SQLParser | An SQL Parser/Lexer for C | SQL Database library

The LANGUAGE PYTHON UDF's are a nice development feature in MonetDB's server installation but this feature requires an additional Python module to be loaded. And there is currently no way to configure monetdbe to load the required python module.

However assuming you have performance requirement for some production setting that are not met with the out-of-the-box SQL toolset in monetdbe, it makes more sense to implement a custom UDF extension written in C/C++. In regular MonetDB's server installation, the database server mserver5 can load an arbitrary extension module using the --loadmodule= command option. But there is no equivalent monetdbe_option as of yet.

You might consider adding a feature request for this on monetdbe-python's github repository.

However there seems to exist a functioning undocumented workaround for adding UDF extensions to monetdbe. During its initialization, monetdbe attempts to load a set of hard coded modules. One of those is a module named "udf". You can create your own implementation of this module and load it into monetdbe.

Creating a native UDF extension is outside of the scope of this question and answer but there exist a nice up-to-date tutorial for writing UDF extensions for MonetDB here. Following the steps described in that tutorial, you end up with a SQL function revstr which has a user defined native implementation. The following Python script demonstrate its use:

Only one method in RelBuilder uses a RelOptSchema: scan(String...) (and its variant Scan(Iterable)). Which makes sense when you consider that the purpose of RelOptSchema is as a directory service, converting a table name (or table path, consisting of a table name qualified with catalog and/or schema names) into a RelOptTable object.

If you have 'free-standing' table objects that are not accessed via a namespace then you can create TableScan relational expressions directly and then call RelBuilder.push(RelNode) to add them to the stack. Since you never call RelBuilder.scan you can create RelBuilder with a null RelOptSchema.

But in your case, it looks as if you don't have free-standing table objects. That's a problem for Calcite, because it needs to know that your "EMP" table has a field called "DEPTNO" and it has type INTEGER.

So I suggest that you create a 'virtual' schema that contains type information but is not necessarily backed by real tables. The MockCatalogReader class, used in several of Calcite's tests, is a good example to follow.

The variable a has type &Box, so when you dereference it, *a, it only removes the reference leaving just Box. You unfortunately cannot do pattern matching on values in a Box (see How to pattern match a Box to get a struct's attribute?), so you have to dereference it again to get an Expr:

I'd suggest taking a look at TokenStreamRewriter.

First, let's get the grammars ready.

1 - with TokenStreamRewriter we'll want to preserve whitespace, so let's change the -> skip directives to ->channel(HIDDEN)

At the end of the Lexer grammar:

I think you're looking for parser.parsePlan instead of parser.parseExpression. The second query doesn't show an error for parsePlan.

However, note that

this seems to work

org.apache.spark.sql.catalyst.plans.logical.Project.getClass calls the getClass method of the companion object for org.apache.spark.sql.catalyst.plans.logical.Project which is a singleton instance of the class org.apache.spark.sql.catalyst.plans.logical.Project$ (in this case, looking at Spark's code shows that to be a case class with a synthetic companion object).

You can get the class object for org.apache.spark.sql.catalyst.plans.logical.Project with:

I'm still learning Android and Kotlin but I think the problem lies in the Database class. However, I'm not sure but hopefully it will help.

Try:

You'll have to override all Visit...(... context) calls (at least all the ones that your parse tree has in it). Let's say you have this grammar: