Which came first? The Select or the Into?

This morning while discussing things vaguely reminiscent of work with a colleague of mine the topic of Russell’s paradox was broached. Russell’s paradox relies the concept of a set of sets that do no contain themselves as members. It is a paradox because if that set does not contain itself it satisfies the requirement of membership and then must therefore contain itself, which would be a violation of the rules required for membership. Long story short humanity innovated and a guy named Ernst Zermelo contributed some rules that don’t allow such paradoxes. Zermelo actually discovered the paradox first but didn’t publish. That was confirmed a guy named David Hilbert who drew a really great looking space filling curve with a few simple steps repeated infinitely. The talk of Russell’s paradox prompted me to write the following SQL:

SELECT * INTO #temp FROM tempdb.sys.tables;

SELECT * FROM #temp;

I then posed the question: Will #temp appear as a member of the rows returned from #temp?


You can see an instance of #temp does in fact appear. The suffix appended to it uniquefies the name. This is what allows for multiple temp tables of the same name to be created. We can see this by running the above query in another tab (which opens another session).

Capture2The reason that we don’t see the table excluded from the results set is that in order for the temp table to hold the data from the sys.tables catalog view it must exist. And if it exists it will appear in the catalog. This helps to illustrate that SQL is not (always) a procedural language, or that INTO does not logically mean CREATE and INSERT.

SQL is usually typified of a declarative programming language. Being a declarative programming language means that we are not telling the database how to do what we want, only what we want. In this case we told it we want a table that matches the definition of sys.tables and we want it filled with the contents of sys.tables. Contrast this with the having asked SQL to take the contents of sys.tables and put it into a table that matches the definition of sys.tables. In the first case we have to first make the table which puts it into sys.tables, in the second we take the contents of sys.tables and then put it into our newly created table.

This might become clearer if we examine a query plan.


We can clearly see from the plan that sys.tables is a view that draws from the inaccessible mssqlsystemresource system database. It might look like all of those branches contribute to building the table to insert into, but that is just where the data comes from. A simpler select into plan can be seen if we select from a table valued function (TVF).


Here we can see that the Table Insert showplan operator occurs before the Select Into operator. That doesn’t actually create the table, though. The table is created by a Data Definition Language (DDL) operation. DDL operations don’t produce showplan operators. So there you have it, moderately conclusive proof that the table does exist before the read of data to insert occurs, and some vague syntactic hints at why.



Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s