value exception throws when using batch-insert to insert one row about nanodbc

[HGeorge-Adapdix]

I came across this when looking at reporting a similar issue.

Environment

nanodbc version:2.14
DBMS name/version:Postgres
ODBC connection string:Driver={PostgreSQL ANSI};Uid=postgres;Server=localhost;Port=36947
OS and Compiler: Ubuntu 22.04
CMake settings:

TEST(Example, TestStringBindSingleNullRowFails)
{
	const std::string pgConnectionString{
		"Driver={PostgreSQL ANSI};Uid=postgres;Server=localhost;Port=36947"};
	nanodbc::connection conn(pgConnectionString);

	nanodbc::just_execute(conn,
						  "create temporary table test_table (id integer, name varchar(255))");
	nanodbc::just_execute(conn, "truncate test_table");

	nanodbc::transaction transaction(conn);

	nanodbc::statement statement(conn, "insert into test_table (id, name) values (?, ?)");

	std::vector<std::string> ids;
	ids.emplace_back("1");
	const std::array<bool, 1> notNull{{false}};

	statement.bind_strings(0, ids, notNull.data());
	bool pass(false);
	if (pass) {
		// Works if bind null
		statement.bind_null(1, 1);
	} else {
		// Fails if bind string with all nulls
		const std::array<bool, 1> null{{true}};
		statement.bind_strings(1, ids, null.data());
	}

	nanodbc::just_execute(statement, static_cast<int>(ids.size()));

	transaction.commit();

	nanodbc::result result(nanodbc::execute(conn, "select * from test_table"));
	ASSERT_TRUE(result.next());
	ASSERT_FALSE(result.is_null(0));

	// This fails
	ASSERT_TRUE(result.is_null(1));
}

If you use the bind_null interface it works, but bind strings does not insert a null and instead will insert the string ignoring the "null" aspect.

[HGeorge-Adapdix]

I believe the issue is in bind_parameter

diff --git a/nanodbc/nanodbc.cpp b/nanodbc/nanodbc.cpp
index 9ece354..ed5ad3c 100644
--- a/nanodbc/nanodbc.cpp
+++ b/nanodbc/nanodbc.cpp
@@ -2239,7 +2239,7 @@ public:
             param.scale_,        // decimal digits
             (SQLPOINTER)buffer.values_, // parameter value
             buffer_size,                // buffer length
-            (buffer.size_ <= 1 ? nullptr : bind_len_or_null_[param.index_].data()));
+            (buffer_size <= 1 ? nullptr : bind_len_or_null_[param.index_].data()));
 
         if (!success(rc))
             NANODBC_THROW_DATABASE_ERROR(stmt_, SQL_HANDLE_STMT);

In the one row example the buffer.size_ is 1 and the buffer_size is 2.

Changing it fixes it, so that code is what is causing the issue but I am not 100% sure what the code is supposed to be doing (hence no PR). I am unsure as to why you wouldn't just always pass in the bind_len_or_null_ value but looking through the history that appears to have been there for many years (tracked it back as far as 2bd4e2e).

Not sure if anyone has any insight here.

[mloskot]

@x-awaken & @HGeorge-Adapdix Thank you for your reports.
Hmm, this is interesting as it looks like an overlooked place that miss the change from buffer.size_ to local variable buffer_size:

nanodbc/nanodbc/nanodbc.cpp

Line 2242 in 71b7f0b

(buffer.size_ <= 1 ? nullptr : bind_len_or_null_[param.index_].data()));

nanodbc/nanodbc/nanodbc.cpp

Line 2888 in 71b7f0b

(buffer.size_ <= 1 ? nullptr : bind_len_or_null_[param.index_].data()));

I do not recall anything specific on the use of buffer.size_ in those places, instead of buffer_size_.

What I'd suggest is to submit PR with

• change from #347 (comment)
• test case from #347 (comment)

and let's see what the CI builds think about it.
I will also try to run local tests of such PR and perhaps add more test cases.