mirror of
https://github.com/hanzoai/datastore.git
synced 2026-09-26 13:55:07 +00:00
Raúl MarínandClaude Opus 4.8
bebe59d8b0
Defense-in-depth follow-up to the per-format depth fixes. `Poco::JSON::Parser` defaulted to unlimited nesting depth, and many sites parse untrusted JSON through it without calling `setDepth` - datalake metadata (Delta/Iceberg/ Paimon) and their REST/Unity catalogs, the Confluent/Avro schema registry, HTTP/cloud responses, and ~30 other sites across src/. Any of them could overflow the native stack on a deeply nested document. Rather than chase each call site, default `ParserImpl`'s depth to `JSON_DEFAULT_DEPTH` (1000) so every Poco JSON parse is bounded; callers that genuinely need more can opt in via `setDepth`. The enforcement itself was wired into Poco's `handleArray`/`handleObject` recursion in the previous commit; this just changes the default. 1000 is far above any legitimate JSON (real nesting is well under 100, and ClickHouse's JSON type uses simdjson/ rapidjson, not Poco), and part metadata (`serialization.json`) is a flat structure, so internal use is unaffected (verified by a nested-type part round-trip). This subsumes the explicit Delta/Iceberg/Paimon `setDepth` calls added earlier, which are removed here as redundant; the datalake regression test (04340) now passes via the default and additionally covers catalogs, schema registry, and the other Poco JSON parsers. Co-Authored-By: Claude Opus 4.8 <[email protected]>