@joshbressers
Today my company #wontfix-ed a bug report for a proprietary library similar to the hypothetical in your OSS podcast 456 "What if XZ happened..."
In our case, AWS Aurora (proprietary postgres wrapper) skips rows containing valid double precision float (IEEE 754) that it thinks are too small (subnormal) or improperly serialized (trailing zeros in scientific notation), despite #postgres handling these quirky values just fine. Imagine skipped rows in an RBAC table...
@kurtseifried
@hobs @joshbressers @kurtseifried
Does this happen silently?
@SpaceLifeForm
Fortunately not. It spits out the row numbers and field name for the IEEE 754 values that are incompatible with its Fractional data type (it's equivalent of a double).
@joshbressers @kurtseifried
@SpaceLifeForm
We're using it in a SageMaker ML "FeatureStore", so I don't really know where the bug might be fixed within the tangle of proprietary libraries and dark patterns that is AWS...
@joshbressers @kurtseifried
@SpaceLifeForm
It's very difficult to reproduce because the quirky values very rarely occur in the real world.
@joshbressers @kurtseifried
@hobs @joshbressers @kurtseifried
If I am understanding this correctly, it does not fail fast.
So, there must be some kind of warning/error returned that must be checked for. And then review of logs.
Not an ideal approach in my book.
@hobs @SpaceLifeForm @kurtseifried youch
This lines up with how I’ve seen commercial entities deal with weird bugs. It’s probably expensive to fix and cheap to ignore
I do think it’s something open source can do better than closed. It’s probably hard to fix and barely affects anyone, but if it affects the right nerd, they’ll get it done because it’s interesting