STRICT_MODE_CONVERSION_PLAN.md

AI Native Lang (AINL) Strict-Mode Conversion Plan

Target: Convert all AINL programs to strict_mode=True + strict_reachability=True compliance. Based on analysis date: 2026-03-03 (compiler_v2 with grammar_constraint updates) Status: Grammar stable; ready for execution.

---

Summary of Breaking Changes

1. If conditions must use precomputed booleans

   • ❌ If (core.gt x y) ->L_then ->L_else
   • ✅ X _cond (core.gt x y); If _cond ->L_then ->L_else

2. Adapter calls must use exactly two tokens: R group verb

   • ❌ R core.now, R email.G, R cache.get, R queue.Put, R db.F, R svc caddy (inside R it's two tokens? Actually R svc caddy is correct; but combined like cache.get as one token must be split.)
   • ✅ R core now, R email G, R cache get, R queue Put, R db F, R svc caddy
   • Note: The grammar expects adapter = group name and target = verb name. Do not combine group and verb with a dot in the R op. The combined forms cause IR 'adapter': '?' and runtime gate errors.

3. Parentheses disallowed in many expression contexts

   • ❌ (core.concat a b) in argument position
   • ✅ Either core.concat a b directly, or X tmp (core.concat a b); use tmp
   • Applies to nested expressions in Filter, Sort, arithmetic, etc.

4. Strict reachability

   • All variables must be defined on all control‑flow paths before use.
   • Add default initializations: Set var default early (often at label 0) to satisfy all reads.
   • Ensure every J target is reachable and contains exactly one J as its entry point.

5. Label naming

   • Strict mode expects numeric label IDs only (L0, L1, …). Friendly names (L_skip, L_healthy) are not permitted.
   • Convert friendly labels to numeric and update all references consistently.

---

Conversion Order

Priority 1: demo/monitor_system.strict.lang (strict version of the production monitor) Priority 2: examples/openclaw/daily_digest.strict.lang Priority 3: examples/openclaw/lead_enrichment.strict.lang Priority 4: examples/openclaw/infrastructure_watchdog.strict.lang Priority 5: examples/openclaw/webhook_handler.strict.lang

Note: The non‑strict files (demo/monitor_system.lang and examples/openclaw/*.lang) are already syntactically correct (two‑token R steps) and run in production. Strict variants need the additional strict-mode transforms.

---

Step‑by‑Step Conversion Protocol

For each strict‑target file (*.strict.lang):

1. Parse with strict mode to generate full error list: AICodeCompiler(strict_mode=True, strict_reachability=True).compile(code).
2. Split any combined R tokens: convert R group.verb into R group verb. Also ensure any R wasm.CALL becomes R wasm CALL.
3. Extract all inline If conditions into X precomputations: X <var> (expr); If <var> ->L_then ->L_else.
4. Remove parentheses in expression arguments: wherever a nested call like (core.concat ...) appears, either flatten or bind to an intermediate X.
5. Add missing variable initializations: for variables read on some paths (e.g., needs_notify, failed_services, cond vars), add Set var default near the start of the graph (usually label 0).
6. Rename friendly labels to numeric: replace L_skip, L_healthy, L_done, L_notify, etc. with L<n> and update all jumps. Keep a mapping.
7. Re‑run strict validation until zero errors.
8. Verify the non‑strict file still compiles (optional sanity check).

---

Adapter Verb Reference (from ADAPTER_REGISTRY.json)

Adapter	Verbs	Effect	Strict‑mode example
core	now, add, sub, mul, div, idiv, mod, pow, len, concat, join, stringify, lt, gt, lte, gte, eq, ne	pure‑compute	R core now; expressions: core.sub x y
email	G	io‑read	R email G ->emails
calendar	G	io‑read	R calendar G ->events
social	G	io‑read	R social G ->mentions
db	F (read), P (write)	io‑read / io‑write	R db F Leads * ->leads; R db P Leads data ->result
svc	caddy, cloudflared, maddy	io‑read	R svc caddy ->status
cache	get, set	io‑read / io‑write	R cache get "key" ->val; R cache set "key" val ->_
queue	Put	io‑write	R queue Put notify payload ->_
wasm	CALL	pure‑compute	R wasm CALL module func args ->result

Important: In strict mode, the R op requires separate tokens for group and verb. Do not use dot‑notation in the R op.

