- Add Sprint 3 plan (37 steps, global 39-75) with canonical memory annotations, test quality requirements, global step numbering, and Sprint 2 overlap notes - Refactor all docs to use canonical annotation families (@Deallocate, @Lifetime, @Reclaim, @Owner, @Allocate) replacing simplified @deref 4-strategy system - Replace @perf with canonical @Hot/@Cold, @Inline, @Pure from annotations/6 optimization - Replace @memory-footprint, @execution-mode, @deref-explicit with canonical equivalents - Update REQUIREMENTS_OVERVIEW, SPRINT_1_REQUIREMENTS, SPRINT_2_PLAN, SPRINT_2_VISION, C++ Implementation Roadmap, example files, and progress report - Remove duplicate bonus steps 41-42, consolidate Phase 3h from 7 to 4 steps Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
18 KiB
Sprint 1: Python ↔ C++ Dual Projection
Sprint Goal
Enable a junior developer to write working logic in Python, a senior developer to optimize it in C++, and the junior developer to continue modifying safe areas while receiving warnings about optimization-locked regions.
Scope
| In Scope | Out of Scope |
|---|---|
| Python and C++ projections | Other languages |
| Tree-sitter parsing for ingestion | Custom parser implementation |
Memory strategy annotations (@Deallocate, @Lifetime, @Reclaim, @Owner, @Allocate) |
Full SemAnno schema |
| Warning system for locked nodes | Hard locks / approval workflows |
| Basic AST nodes (functions, loops, variables, expressions) | Advanced nodes (generics, macros, templates) |
| MPS projectional editor | IDE plugins (VS Code, etc.) |
User Stories
US-1: Python Developer Writes Logic
As a junior Python developer I want to write business logic in a Python-like projection So that I can focus on correctness without worrying about memory management
US-2: C++ Developer Optimizes
As a senior C++ developer I want to switch to a C++ projection and add optimization annotations So that the generated code is production-ready and performant
US-3: Python Developer Sees Warnings
As a junior Python developer I want to see warnings when I try to modify optimized nodes So that I understand the impact of my changes without being blocked
US-4: Import Existing Code
As a developer I want to import existing Python or C++ files into the AST So that I can work with legacy code in Whetstone
Technical Requirements
TR-1: Core AST Nodes
Define MPS concepts for a minimal but complete AST:
Module
├── Function
│ ├── name: string
│ ├── parameters: Parameter[]
│ ├── returnType: TypeReference
│ ├── body: Statement[]
│ └── annotations: SemAnno[]
├── Variable
│ ├── name: string
│ ├── type: TypeReference
│ ├── initializer: Expression?
│ └── annotations: SemAnno[]
└── annotations: SemAnno[]
Statement (abstract)
├── Assignment
├── IfStatement
├── WhileLoop
├── ForLoop
├── Return
├── ExpressionStatement
└── Block
Expression (abstract)
├── BinaryOperation
├── UnaryOperation
├── FunctionCall
├── VariableReference
├── Literal (int, float, string, bool)
├── ListLiteral
├── IndexAccess
└── MemberAccess
TR-2: Memory Strategy Annotations
The canonical memory annotation system (from annotations/Memory strategy.md) uses distinct annotation families:
DeallocateAnnotation — strategy: "Explicit" (manual free/delete)
LifetimeAnnotation — strategy: "RAII" (destructor-based cleanup)
ReclaimAnnotation — strategy: "Tracing"|"Cycle"|"Escape" (GC variants)
OwnerAnnotation — strategy: "Single"|"Shared_ARC" (ownership/ARC)
AllocateAnnotation — strategy: "Static"|"Register"|"Allocator" (allocation)
Common fields:
├── deallocationTime: Expression? (required if @Deallocate(Explicit))
├── deallocationLocation: string? (required if @Deallocate(Explicit))
└── owner: AgentReference?
