Add test_validate feature flag + CI check

[tustvold]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

For better or worse arrow-rs makes use of unsafe in various places. Whilst there are ongoing efforts to reduce the use of unsafe, occasionally bugs will slip through. Both arrow-rs and arrow-datafusion run with MIRI in order to catch many of these, however, MIRI is extremely slow and fiddly to setup which acts as a limiting factor on crowd-sourced test coverage. Additionally it will not catch the arguably more common types of logic bug, e.g. inconsistent null counts, non-consecutive array offsets, etc...

Describe the solution you'd like

Add a test_validate feature flag that when enabled will cause various "unchecked" methods to actually perform validation. The most obvious being ArrayDataBuilder::build_unchecked. This will allow projects to contribute to crowd-sourced nasal demon hunting, without imposing on them all the costs of setting up and running MIRI.

Describe alternatives you've considered

This could also be enabled based on debug_assertions being enabled, but this is a global setting that is on by default in debug builds. A feature flag gives people the ability to opt-out of the validation. FWIW this is what the parquet crate currently does.

[alamb]

"nasal demons" for anyone else who is curious http://www.catb.org/jargon/html/N/nasal-demons.html

[alamb]

Maybe force_validate might be a more descriptive name for this flag.

I actually did this (forced ArrayData::new_unchecked to actually check) to manually test the original validation logic

[jhorstmann]

Agree, this sounds useful. Making it independent of debug_assertions is also a good idea and would allow enabling it also in some staging/pre-production environments.

[tustvold]

Had a brief stab at adding this, the good news is it doesn't show any new problems, the bad news is it causes a lot of the tests of the validation logic to fail 😆 I need to have a think about how best to handle this...

[alamb]

Had a brief stab at adding this, the good news is it doesn't show any new problems, the bad news is it causes a lot of the tests of the validation logic to fail 😆 I need to have a think about how best to handle this...

Maybe we could just #cfg out the test suite for ArrayData 🤔