Fix issue #1032

[eeeebbbbrrrr]

Blindly use the FMGR_ABI_EXTRA constant as part of the magic block. Hardcoding to "PostgreSQL" isn't exactly correct when using bindings from, for example, a Postgres fork that has purposely changed its ABI "name".

[workingjubilee]

This violates the compile-time guarantees we have provided until now. That is: we have previously assumed the responsibility for making PGX work against Postgres.

We should not enable this change except conditionally on the user explicitly requesting this be overwritten at compile time, in a way that we can detect when an issue is opened against PGX. It should be the user's responsibility, not ours, for whether their extension works against a bespoke Postgres fork.

[eeeebbbbrrrr]

This violates the compile-time guarantees we have provided until now. That is: we have previously assumed the responsibility for making PGX work against Postgres.

We should not enable this change except conditionally on the user explicitly requesting this be overwritten at compile time, in a way that we can detect when an issue is opened against PGX. It should be the user's responsibility, not ours, for whether their extension works against a bespoke Postgres fork.

What are you thinking here? --features this_isnt_postgres?

[workingjubilee]

I think a feature name can use all of Unicode, so we could go with "unsafe-postgres-doppelgänger" so it's annoying to type.

[eeeebbbbrrrr]

and do we raise a compilation error if it's not "PostgreSQL" and that feature doesn't exist?

Doing this pushes the burden of supporting Postgres forks onto pgx extension developers. Maybe that's where it belongs tho. We can't know everything.

[workingjubilee]

Yes, I think so. We could also lift this checking into cargo pgx, effectively, and then it's the fault of whoever runs the compiler.

[eeeebbbbrrrr]

Yes, I think so. We could also lift this checking into cargo pgx, effectively, and then it's the fault of whoever runs the compiler.

Well, one can compile a pgx extension without cargo-pgx, so I don't think that's super valuable. Lets go with a feature flag of --features unsafe-postgres and a compilation error without it.

I'll update the PR and we can see what it looks like.

[eeeebbbbrrrr]

this isn't working out like I had hoped. #grr

[thomcc]

Looks good to me (but I did write some of it 😅)

[eeeebbbbrrrr]

You wrote all of it. In the discord input box!