---

Conversion Examples

If expression with precompute

Before:

L7: If (core.gt email_count Config.email_threshold) ->L8 ->L9

After:

L7: X _cond (core.gt email_count Config.email_threshold)
    If _cond ->L8 ->L9

Adapter token splitting

Before (strict draft):

L2: R email.G ->emails
L5: R cache.get "state" "last_email_check" ->last_check
L57: R queue.Put notify {...} ->_

After:

L2: R email G ->emails
L5: R cache get "state" "last_email_check" ->last_check
L57: R queue Put notify {...} ->_

Numeric labels

Before:

L34: If (core.gt stalled_count 0) ->L_skip ->L35
L_skip: Set skipped 1 J L36

After:

L0: Set skipped 0   # default
...
L34: X cond (core.gt stalled_count 0)
    If cond ->L50 ->L35
L50: Set skipped 1 J L36

Parentheses removal

Before:

X digest (core.join "\n\n" [
    (core.concat "📧 " (core.stringify email_count))
])

After:

X part1 (core.concat "📧 " (core.stringify email_count))
X part2 (core.concat "📅 " (core.stringify cal_count))
X digest (core.join "\n\n" [part1 part2])

---

Specific Files & Action Items

demo/monitor_system.strict.lang

• Split all R ops (core.now, email.G, calendar.G, social.G, cache.get/set, db.F, queue.Put, svc.* are already two‑token? Monitor non‑strict uses two‑token; the strict draft currently uses combined forms. Update to two‑token.)
• Precompute all If conditions with X.
• Add default Set needs_notify 0 and Set failed_services "" at entry (label 0) to satisfy reachability.
• Rename any friendly labels (if present) to numeric.
• Ensure every target label contains exactly one J.
• Verify bulk‑email filter doesn’t use parentheses around core.sub inside Filter; either flatten or bind intermediate.

examples/openclaw/daily_digest.strict.lang

• Convert R email G, calendar G, social.G → R social G (some already two‑token? but watch for combined).
• Replace cache.F usage with cache.get or cache.set explicitly.
• Extract If in L3 (throttle check) with X; convert friendly L_skip to numeric.
• Remove parentheses around any core.concat inside arrays.
• Add required initializations (e.g., summary = "") if needed.

examples/openclaw/lead_enrichment.strict.lang

• Fix If (core.ge i total_leads) ->Label to precomputed X.
• db.F is correct verb (F); ensure two‑token: R db F.
• social.F → R social G.
• cache.F → R cache get / set as appropriate.
• queue.F → R queue Put.
• Numeric labels.

examples/openclaw/infrastructure_watchdog.strict.lang

• Replace friendly labels (L_healthy, L_notify, L_send, L_skip) with numeric.
• Transform all If (expr) to X.
• queue.PUT → R queue Put.
• cache.F → R cache get.
• svc caddy etc. already two‑token? Should be R svc caddy (correct). Ensure no dot syntax.
• Ensure J targets singletons.

examples/openclaw/webhook_handler.strict.lang

• Check endpoint return flow; ensure all paths lead to valid response label.
• db.F and db.P already two‑token if written R db F / R db P; if written as db.F, split.
• cache.F → cache.get/set.
• queue.PUT → R queue Put.
• Numeric labels and single J per target.

---

Deliverables After Conversion

• Every .strict.lang file compiles in strict mode with zero errors.
• scripts/run_test_profiles.py --profile openclaw passes for both strict and non‑strict variants (if strict variant included in profile).
• agent_reports/AI_CONSULTANT_REPORT_APOLLO.md updated to reflect strict‑mode compliance and include strict‑mode example snippets.
• OPENCLAW_AI_AGENT.md references updated if necessary.

---

Timing

• Execute conversions in priority order; commit after each file’s strict validation.
• Re-run tests after each conversion.
• Keep non‑strict files unchanged; place strict variants alongside with .strict.lang suffix.

---

Prepared by: Apollo Date: 2026‑03‑03 Context: Two‑token R syntax required by compiler; non‑strict monitor already updated and running