_not_null` | Page is entirely null (nullCount = rowCount) |
 
-String columns also have min/max stats (lexicographic). `like` and `not_like` filters don't benefit from page skip.
+String columns also have min/max stats (lexicographic). `like` patterns with a fixed prefix (e.g., `'abc%'`) can skip pages where the string range doesn't overlap the prefix. Patterns starting with a wildcard (e.g., `'%xyz'`) cannot be pruned. `not_like` does not benefit from page skip.
 
 ### OR filters and page skip
 

src/decode.test.ts

@@ -199,6 +199,25 @@ describe("canSkipPage", () => {
     const withNulls: PageInfo = { byteOffset: 0n, byteLength: 100, rowCount: 50, nullCount: 5, minValue: 10, maxValue: 90 };
     expect(canSkipPage(withNulls, [{ column: "x", op: "is_not_null", value: 0 }], "x")).toBe(false);
   });
+
+  it("skips page with LIKE prefix when range doesn't overlap", () => {
+    const strPage: PageInfo = { byteOffset: 0n, byteLength: 100, rowCount: 50, nullCount: 0, minValue: "apple", maxValue: "banana" };
+    // Prefix "car" is entirely above [apple, banana]
+    expect(canSkipPage(strPage, [{ column: "x", op: "like", value: "car%" }], "x")).toBe(true);
+    // Prefix "aa" is entirely below [apple, banana]
+    expect(canSkipPage(strPage, [{ column: "x", op: "like", value: "aa%" }], "x")).toBe(true);
+  });
+
+  it("does not skip page with LIKE prefix when range overlaps", () => {
+    const strPage: PageInfo = { byteOffset: 0n, byteLength: 100, rowCount: 50, nullCount: 0, minValue: "apple", maxValue: "banana" };
+    expect(canSkipPage(strPage, [{ column: "x", op: "like", value: "ba%" }], "x")).toBe(false);
+    expect(canSkipPage(strPage, [{ column: "x", op: "like", value: "app%" }], "x")).toBe(false);
+  });
+
+  it("does not skip page with LIKE pattern starting with wildcard", () => {
+    const strPage: PageInfo = { byteOffset: 0n, byteLength: 100, rowCount: 50, nullCount: 0, minValue: "apple", maxValue: "banana" };
+    expect(canSkipPage(strPage, [{ column: "x", op: "like", value: "%xyz%" }], "x")).toBe(false);
+  });
 });
 
 describe("assembleRows", () => {

src/decode.ts

@@ -71,6 +71,16 @@ export function canSkipPage(page: PageInfo, filters: QueryDescriptor["filters"],
         if (min >= lo && max <= hi) return true;
         break;
       }
+      // like: skip when the fixed prefix doesn't overlap the page's string range
+      case "like": {
+        if (typeof filter.value !== "string" || typeof min !== "string") break;
+        const prefix = extractLikePrefix(filter.value);
+        if (!prefix) break;
+        // Page max < prefix or page min >= prefix + '\uffff' → no match possible
+        const prefixEnd = prefix.slice(0, -1) + String.fromCharCode(prefix.charCodeAt(prefix.length - 1) + 1);
+        if ((max as string) < prefix || (min as string) >= prefixEnd) return true;
+        break;
+      }
     }
   }
   return false;
@@ -547,6 +557,18 @@ export function rowPassesFilters(row: Row, filters: FilterOp[], filterGroups?: F
   return true;
 }
 
+/** Extract the fixed prefix from a LIKE pattern (before the first wildcard). Returns null if no useful prefix. */
+function extractLikePrefix(pattern: string): string | null {
+  let prefix = "";
+  for (let i = 0; i < pattern.length; i++) {
+    const ch = pattern[i];
+    if (ch === "%" || ch === "_") break;
+    if (ch === "\\" && i + 1 < pattern.length) { prefix += pattern[++i]; continue; }
+    prefix += ch;
+  }
+  return prefix.length > 0 ? prefix : null;
+}
+
 /** Cache compiled LIKE regexes — avoids re-compilation per row. */
 const likeRegexCache = new Map<string, RegExp>();