Because this is a potentially serious issue if the user doesn't use separate databases, I'm going to approve the docs PR as an interim measure. We can remove the warning from the docs when this work is done.

Another case that could have been prevented by this: #20842

We already check that url ≠ shadow database url (https://github.com/prisma/prisma-engines/blob/d7eca61ec34d27473cb990402ce406299191[…]ed/schema-engine/connectors/sql-schema-connector/src/flavour.rs — and in the submodules to see where that is used). That wasn't implemented for directUrl when directUrl was added.

@tomhoule Can you provide some more guidance on what approach I can take to try resolve this issue?

The gist of the change would be that if directUrl is defined, we would need to compare (in the function linked above) the directUrl with the shadowDatabaseUrl, and otherwise keep the existing comparison (between url and shadowDatabaseUrl).

@tomhoule I am keen on contributing to this issue, but from what I understand you are saying that this validation should be performed at the engine side right therefore this issue belong to the engine repo? As I see the current client implementation also just feeds the entire schema as a string to the engine, so to perform validation at client side would require some parsing which I assume you think is not the best approach?

Hi @zlimez — I don't work at Prisma anymore so I'm not responsible for the direction. It could be done either way, the TS code has access to some of the engines parser API.

Oh okok thank you I understand @janpio may I know what would be the desired approach adding to the engine backend check or validate before passing the schema to the engine

Yes, this should happen in the engine.
We track all issues in this repository.