Should datum_t() == datum_t() crash or not?

[mlucy]

Opinions?

[timmaxw]

I'd say "no". A related question is whether datum_t() < datum_t::null() should crash.

[danielmewes]

I think we should probably allow all comparisons with uninitialized datums, including < != and ==.
I don't think it matters where they sort though.

Additionally we can consider adding an assertion (but not a guarantee) that uninitialized datum_ts don't get < compared.

[mlucy]

I'm a little torn on this; datum_t() is basically replacing counted_t<datum_t>() from before, so in my mind datum_t() < datum_t::null() is equivalent to *counted_t<datum_t>() == *datum_t::null(), which should clearly crash. (Also, in general, it might make sense to force people to explicitly handle the uninitialized case rather than using uninitialized data like normal data.)

[mlucy]

Also also, the current system has the pleasant property that if uninitialized data slips through into the normal ReQL code you get an error immediately rather than it being treated normally until something tries to send it to the user or whatever.

[mlucy]

Moving to polish because we obviously wouldn't hold the release for an internal issue like this. Not sure why I added it to 1.16 initially.