Validate that unique constraint name is unique #427
Conversation
|
good catch! I guess we want the same for the other constraint types? checks & foreign keys |
|
Yep, those are mentioned in this issue too. I just wanted to make sure this was the correct approach before doing the others. |
|
On the one hand, this is nice because we are catching issues early. On the other hand, it means we need to manually think of all the cases where things could go wrong and check for them during validation. I wonder if there is maybe a way to let Postgres find these issues early by running our complete step in an isolated transaction with no intention of committing it first? |
|
I guess this depends on how likely this is, detecting an issue at validate time is more valuable as we don't even touch the DB and can potentially be used while editing a migration. It feels like this doesn't get much more complex? we are probably covering all constraint types in that issue |
|
I should have been a little clearer, I didn't mean for just the constraint checks. I'm sure there are a range of things that could go wrong at completion time that we might be able to have Postgres check for us. Something to keep in mind, but agree for this specific issue we can do it manually. |
|
ah yes, definitely we should try to detect any possible issue at validate or start time. Ideally, a successful start should guarantee complete! so that's something to take into account |
I'm not sure this is exactly what we want, but putting it up here for review.
This will now catch the case when a unique constraint is added with a name that already exists on the table.
All this really does in shift the check to the
Startoperation instead of theCompleteoperation.Before this changed we get this error:
Afterwards we get this:
Part of #105