Re: Weird behaviour of ROLLUP/GROUPING - Mailing list pgsql-general

From Guillaume Lelarge
Subject Re: Weird behaviour of ROLLUP/GROUPING
Date
Msg-id CAECtzeWeLfm3_VOg5RzNVAgh5Zs051qrkoOmMZTk65cS_tZ8PA@mail.gmail.com
Whole thread Raw
In response to Re: Weird behaviour of ROLLUP/GROUPING  (Andrew Gierth <andrew@tao11.riddles.org.uk>)
Responses Re: Weird behaviour of ROLLUP/GROUPING  (Andrew Gierth <andrew@tao11.riddles.org.uk>)
List pgsql-general
Le mer. 16 janv. 2019 à 17:40, Andrew Gierth <andrew@tao11.riddles.org.uk> a écrit :
>>>>> "Andrew" == Andrew Gierth <andrew@tao11.riddles.org.uk> writes:
>>>>> "Guillaume" == Guillaume Lelarge <guillaume@lelarge.info> writes:

 Guillaume> CASE grouping(to_char(b, 'MMYYYY'))

 Guillaume> ERROR:  arguments to GROUPING must be grouping expressions of the
 Guillaume> associated query level
 Guillaume> LINE 3: CASE grouping(to_char(b, 'MMYYYY')) WHEN 1 THEN 'some date' ...

 Guillaume> AFAICT, both queries should behave the same, though their
 Guillaume> actual behaviours are quite opposite. Working fine for the
 Guillaume> first, erroring out on the second.

 Guillaume> Does anyone has any idea what's going on here?

 Andrew> Not yet. But I will find out, since it seems to be a bug.

It is definitely a bug, to do with assignment of collations. It
specifically happens when you use GROUPING which contains any
subexpression of a collatable type, inside a CASE expr WHEN clause,
since that assigns collations to the expression much earlier in parsing
than the rest of the query, so the code that validates GROUPING ends up
trying to compare an expression which has had collations assigned to it
to one which has not, and so it thinks they differ.

I will see about fixing this, somehow.


Thanks a lot.


--
Guillaume.

pgsql-general by date:

Previous
From: Andrew Gierth
Date:
Subject: Re: Weird behaviour of ROLLUP/GROUPING
Next
From: Sam Patterson
Date:
Subject: Varlena with recursive data structures?