Horton

Re: ALTER COLUMN ... SET EXPRESSION to alter stored generated column's expression

Peter Eisentraut <peter@eisentraut.org>

View thread
From: Peter Eisentraut <peter@eisentraut.org>
To: Amul Sul <sulamul@gmail.com>
Cc: Robert Haas <robertmhaas@gmail.com>, Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>, Vaibhav Dalvi <vaibhav.dalvi@enterprisedb.com>, pgsql-hackers@postgresql.org
Date: 2023-11-20T07:42:00Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. ALTER TABLE command to change generation expression

  2. Refactor: separate function to find all objects depending on a column

  3. Turn AT_PASS_* macros into an enum

On 17.11.23 13:25, Amul Sul wrote:
> To fix this we should be doing something like ALTER COLUMN TYPE and the pass
> should be AT_PASS_ALTER_TYPE (rename it or invent a new one near to that) so
> that in ATRewriteCatalogs(), we would execute ATPostAlterTypeCleanup().
> 
> I simply tried that by doing blind copy of code from 
> ATExecAlterColumnType() in
> 0002 patch.  We don't really need to do all the stuff such as re-adding
> indexes, constraints etc, but I am out of time for today to figure out the
> optimum code and I will be away from work in the first half of the coming
> week and the week after that. Therefore, I thought of sharing an approach to
> get comments/thoughts on the direction, thanks.

The exact sequencing of this seems to be tricky.  It's clear that we 
need to do it earlier than at the end.  I also think it should be 
strictly after AT_PASS_ALTER_TYPE so that the new expression can refer 
to the new type of a column.  It should also be after AT_PASS_ADD_COL, 
so that the new expression can refer to any newly added column.  But 
then it's after AT_PASS_OLD_INDEX and AT_PASS_OLD_CONSTR, is that a problem?

(It might be an option for the first version of this feature to not 
support altering columns that have constraints on them.  But we do need 
to support columns with indexes on them.  Does that work ok?  Does that 
depend on the relative order of AT_PASS_OLD_INDEX?)