Skip to content

Column and File Metadata RFC#62

Draft
mprammer wants to merge 1 commit into
developfrom
mp/rfc-column-file-metadata
Draft

Column and File Metadata RFC#62
mprammer wants to merge 1 commit into
developfrom
mp/rfc-column-file-metadata

Conversation

@mprammer

@mprammer mprammer commented Jul 6, 2026

Copy link
Copy Markdown

Proposes RFC 0062: Column and File Metadata — generic, user-defined metadata at two granularities:

  • Per-column, serialized co-located on each DType node but held out of the in-memory DType, so Eq/Hash and every compute/scan/chunk path are unaffected.
  • File-level, a postscript-referenced footer segment fetched on demand.

Both are an opaque map<string, [u8]>. The change is additive with no file-format version bump, and it unblocks three independent needs: Iceberg field IDs / schema evolution, lossless Arrow Field metadata round-trip, and file-level sidecar metadata. iceberg.id rides as a convention, modeled on ORC.

Opening as a draft for discussion. The design is grounded against the current dtype.fbs, footer.fbs, and the Arrow bridge; open questions (in-memory carrier shape, file-level segment granularity, Arrow non-UTF-8 export, reserved namespaces) are called out in the RFC.

🤖 Generated with Claude Code

Add RFC 0062 proposing generic per-column and file-level metadata (opaque map<string,[u8]>) for Iceberg field IDs, Arrow field-metadata round-trip, and file-level sidecar metadata.

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: mprammer <martin@spiraldb.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant