Deprecate upsert?

[MaxGabriel]

The upsert function inserts or updates an existing record. It works fine if there's one uniqueness constraint, but if there are two it doesn't know which to use, and throws an exception. This is pretty lame: you wouldn't think that adding a new uniqueness constraint would cause a runtime exception in a related area of your code. Users should just use upsertBy, or if their database supports a way of not specifying it, use that functionality (e.g. the MySQL module exposes insertOnDuplicateKeyUpdate; I think Postgres requires specifying the column(s)).

This would resolve issues like #768 and avoid all the documentation of this caveat in #785.

[psibi]

In-spite of these caveats, I would still like to use upsert and don't want it to be deprecated.

It works fine if there's one uniqueness constraint, but if there are two it doesn't know which to use, and throws an exception.

I would like an API in which we can explicitly specify which uniqueness column it should select.

[MaxGabriel]

I would like an API in which we can explicitly specify which uniqueness column it should select.

Sorry should have clarified: we already have that, it’s upsertBy. We should just direct everyone to use that

[psibi]

In that case, I'm fine with it being deprecated. Never actually used upsertBy and I always thought it had the same caveat as upsert. 👍

[parsonsmatt]

We could keep upsert on the case that there's a single primary key with a bit of type magic. Not sure if it's worth the hassle.

[pythonissam]

@MaxGabriel @psibi
How's the status of this issue? Sadly enough, I haven't received any response with my PR for this two months...

[bitemyapp]

@pythonissam I'd lean toward deprecating upset. I think I know the kind of design @parsonsmatt alluded to and I'm not a fan. I've been working with some 9-5 code lately that uses Groundhog's equivalent to an implicit upset and it has not made my life easier or simpler.

I'm going to comment separately on your repsert PR.

[pythonissam]

@bitemyapp

I'd lean toward deprecating upset.

All right. But I think there still exists bugs around these functions. I'm going to organize my thoughts and re-submit issues separately.