Roadmap poll: what should we build next in django-orm-lens?

[FROWNINGdev]

tl;dr — django-orm-lens сегодня закрывает три ниши на одном static-analysis парсере: editor sidebar + ER diagram, nplusone / migration-risk CLI, MCP server for AI agents (Cursor / Aider / Claude Desktop / Zed). Больше нет tool'а, который делает всё это сразу.

But we can go further. Below is a shortlist of candidates for v0.8 – v1.0. Comment 👍 / 👎 / notes on each, propose your own, or roast the roadmap — everything is on the table.

Candidate features

🧭 Deep static analysis

• A. Query hint suggestions — walk .filter() / .values() calls, warn when .only(...) or .defer(...) would eliminate 90% of column IO. Runtime tools like django-silk measure this after the fact; we could catch it in the editor as you type.
• B. Missing index inference from queryset patterns — currently suggest_indexes only reads Meta.indexes + db_index=. Next step: analyse .filter(foo__gt=) / .order_by(...) chains across the repo and recommend composite indexes that would kill sequential scans.
• C. select_related / prefetch_related autocomplete inside .filter() / .annotate() — the top-voted feature request in #3. Would ship as a VS Code CompletionProvider driven by the same relation graph we already build.

🧪 CI / DX

• D. Pre-commit hook wrapper — one-line install: django-orm-lens precommit install writes a .pre-commit-config.yaml block that runs nplusone + migration-risk on staged files only. Zero-config.
• E. GitHub Action — a marketplace action that comments on a PR when it introduces N+1 patterns or risky migrations. Companion to D for cloud-only teams.
• F. --format sarif — output CLI findings as SARIF so GitHub Code Scanning + GitLab CI + JetBrains Qodana ingest us natively.

🎨 Editor UX

• G. Inline diagnostics for related_name typos — squiggle under Book.objects.filter(authro__name=). Would need light Django-project resolution but stays 100% static.
• H. Save ER diagram as PNG / SVG — currently Mermaid renders in the webview; export button would let people drop it into design docs and PRs.
• I. Multi-workspace diagram merge — for microservice repos where models live across 3-5 repos, merge them into one diagram.

🤖 MCP / AI agents

• J. Streaming describe_model responses — for models with 40+ fields the current single-shot response is over the token budget. Chunked would let agents keep context clean.
• K. explain_migration_plan MCP tool — given a set of pending migrations, produce a plain-English explanation the agent can quote back to the human before running.

🌐 Framework reach

• L. SQLAlchemy support — same parser, different regex/AST. Roughly 3-4 days of work. Would double the addressable audience.
• M. Pydantic model diagrams — data-classes have less relational structure but the "explore my types visually" surface is shared.

What we intentionally do NOT plan

• No runtime instrumentation (that is django-silk / django-debug-toolbar — they do it well; we stay static)
• No admin UI generator (out of scope)
• No paid tier (v0.8+ stays MIT, single binary, no telemetry)

How to vote

React on this post: 👍 = yes, ship; 👎 = meh. Reply with a letter + one line to add nuance: "C. yes but only for .filter(), not .annotate()".

If your favorite thing isn't listed — propose it. Every idea in Ideas with 3+ 👍 gets a labelled issue and lands on the Roadmap.

Also — if you tried the tool and it broke, file a bug or reply here. We shipped 8 patches yesterday alone off a single Reddit post; we can turn one from here around in a day.