PostgreSQL notes (10) dynamically execute syntax parsing process

background

Postgresql in PLPGSQL Support dynamic splicing SQL And implement ：
https://www.postgresql.org/docs/current/plpgsql-statements.html#PLPGSQL-STATEMENTS-EXECUTING-DYN
for example ：

EXECUTE 'SELECT count(*) FROM mytable WHERE inserted_by = $1 AND inserted <= $2'
   INTO c
   USING checked_user, checked_date;

This article briefly analyzes EXECUTE Execute the process .

test case

drop table u1tbl;
create table u1tbl(i int);
insert into u1tbl values (1);
insert into u1tbl values (2);
insert into u1tbl values (3);

drop function f1();
CREATE OR REPLACE FUNCTION f1() RETURNS int AS $$
DECLARE
	id int;
	i1 int := 1;
	i2 int := 2;
BEGIN
    EXECUTE 'SELECT sum(i) FROM u1tbl WHERE i >= $1 AND i <= $2'
        INTO id
        USING i1, i2;	
    return id;
END;
$$ LANGUAGE plpgsql;

select f1();

postgres=# select f1();
 f1 
----
  3
(1 row)

EXECUTE Analytical process

EXECUTE 'SELECT sum(i) FROM u1tbl WHERE i >= $1 AND i <= $2' INTO id USING i1, i2;

1、 for the first time read_sql_construct Will be called many times lex Put what you need SQL Extract it all , The actual situation is for the first time lex after , Will be able to EXECUTE after '' Take out all the text in , Give a SCONST Of token.

The process is as follows ：

EXECUTE 'SELECT sum(i) FROM u1tbl WHERE i >= $1 AND i <= $2' INTO id USING i1, i2;	
         |---------------     SCONST     -----------------|  K_INTO  K_USING


[pl_gram.y]     stmt_dynexecute : K_EXECUTE      read_sql_construct
[   scan.l]     {xqstart}                        BEGIN(xq);
[   scan.l]     <xq,xus>{xqinside}               addlit(yytext, yyleng, yyscanner);
[   scan.l]     <xb,xh,xq,xe,xus>{quote}         BEGIN(xqs);
[   scan.l]     <xqs><<EOF>>                     BEGIN(INITIAL);  return SCONST;
[pl_gram.y]

2、 Later in the circulation body read_sql_construct Will be able to USING Everything in the back SQL Or read the variable name , Press , Separate . every last SQL Or the variable is recorded as a expr Hang it as a linked list PLpgSQL_stmt_dynexecute->param after .

Last returned PLpgSQL_stmt_dynexecute structure ：

{
  cmd_type = PLPGSQL_STMT_DYNEXECUTE, 
  lineno = 7, 
  stmtid = 1, 
  query = 0x1a3a328, <PLpgSQL_expr> 'SELECT sum(i) FROM u1tbl WHERE i >= $1 AND i <= $2'
  into = true, 
  strict = false, 
  target = 0x1a3a490, <PLpgSQL_variable> dno=4 --> ((PLpgSQL_row*)plpgsql_Datums[4])
  params = 0x1a3a600  List: <PLpgSQL_expr> <PLpgSQL_expr>
}

pl_gram.y Related to the source code

stmt_dynexecute : K_EXECUTE
					{
						PLpgSQL_stmt_dynexecute *new;
						PLpgSQL_expr *expr;
						int endtoken;

						expr = read_sql_construct(K_INTO, K_USING, ';',
												  "INTO or USING or ;",
												  RAW_PARSE_PLPGSQL_EXPR,
												  true, true, true,
												  NULL, &endtoken);
	
						new = palloc(sizeof(PLpgSQL_stmt_dynexecute));
						new->cmd_type = PLPGSQL_STMT_DYNEXECUTE;
						new->lineno = plpgsql_location_to_lineno(@1);
						new->stmtid = ++plpgsql_curr_compile->nstatements;
						new->query = expr;
						new->into = false;
						new->strict = false;
						new->target = NULL;
						new->params = NIL;
	
						/*
						 * We loop to allow the INTO and USING clauses to
						 * appear in either order, since people easily get
						 * that wrong.  This coding also prevents "INTO foo"
						 * from getting absorbed into a USING expression,
						 * which is *really* confusing.
						 */
						for (;;)
						{
							if (endtoken == K_INTO)
							{
								if (new->into)			/* multiple INTO */
									yyerror("syntax error");
								new->into = true;
								read_into_target(&new->target, &new->strict);
								endtoken = yylex();
							}
							else if (endtoken == K_USING)
							{
								if (new->params)		/* multiple USING */
									yyerror("syntax error");
								do
								{
									expr = read_sql_construct(',', ';', K_INTO,
															  ", or ; or INTO",
															  RAW_PARSE_PLPGSQL_EXPR,
															  true, true, true,
															  NULL, &endtoken);
									new->params = lappend(new->params, expr);
								} while (endtoken == ',');
							}
							else if (endtoken == ';')
								break;
							else
								yyerror("syntax error");
						}
	
						$$ = (PLpgSQL_stmt *)new;
					}
				;

SELECT sum(i) FROM u1tbl WHERE i >= $1 AND i <= $2’