From 9421f9ef62f5569e069f8b651eaeb6e5fa2eb89c Mon Sep 17 00:00:00 2001
From: chuanghiduoc <levanbaotrong03@gmail.com>
Date: Fri, 23 Oct 2026 16:55:51 +0700
Subject: [PATCH] fix(csv-parse): wait for full CRLF before auto-detecting
 record delimiter

---
 packages/csv-parse/lib/api/index.js        | 18 +++++++-------
 packages/csv-parse/test/option.encoding.ts | 28 ++++++++++++++++++++++
 2 files changed, 37 insertions(+), 9 deletions(-)

diff --git a/packages/csv-parse/lib/api/index.js b/packages/csv-parse/lib/api/index.js
index ca97838cd..17d59ffe8 100644
--- a/packages/csv-parse/lib/api/index.js
+++ b/packages/csv-parse/lib/api/index.js
@@ -48,21 +48,21 @@ const transform = function (original_options = {}) {
       const { quoting, needMoreDataSize, recordDelimiterMaxLength } =
         this.state;
       const numOfCharLeft = bufLen - i - 1;
+      // When the record delimiter has not been discovered yet, auto-detection
+      // needs enough bytes to distinguish "\r\n" from "\r" in the current
+      // encoding. This also applies to the delimiter following a closing quote.
+      const recordDelimiterLookahead =
+        recordDelimiterMaxLength === 0
+          ? Buffer.from("\r\n", encoding).length
+          : recordDelimiterMaxLength;
       const requiredLength = Math.max(
         needMoreDataSize,
         // Skip if the remaining buffer smaller than record delimiter
-        // If "record_delimiter" is yet to be discovered:
-        // 1. It is equals to `[]` and "recordDelimiterMaxLength" equals `0`
-        // 2. We set the length to windows line ending in the current encoding
-        // Note, that encoding is known from user or bom discovery at that point
-        // recordDelimiterMaxLength,
-        recordDelimiterMaxLength === 0
-          ? Buffer.from("\r\n", encoding).length
-          : recordDelimiterMaxLength,
+        recordDelimiterLookahead,
         // Skip if remaining buffer can be an escaped quote
         quoting ? (escape === null ? 0 : escape.length) + quote.length : 0,
         // Skip if remaining buffer can be record delimiter following the closing quote
-        quoting ? quote.length + recordDelimiterMaxLength : 0,
+        quoting ? quote.length + recordDelimiterLookahead : 0,
       );
       return numOfCharLeft < requiredLength;
     },
diff --git a/packages/csv-parse/test/option.encoding.ts b/packages/csv-parse/test/option.encoding.ts
index 1d3c53491..55099c39b 100644
--- a/packages/csv-parse/test/option.encoding.ts
+++ b/packages/csv-parse/test/option.encoding.ts
@@ -63,6 +63,34 @@ describe("Option `encoding`", function () {
     });
   });
 
+  describe("auto record delimiter", function () {
+    it("utf16le with a quoted record split across writes", function (next) {
+      const input = Buffer.from('"b"\r\n', "utf16le");
+      const parser = parse({ encoding: "utf16le" }, (err, records) => {
+        if (err) return next(err);
+        records.should.eql([["b"]]);
+        next();
+      });
+      for (let i = 0; i < input.length; i++) {
+        parser.write(input.subarray(i, i + 1));
+      }
+      parser.end();
+    });
+
+    it("utf16le with multiple quoted records split across writes", function (next) {
+      const input = Buffer.from('"b"\r\n"c"\r\n', "utf16le");
+      const parser = parse({ encoding: "utf16le" }, (err, records) => {
+        if (err) return next(err);
+        records.should.eql([["b"], ["c"]]);
+        next();
+      });
+      for (let i = 0; i < input.length; i++) {
+        parser.write(input.subarray(i, i + 1));
+      }
+      parser.end();
+    });
+  });
+
   describe("with BOM", function () {
     it("utf16le auto detected", function (next) {
       const parser = parse(