At the RelNode level, you can use RelMetadataQuery.getColumnOrigins.
But I’m sure that there’s something available at the SqlNode level, after
validation. PreparedExplain.getFieldOrigins() is what the JDBC driver uses to
provide ResultSetMetadata.getColumnName(int) etc.
Julian
> On Apr 16, 2020, at 4:17 PM, Feng Zhu <wellfeng...@gmail.com> wrote:
>
> You can customize a RelVisitor to visit the RelNode, and collect the
> columns from TableScan's rowtype.
>
> 王驰 <wang...@kuaishou.com> 于2020年4月16日周四 下午11:23写道：
>
>> Hi folks,
>>
>>
>> We're using CalCite in our project for SQL parsing and validating. We have
>> a question: is there any way to get all used columns from a SQL?
>>
>>
>> Consider the following case:
>>
>> we have two tables `user` and `user_region`. And the SQL is like
>>
>>
>> ```
>>
>> SELECT
>>
>> id, name, age, country, province, city
>>
>> FROM
>>
>> user
>>
>> INNER JOIN user_region ON user.id = user_region.user_id
>>
>> WHERE age > 18;
>>
>> ```
>>
>>
>> The result will be a Map with two keys:
>>
>>
>> - 'user' --> List('id', 'name', 'age')
>> - 'user_region' --> List('country', 'province', 'city', 'user_id')
>>
>>
>> ======
>>
>> I've tried with SqlValidator (along with SqlValidatorScope,
>> SqlValidatorNamespace) but found no easy ways to do this.
>> I also tried to replay the validate process in our code but it seems
>> impossible since most classes used are package private.
>>
>>
>> Could you please give us some suggestions? Thanks!
>>
>>
