Problems with loading large clob values via returning clauses

[weiznich]

I've tried the following code:

use oracle::*; // oracle = "0.5.6"

const CREATE_TEST_TABLE: &str = "CREATE TABLE test_table (\
        text clob
    )";

fn repro(conn: Connection) {
    let _ = conn.execute("DROP TABLE test_table", &[]);
    conn.execute(CREATE_TEST_TABLE, &[]).unwrap();
    let mut stmt = conn
        .statement("INSERT INTO test_table(text) VALUES (:in1) RETURNING text INTO :out1")
        .build()
        .unwrap();
    let long_text = std::iter::repeat('a').take(4000).collect::<String>();
    stmt.execute_named(&[("in1", &long_text), ("out1", &None::<String>)])
        .unwrap();
    let s: &String = &stmt.returned_values("out1").unwrap()[0];
    assert_eq!(s, &long_text);
}

I expect that this code passes without an error for a valid connection. Instead of that I get the following error message:

thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: OciError(DbError { code: 22835, offset: 53, message: "ORA-22835: Buffer too small for CLOB to CHAR or BLOB to RAW conversion (actual: 4000, maximum: 2000)", fn_name: "dpiStmt_execute", action: "execute" })', src/main.rs:16:10

(where line main.rs:16 refers to the unwrap after the execute_named)

[kubo]

There are three ways. The second and third ones are usually unacceptable.

1. Use OracleType::CLOB in place of None::<String> and get the CLOB value as Clob.
2. Use OracleType::Varchar2(4000) in place of None::<String> if the maximum size of CLOB data is 4000.
3. Use OracleType::Varchar2(32767) in place of None::<String> and change the Oracle initialization parameter MAX_STRING_SIZE to EXTENDED if the maximum size of CLOB data is 32767. (I have not tested this...)

use oracle::sql_type::{Clob, OracleType};
use oracle::Connection;
use std::io::Read; // for clob.read_to_string

const CREATE_TEST_TABLE: &str = "CREATE TABLE test_table (\
        text clob
    )";

fn repro(conn: Connection) {
    let _ = conn.execute("DROP TABLE test_table", &[]);
    conn.execute(CREATE_TEST_TABLE, &[]).unwrap();
    let mut stmt = conn
        .statement("INSERT INTO test_table(text) VALUES (:in1) RETURNING text INTO :out1")
        .build()
        .unwrap();
    let long_text = std::iter::repeat('a').take(4000).collect::<String>();
    stmt.execute_named(&[("in1", &long_text), ("out1", &OracleType::CLOB)])
        .unwrap();
    // https://docs.rs/oracle/latest/oracle/sql_type/struct.Clob.html
    let mut clob: Clob = stmt.returned_values("out1").unwrap().remove(0);
    let mut s = String::new();
    clob.read_to_string(&mut s).unwrap();
    assert_eq!(&s, &long_text);
}

[weiznich]

Thanks for the fast answer. For context the real code casing this is coming from diesel-oci, so it's much more generic and cannot assume much about the actual fields. Solution 2 will not work there. Solution 1 works for CLOB columns, but won't work if the user tries to load a VARCHAR column (diesel-oci does currently not differentiate between these two types). Using the maximal-size variant sounds like it has some negative implications, is that correct?