On Tue, 2006-02-14 at 14:32 -0500, Kris Jurka wrote:
> This patch adds most of the options available for regular CREATE TABLE 
> syntax to the CREATE TABLE x AS SELECT ... and AS EXECUTE ... 
> Specifically this allows specification of on commit behavior for temp 
> tables and tablespaces for regular tables to these two statements.

The implementation is pretty ugly -- it clutters ExecuteStmt and Query
with fields that really do not belong there. Per previous discussion, I
think it would be better to refactor the CREATE TABLE AS implementation
to be essentially a CREATE TABLE followed by a INSERT ... SELECT.

(That's not necessarily a reason to reject the patch, but the patch does
increase the benefit of performing that refactoring.)

A few cosmetic comments:

typedef struct ExecuteStmt
{
        NodeTag                 type;
        char               *name;
        RangeVar           *into;
        ContainsOids    intocontainsoids;
        bool                    intohasoids;
        OnCommitAction  intooncommit;
        char               *intotablespacename;
        List               *params;
} ExecuteStmt;

I think we ought to use either camel-case or underscore characters to
separate words.

parser/analyze.c, circa 1822:

    if (stmt->intoTableSpaceName)
        qry->intoTableSpaceName = pstrdup(stmt->intoTableSpaceName);
    else
        qry->intoTableSpaceName = NULL;

You can omit the "else", as makeNode() zeroes all the fields of the new
node. (You could argue that leaving the assignment is more readable, but
I personally don't think so: this behavior of makeNode() is used in a
several places in the backend.)

-Neil



---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
       choose an index scan if your joining column's datatypes do not
       match

Reply via email to