Codify why we don't commit directly after a read call is finished

src/Npgsql/Internal/PgReader.cs

@@ -12,6 +12,9 @@ namespace Npgsql.Internal;
 
 public class PgReader
 {
+    // We don't want to add a ton of memory pressure for large strings.
+    internal const int MaxPreparedTextReaderSize = 1024 * 64;
+
     readonly NpgsqlReadBuffer _buffer;
 
     bool _resumable;
@@ -210,11 +213,8 @@ public ValueTask<TextReader> GetTextReaderAsync(Encoding encoding, CancellationT
 
     async ValueTask<TextReader> GetTextReader(bool async, Encoding encoding, CancellationToken cancellationToken)
     {
-        // We don't want to add a ton of memory pressure for large strings.
-        const int maxPreparedSize = 1024 * 64;
-
         _requiresCleanup = true;
-        if (CurrentRemaining > _buffer.ReadBytesLeft || CurrentRemaining > maxPreparedSize)
+        if (CurrentRemaining > _buffer.ReadBytesLeft || CurrentRemaining > MaxPreparedTextReaderSize)
             return new StreamReader(GetColumnStream(), encoding, detectEncodingFromByteOrderMarks: false);
 
         if (_preparedTextReader is { IsDisposed: false })

test/Npgsql.Tests/CopyTests.cs

@@ -562,6 +562,21 @@ public async Task ReadMoreColumnsThanExist()
         }
     }
 
+    [Test]
+    public async Task StreamingRead()
+    {
+        if (IsMultiplexing)
+            Assert.Ignore("Multiplexing: fails");
+        using var conn = await OpenConnectionAsync();
+
+        var str = new string('a', PgReader.MaxPreparedTextReaderSize + 1);
+        var reader = conn.BeginBinaryExport($"""COPY (values ('{str}')) TO STDOUT BINARY""");
+        while (reader.StartRow() != -1)
+        {
+            using var _ = reader.Read<TextReader>(NpgsqlDbType.Text);
+        }
+    }
+
     [Test, IssueLink("https://github.com/npgsql/npgsql/issues/2330")]
     public async Task Wrong_format_binary_export()
     {