Thread: Allow relocatable extension to use @extschema@?
Hi! During work on knn-btree patchset we've faced the need to move functions/operators from contrib to core [1]. In the extension upgrade script we need to use @extschema@ in order to distinguish contrib and core objects. However, it appears to be possible to use @extschema@ only in non-relocatable extensions. Comment in extension.c says: "For a relocatable extension, we needn't do this. There cannot be any need for @extschema@, else it wouldn't be relocatable.". I've explored that we've marked extension as non-relocatable solely to use @extschema@ in script before [2]. So, it appears that comment in extension.c isn't true. There is at least two precedents when relocatable extension needs to use @extschema@. We've marked possibly relocatable extension as non-relocatable once. And we could do it at the second time. Extension relocatability doesn't seem to me much value to sacrifice. But nevertheless should we allow relocatable extension to use @extschema@. Any thoughts? Links. 1. https://www.postgresql.org/message-id/CAPpHfdsWsb9T1eHdX%2Br7wnXbGJKQxSffc8gTGp4ZA2ewP49Hog%40mail.gmail.com 2. https://git.postgresql.org/gitweb/?p=postgresql.git;a=commitdiff;h=de623f33353c96657651f9c3a6c8756616c610e4;hp=0024e348989254d48dc4afe9beab98a6994a791e ------ Alexander Korotkov Postgres Professional: http://www.postgrespro.com The Russian Postgres Company
Alexander Korotkov <a.korotkov@postgrespro.ru> writes:
> But nevertheless should we allow relocatable extension to use
> @extschema@. Any thoughts?
No. The reasoning in the comment still holds good: if you embed
@extschema@ in an object's definition, it becomes nonrelocatable.
regards, tom lane
On Tue, Dec 3, 2019 at 6:18 PM Tom Lane <tgl@sss.pgh.pa.us> wrote: > Alexander Korotkov <a.korotkov@postgrespro.ru> writes: > > But nevertheless should we allow relocatable extension to use > > @extschema@. Any thoughts? > > No. The reasoning in the comment still holds good: if you embed > @extschema@ in an object's definition, it becomes nonrelocatable. I see, allowing @extschema@ in non-relocatable extension provides easy way to shoot yourself in the foot. However, it might be still useful to be able to distinguish extension and core object in upgrade script of relocatable extensions. What about (optional) way to set search_path to @extschema@, pg_catalog instead of just @extschema@? ------ Alexander Korotkov Postgres Professional: http://www.postgrespro.com The Russian Postgres Company
On Tue, Dec 3, 2019 at 10:18 AM Tom Lane <tgl@sss.pgh.pa.us> wrote: > Alexander Korotkov <a.korotkov@postgrespro.ru> writes: > > But nevertheless should we allow relocatable extension to use > > @extschema@. Any thoughts? > > No. The reasoning in the comment still holds good: if you embed > @extschema@ in an object's definition, it becomes nonrelocatable. That's demonstrably false, and the links included in the OP prove it. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
On Thu, Dec 05, 2019 at 09:52:50AM -0500, Robert Haas wrote: > On Tue, Dec 3, 2019 at 10:18 AM Tom Lane <tgl@sss.pgh.pa.us> wrote: > > Alexander Korotkov <a.korotkov@postgrespro.ru> writes: > > > But nevertheless should we allow relocatable extension to use > > > @extschema@. Any thoughts? > > > > No. The reasoning in the comment still holds good: if you embed > > @extschema@ in an object's definition, it becomes nonrelocatable. > > That's demonstrably false, and the links included in the OP prove it. Right. Or, to add further precision, embedding @extschema@ persistently in the system catalogs does make the extension nonrelocatable. The OP examples merely embed @extschema@ in extension scripts, and they don't impair relocation. I'm okay with the proposal, but I still wouldn't mind deprecating relocatable=true.