Make Polaris safe against certain unparseable locations

[snazy]

Using java.net.URI for paths/locations generated/used by Apache Iceberg is particularly unsafe.

Characters that are illegal in URI string representations are not escaped by Iceberg and lead to runtime exceptions when parsed for example by java.net.URI.create(String).

Example: A legit column/partition-field name "_foo_bar_" will end "as is" in any path generated by Iceberg, leading to Caused by: java.lang.IllegalArgumentException: Illegal character in path at index ....

The safest approach to prevent this issue entirely is to have a dedicated class that deals with "unsafe encodings", maybe call it "StorageUri".

Mixing "unsafe" and "safe" encodings will cause errors.

[flyrain]

Hi @snazy , can you provide more details of which part of Polaris code has this issue?

[flyrain]

Related PR: #1586

[dimas-b]

More specific issue: #1545

PR #1586 may not be sufficient for this issue. If it is merged with its current scope more end-to-end testing will still be required with odd base URI to ensure correct operation. Before #1586, end-to-end testing is not possible.