Обсуждение: The document contains false statements
The following documentation comment has been logged on the website: Page: https://www.postgresql.org/docs/17/sql-createfunction.html Description: > This is effectively another way of declaring a named OUT parameter Actually OUT works twice slower in compare to RETURNS TABLE https://stackoverflow.com/q/79763947/4632019 DB<>fiddle for [`OUT`](https://dbfiddle.uk/fz9L_wm0) and [`RETURNS TABLE`](https://dbfiddle.uk/uTkU1MT8) cases. *I hope after the fix, data centers will consume 2 times less electricity :D.
PG Doc comments form <noreply@postgresql.org> writes: > The following documentation comment has been logged on the website: > Page: https://www.postgresql.org/docs/17/sql-createfunction.html > Description: >> This is effectively another way of declaring a named OUT parameter > Actually OUT works twice slower in compare to RETURNS TABLE You stopped reading halfway through that sentence... it says This is effectively another way of declaring a named OUT parameter, except that RETURNS TABLE also implies RETURNS SETOF. The difference between a set-returning and not-set-returning function might explain what you're seeing, particularly because the rules for inlining SQL functions are different in the two cases. A documentation comment is not the place to pursue this further, but you could ask for help on our pgsql-performance mailing list. regards, tom lane
On Saturday, September 13, 2025, PG Doc comments form <noreply@postgresql.org> wrote:
The following documentation comment has been logged on the website:
Page: https://www.postgresql.org/docs/17/sql-createfunction. html
Description:
> This is effectively another way of declaring a named OUT parameter
Actually OUT works twice slower in compare to RETURNS TABLE
https://stackoverflow.com/q/79763947/4632019
DB<>fiddle for [`OUT`](https://dbfiddle.uk/fz9L_wm0) and [`RETURNS
TABLE`](https://dbfiddle.uk/uTkU1MT8) cases.
*I hope after the fix, data centers will consume 2 times less electricity
The statement is not false - it contains an “except” clause that you’ve ignored which makes it true for exactly this reason. The fact you are comparing a set-returning function to one that doesn’t return a set has invalidated the test.
The fundamental issue here is “select (composite_func()).*” where the function is not set-returning if known to be broken - the “*” expansion during planning results in the function being executed multiple times once for each output column. (I may be missing some nuances here as, since the inclusion of lateral joins, this almost never comes up anymore.)
Non-trivial function calls should be placed in the FROM clause of a query; in part to ensure avoidance of this problematic behavior.
This is not at all limited to RLS.
In short, I don’t know how to improve the documentation to prevent people from writing bad queries of this type. Concrete suggestions are welcome, but removing this sentence, or re-wording it, doesn’t seem like it would make any difference.
David J.
Hello Tom, Saturday, September 13, 2025, 7:25:08 PM, you wrote: > PG Doc comments form <noreply@postgresql.org> writes: >> The following documentation comment has been logged on the website: >> Page: https://www.postgresql.org/docs/17/sql-createfunction.html >> Description: >>> This is effectively another way of declaring a named OUT parameter >> Actually OUT works twice slower in compare to RETURNS TABLE > You stopped reading halfway through that sentence... it says > This is effectively another way of declaring a named OUT parameter, > except that RETURNS TABLE also implies RETURNS SETOF. > The difference between a set-returning and not-set-returning > function might explain what you're seeing, particularly because > the rules for inlining SQL functions are different in the two > cases. > A documentation comment is not the place to pursue this further, > but you could ask for help on our pgsql-performance mailing list. > regards, tom lane Let me continue how to improve this in David G. thread. -- Best regards, Eugen Konkov
Hello David, Saturday, September 13, 2025, 7:26:21 PM, you wrote: > On Saturday, September 13, 2025, PG Doc comments form <noreply@postgresql.org> wrote: > The following documentation comment has been logged on the website: > Page: https://www.postgresql.org/docs/17/sql-createfunction.html > Description: >> This is effectively another way of declaring a named OUT parameter > Actually OUT works twice slower in compare to RETURNS TABLE > https://stackoverflow.com/q/79763947/4632019 > DB<>fiddle for [`OUT`](https://dbfiddle.uk/fz9L_wm0) and [`RETURNS > TABLE`](https://dbfiddle.uk/uTkU1MT8) cases. > *I hope after the fix, data centers will consume 2 times less electricity > The statement is not false - it contains an “except” clause that > you’ve ignored which makes it true for exactly this reason. The > fact you are comparing a set-returning function to one that doesn’t > return a set has invalidated the test. > The fundamental issue here is “select (composite_func()).*” where > the function is not set-returning if known to be broken - the “*” > expansion during planning results in the function being executed > multiple times once for each output column. (I may be missing some > nuances here as, since the inclusion of lateral joins, this almost never comes up anymore.) > Non-trivial function calls should be placed in the FROM clause of a > query; in part to ensure avoidance of this problematic behavior. > This is not at all limited to RLS. > In short, I don’t know how to improve the documentation to prevent > people from writing bad queries of this type. Concrete suggestions > are welcome, but removing this sentence, or re-wording it, doesn’t > seem like it would make any difference. > David J. Thanks for more information on this. You and Tom both pointed me to `RETURNS SET OF` part. I agree with Tom that documentationis not the place to teach user how to write SQL. But I would appeal that the documentation should be meaning full. And for me the part after "except" looks the same as itwould be written in Arabic "إرجاع مجموعة من" (RETURNS SETOF). The question from David: How it could be done better? is good. Let me describe how I see it from my point of view and experience. The documentation above highlights as RETURNS SETOF as something special. I never used SETOF. I tried to google and findalmost nothing in the official documentation, except these two https://www.postgresql.org/docs/current/xfunc-sql.html#XFUNC-SQL-FUNCTIONS-RETURNING-SET https://www.postgresql.org/docs/current/xfunc-sql.html#XFUNC-SQL-FUNCTIONS-RETURNING-TABLE with just a mention that it can return 0, 1 or more rows. And the most informative one is https://www.postgresql.org/docs/current/queries-table-expressions.html#QUERIES-TABLEFUNCTIONS The latest one my my mind should be reffered from https://www.postgresql.org/docs/current/xfunc-sql.html#XFUNC-SQL-FUNCTIONS-RETURNING-TABLE, Eg. Please read 7.2.1.4 Table Functions for more information. For the original problem: >The name of an output column in the RETURNS TABLE syntax. This is effectively another way of declaring a named OUT parameter, > except that RETURNS TABLE also implies RETURNS SETOF. My proposition is to add link to 7.2.1.4 Table Functions for more information and extend "7.2.1.4 Table Functions for moreinformation" with the information that "in certain circumstances RETURNS TABLE is a subject for optimizer and could be inlined https://wiki.postgresql.org/wiki/Inlining_of_SQL_functions#Inlining_conditions_for_table_functions". Having this it would be clear that `RETURNS TABLE` and `RETURNS SETOF` are sort of fuctions which are called Table functions. Probably official documentation should have in future a section somewhere "How to optimize your queries" and links to benchmarkslike this I did in my question on SO: https://pastebin.com/n3sxBxt6 https://dbfiddle.uk/xfy-qw75 Without this information and statement that these two: OUT and RETURNS TABLE are just taste of syntax, users will use eitherwithout knowing consequences. (like me until I benchmarked it) -- Best regards, Eugen Konkov -- Best regards, Eugen Konkov