Thread: WITH RECURSIVE (Documentation section 7.8.1) Note
The following documentation comment has been logged on the website: Page: https://www.postgresql.org/docs/9.4/queries-with.html Description: In section 7.8.1, the author comments "Note: Strictly speaking, this process is iteration not recursion, but RECURSIVE is the terminology chosen by the SQL standards committee." This comment is not helpful (at best) and misleading (at worst). The "process" which is the subject of the comment is based on the implementation of WITH RECURSIVE in PostgreSQL -- the implementation happens to execute a sequence of queries and a temporary table, and one could arguably describe this as iteration. However, the use of the term RECURSIVE in the syntax, refers not to any particular implementation, but to an observation that WITH RECURSIVE is effectively an example of a recursively defined function in set-theoretic terms. Consider the first of the UNIONed statements to produce a row set R_0. Consider the second of the UNIONed statements to be function F that produces a row set R_k = F(R_k-1). (The underscore is used here to denote a subscript). Example: R_2 = F(F(R_0)) Example: R_5 = F(F(F(F(F(R_0)))) This is textbook recursion, not just some folly of the ANSI standards committee as suggested by the note in the text.
PG Doc comments form <noreply@postgresql.org> writes: > In section 7.8.1, the author comments "Note: Strictly speaking, this process > is iteration not recursion, but RECURSIVE is the terminology chosen by the > SQL standards committee." I think I'm to blame for that wording. > Example: R_2 = F(F(R_0)) > Example: R_5 = F(F(F(F(F(R_0)))) > This is textbook recursion, not just some folly of the ANSI standards > committee as suggested by the note in the text. No, it isn't. F() is being applied repeatedly (iteratively). If F() called itself internally, that would be recursion. Another way to look at this is that control of whether the repetition is finished is external to the function/query. If the stop condition were part of the query logic, it'd be more sensible to think of it as recursion, IMO. regards, tom lane