feat: Default to bigint = "integer"

[krlmlr]

Defaulting to integer64 does more harm than good in my opinion. Happy to discuss.

[nbenn]

One issue is that the nanoarrow int conversion is not compatible with the "silent overflow" requirement, which I feel is as good thing, especially if this is to be the default setting. But I am too happy to discuss this. I don't think defaulting to int64 is great. I just felt it was the least-worst option.

[krlmlr]

What do you mean by "silent overflow requirement"?

[nbenn]

What do you mean by "silent overflow requirement"?

With that I mean

In addition, DBI supports the bigint argument that governs how 64-bit integer data is returned. The following values are supported:

• "integer": always return as integer, silently overflow
• "numeric": always return as numeric, silently round
• ...

which I believe is explicitly enforced in connect_bigint_integer.

[krlmlr]

We could invent a new "integer-strict" setting here?

[nbenn]

@krlmlr yes, if some form of bigint = "integer" should be default, we'd need to at least warn the user when they try to pull data that does not fit. From my side this is ready to go. Warnings in case of overflow are the default anyways. I can always muffle the warnings for DBI compliance-sake. Would you want to look into adding this "strict" int mode to DBI/DBItest?

[krlmlr]

A proper error seems like the better way.

Happy to review a proposal for an "integer-strict" mode in adbi, to see which checks now fail, before introducing them to DBI/DBItest.