Re: Pass ParseState as down to utility functions.

On Thu, Dec 12, 2024 at 12:38:06PM +0800, jian he wrote:
> I am using DefineDomain(ParseState *pstate, CreateDomainStmt *stmt) for now.
> we can also pass querystring or another struct.

All the changes in the alternate outputs for collate are always tricky
to track. As far as I can see, you've not missed a spot. So applied
this one.

> 0002:
> passing AlterTableUtilityContext for some ALTER TABLE subroutine.
> add parser_errposition to some existing `ereport(ERROR` places.
> -------------------
> you mentioned ALTER DOMAIN, I have further simplified it at
> https://postgr.es/m/CACJufxG0N_WLfk-NC_k5w6vv26qLvXupbHvnkKtc2npftJQicQ@mail.gmail.com

- likeType = typenameType(NULL, defGetTypeName(likeTypeEl), NULL);
+ likeType = typenameType(pstate, defGetTypeName(likeTypeEl), NULL);

The only test impacted by this change is the CREATE TYPE (LIKE) in
float8. It seems like this should be separated as a change of its own
as it impacts its own command.

For the rest, we're just manipulating ATExecAddOf(),
ATPrepAlterColumnType() and ATExecAlterColumnType(). FWIW, I'm
feeling annoyed with these new make_parsestate() calls, also knowing
that we do it twice for the prep and exec parts of AlterColumnType.
Perhaps that's fine at the end, that's just an increase of calls to
make_parsestate(), still...

> like this command will fail, so we don't need to change
> create_domain.sgml synopsis section?

Yep, right. I was getting the impression that it would be possible to
have these ones go through with the parser allowed them when I looked
at that last Thursday. Will double-check to be sure.
--
Michael