A while ago I created the issue #14424 
(https://github.com/apache/iceberg/issues/14424) regarding the hard-coded 
default database name in the SessionCatalog, and its improper documentation.

Initially, I had feared that people might be relying on this bug, so I wanted 
to raise that issue here. I had a similar issue once in the Python standard 
library, and such technically-correct-but-might-break-stuff changes required an 
approval from the steering council. After digging a bit deeper, I think I'm 
being overly cautious here, but I can't statically prove that this is not a 
breaking change since the the defaultNamespace() function has a very wide 
transitive call tree.

Another thing that I had proposed in my original PR is that we deprecate the 
"default-namespace" config key and align it with Spark, Hudi and Delta by using 
"defaultDatabase". This is now a stacked PR #18222 on top of #17029, which just 
focuses on fixing the hard-coded default namespace in SessionCatalog.

The changes here are extremely simple, but the lines of code have balooned due 
to new tests and having to copy stuff into 4 Spark folders. I had discussed 
this with Manu Zhang but we would appreciate a fresh pair of eyes on this. You 
can accept just the first, the second, or none of the PRs.

Cheers,
Dzeri

Attachment: publickey - [email protected] - 0x5E7E90EC.asc
Description: application/pgp-keys

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to