C10: ModuleInfo.drop_ast() for memory hygiene #25

New issue

Open

opened 2026-05-05 02:01:16 +02:00 by claude · 0 comments

claude commented

2026-05-05 02:01:16 +02:00

Collaborator

Ticket C10

F-06: FunctionInfo.ast_node retained after analysis. For 100k LOC repos + long-running MCP server (post-B6 caching), memory accumulates. Add drop_ast() called by default in analyze().

Source: F-06 in 2026-05-03 audit.

Detailed plan

The full ticket brief — TDD plan, current state with file:line references, target state, acceptance criteria, edge cases, hostile reviewer questions, out-of-scope — lives at .codex/MASTER/PHASE-C/C10-drop-ast.md in the operator's local working tree (the .codex/ directory is gitignored, so this content is intentionally not in repo history; it's pre-execution planning material that gets updated as evidence comes in).

When picking up this issue, read the local file first. If you're an external contributor without access to the operator's .codex/, comment on this issue and the operator will paste the brief.

Phase context

This is part of Phase C (post-Phase-A) work. Execution is paused until dogfood evidence accumulates from real-world pyfallow integration in operator's other repos (window: 2026-05-04 → ~2026-06-15).

After the window closes, dogfood evidence drives ticket prioritization:

Tickets validated by evidence stay (and may be refined)
Tickets contradicted or made obsolete get closed with rationale
New issues with dogfood:* labels may pre-empt Phase C priority

See docs/dogfood.md and docs/dogfood-log-template.md for the protocol.

Labels applied

phase:c — phase grouping
severity:low — severity (audit-derived)
area:engineering — area of impact

— Migrated from .codex/MASTER/PHASE-C/ by Claude Opus 4.7 (orchestrator) on 2026-05-05 per operator request to surface Phase B/C work in Forgejo Issues so it doesn't disappear with session context.

## Ticket C10 F-06: FunctionInfo.ast_node retained after analysis. For 100k LOC repos + long-running MCP server (post-B6 caching), memory accumulates. Add drop_ast() called by default in analyze(). **Source:** F-06 in 2026-05-03 audit. ## Detailed plan The full ticket brief — TDD plan, current state with file:line references, target state, acceptance criteria, edge cases, hostile reviewer questions, out-of-scope — lives at `.codex/MASTER/PHASE-C/C10-drop-ast.md` in the operator's local working tree (the `.codex/` directory is gitignored, so this content is intentionally not in repo history; it's pre-execution planning material that gets updated as evidence comes in). When picking up this issue, read the local file first. If you're an external contributor without access to the operator's `.codex/`, comment on this issue and the operator will paste the brief. ## Phase context This is part of **Phase C** (post-Phase-A) work. Execution is **paused** until dogfood evidence accumulates from real-world pyfallow integration in operator's other repos (window: 2026-05-04 → ~2026-06-15). After the window closes, dogfood evidence drives ticket prioritization: - Tickets validated by evidence stay (and may be refined) - Tickets contradicted or made obsolete get closed with rationale - New issues with `dogfood:*` labels may pre-empt Phase C priority See `docs/dogfood.md` and `docs/dogfood-log-template.md` for the protocol. ## Labels applied - `phase:c` — phase grouping - `severity:low` — severity (audit-derived) - `area:engineering` — area of impact — Migrated from `.codex/MASTER/PHASE-C/` by Claude Opus 4.7 (orchestrator) on 2026-05-05 per operator request to surface Phase B/C work in Forgejo Issues so it doesn't disappear with session context.

claude added the

severity:low

area:engineering

phase:c

labels

2026-05-05 02:01:16 +02:00

claude added this to the Phase C — Show HN polish milestone

2026-05-05 02:08:07 +02:00