Re: Command Triggers, patch v11
От | Dimitri Fontaine |
---|---|
Тема | Re: Command Triggers, patch v11 |
Дата | |
Msg-id | m2fwdhjvyq.fsf@2ndQuadrant.fr обсуждение исходный текст |
Ответ на | Re: Command Triggers, patch v11 (Thom Brown <thom@linux.com>) |
Ответы |
Re: Command Triggers, patch v11
|
Список | pgsql-hackers |
Hi, Please find attached v15 of the patch, addressing all known issues apart from the trigger function argument passing style. Expect a new patch with that taken care of early next week. (The github branch too, should you be using that) Thom Brown <thom@linux.com> writes: > CREATE OPERATOR CLASS now shows objectname as 'hash' instead of its > name where it's declared as "USING hash". This isn't a problem with > ALTER/DROP OPERATOR CLASS. Everything else above works as expected > now. Ah yes that needed a special case, it's properly handled now, and tested. >>> When dropping domains, the name of the domain includes the schema name: Fixed. > Could we change this to "REINDEX DATABASE triggers are not supported"? > This way it would be consistent with the "AFTER CREATE INDEX > CONCURRENTLY" warning. Sure, done. >>> REINDEX on a table seems to show no schema name but an object name for >>> specific triggers: Was a typo, fixed. >>> When REINDEXing an index rather than a table, the table's details are >>> shown in the trigger. Is this expected?: Fixed. > ALTER CAST is still listed and needs removing, not just from the > documentation but every place it's used your code too. I can > currently create a trigger for it, but it's impossible for it to fire > since there's no such command. Removed. > All these corrections I mentioned previously still needs to be made: That's about the docs, I edited them accordingly to your comments. >> Thom Brown <thom@linux.com> writes: > All other command triggers don't fire on read-only standbys, and the > inconsistency doesn't seem right. On the one hand all > CREATE/DROP/ALTER triggers aren't fired because of the "cannot execute > <command> in a read-only transaction" error message, but triggers do > occur before utility commands, which would otherwise display the same > message, and might not display it at all if the trigger causes an > error in its function call. So it seems like they should either all > fire, or none of them should. What are you thoughts? The others trigger don't fire because an ERROR case is detected before they have a chance to run, much like on a primary in some ERROR cases. Thom Brown <thom@linux.com> writes: > It was late last night and I forgot to get around to testing pg_dump, > which isn't working correctly: Fixed. > Also I notice that CREATE/ALTER/DROP COMMAND TRIGGER appears just > before CREATE/ALTER/DROP TRIGGER in the documentation. This breaks > the alphabetical order and I wasn't expecting to find it there when > scanning down the page. Could we move them into an alphabetic > position? I don't see that problem in the source files, could you be more specific? Regards, -- Dimitri Fontaine http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
Вложения
В списке pgsql-hackers по дате отправления: