Plan: Reimagine profile data structure v2

[Copilot]

Design plan for a v2 data structure addressing limitations of the current format: pre-aggregation losing individual samples, nested list columns breaking relational semantics, no memory profiling support, and no provenance tracking.

Proposed v2 schema (data-raw/plan-v2.md)

• sources — new table for provenance (type, URI, timestamp), enabling multi-run combination
• samples — one row per raw sample (no aggregation), linked to source
• sample_values — long-form measurements replacing sample_types + samples$value; extensible to memory profiling without schema changes
• sample_locations — bridge table with depth column, replacing nested list column samples$locations
• locations, functions — unchanged
• dm as primary container with defined PK/FK constraints

Migration

profile_v2_from_v1() expands aggregated samples, unnests locations into the bridge table, infers source type from hidden .rprof/.msg components.

Open questions

Includes decisions needed on dm as Imports vs Suggests, class hierarchy, value column type, and depth indexing convention.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.