[KYLIN-4069] HivePushDownConverter.doConvert will change sql semantics in some scenarios

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: v2.6.2
• Component/s: Query Engine
• Assignee: weibin0516
• Reporter: weibin0516
• Created: 03/Jul/19 02:40
• Updated: 03/Jul/19 02:40

Description

HivePushDownConverter.doConvert source code is as follows:

public static String doConvert(String originStr, boolean isPrepare) {
        // Step1.Replace " with `
        String convertedSql = replaceString(originStr, "\"", "`");

        // Step2.Replace extract functions
        convertedSql = extractReplace(convertedSql);

        // Step3.Replace cast type string
        convertedSql = castReplace(convertedSql);

        // Step4.Replace sub query
        convertedSql = subqueryReplace(convertedSql);

        // Step5.Replace char_length with length
        convertedSql = replaceString(convertedSql, "CHAR_LENGTH", "LENGTH");
        convertedSql = replaceString(convertedSql, "char_length", "length");

        // Step6.Replace "||" with concat
        convertedSql = concatReplace(convertedSql);

        // Step7.Add quote for interval in timestampadd
        convertedSql = timestampAddDiffReplace(convertedSql);

        // Step8.Replace integer with int
        convertedSql = replaceString(convertedSql, "INTEGER", "INT");
        convertedSql = replaceString(convertedSql, "integer", "int");

        // Step9.Add limit 1 for prepare select sql to speed up
        if (isPrepare) {
            convertedSql = addLimit(convertedSql);
        }

        return convertedSql;
    }

It is not advisable to directly replace the sql text. The following example will convert sql to another error sql:

SELECT "CHAR_LENGTH" FROM datasource.a

will convert to

SELECT `LENGTH` FROM datasource.a

Every use of replaceString in doConvert will cause such problems.