Note: Sprint 1 implemented this as a single
DerefStrategyclass with astrategystring field. Sprint 3 refactors into the canonical types above. SeeSPRINT_3_PLAN.mdMigration Notes for the mapping.
TR-3: Optimization Lock Annotation
OptimizationLock
├── lockedBy: AgentReference
├── lockReason: string
├── lockLevel: enum {warning, soft, hard} // Sprint 1: warning only
├── affectedStrategies: string[]
└── timestamp: datetime
TR-4: Language-Specific Idiom Annotation
Placeholder annotations that preserve language-specific features during import without requiring full semantic understanding.
LangSpecific
├── language: enum {python, cpp, rust, ...}
├── idiomType: string // "decorator", "template", "attribute", "pragma", etc.
├── rawSyntax: string // The original syntax as written
├── semanticHint: string? // Optional: "memoization", "generic", "compile-hint"
└── position: enum {before, after, wrapping} // Where it attaches to the node
Examples:
// Python decorator captured during import
@lang_specific(python, "decorator", "@lru_cache(maxsize=128)", hint="memoization", position=before)
// C++ template captured during import
@lang_specific(cpp, "template", "template<typename T>", hint="generic", position=before)
// C++ pragma
@lang_specific(cpp, "pragma", "#pragma omp parallel for", hint="parallelization", position=before)
// Python type hint that has no C++ equivalent
@lang_specific(python, "type_hint", "-> Generator[int, None, None]", hint="generator", position=after)
// C++ attribute
@lang_specific(cpp, "attribute", "[[nodiscard]]", hint="return_value_check", position=before)
Projection Behavior:
| In Native Projection | In Foreign Projection |
|---|---|
| Rendered as original syntax | Rendered as comment with semantic hint |
Python projection of a C++ template function:
# cpp_template: template<typename T> (generic)
# NOTE: Python version uses dynamic typing instead
def compute(value):
...
C++ projection of a Python decorated function:
// python_decorator: @lru_cache(maxsize=128) (memoization)
// NOTE: Implement memoization manually or use std::map cache
int compute(int value) {
...
}
TR-5: Tree-sitter Integration
- Python grammar:
tree-sitter-python - C++ grammar:
tree-sitter-cpp - Map tree-sitter CST nodes to Whetstone AST concepts
- Preserve source locations for round-trip fidelity
TR-6: Generators
Python Generator:
- Emit idiomatic Python from AST
- Ignore deref strategies (Python is GC'd)
- Emit annotations as comments or type hints where applicable
C++ Generator:
- Emit idiomatic C++ from AST
- Translate deref strategies to appropriate constructs
- Inject required includes and memory management code
Memory Strategy Examples
The following examples show the same logical function with different memory annotations (from annotations/Memory strategy.md), demonstrating how the C++ projection changes while Python remains constant.
Example 1: Processing a List of Records
Whetstone AST (Canonical Form)
Function: processRecords
@{MemoryAnnotation} // Strategy applied here — one of the canonical types
Parameter: records -> List<Record>
Body:
ForLoop:
iterator: record in records
body:
Call: record.validate()
Call: record.transform()
Return: records
Python Projection (Same for all strategies)
def process_records(records: list[Record]) -> list[Record]:
"""
# @Reclaim(Tracing) — Annotation visible but doesn't affect Python code
"""
for record in records:
record.validate()
record.transform()
return records
C++ Projection: @Reclaim(Tracing) - Garbage Collected
// Memory managed by shared_ptr reference counting
#include <memory>
#include <vector>
std::vector<std::shared_ptr<Record>> process_records(
std::vector<std::shared_ptr<Record>> records
) {
for (auto& record : records) {
record->validate();
record->transform();
}
return records;
}
C++ Projection: @Owner(Single) - Ownership-Based (Rust-like)
// Ownership transferred, no copies, RAII cleanup
#include <vector>
#include <memory>
std::vector<std::unique_ptr<Record>> process_records(
std::vector<std::unique_ptr<Record>> records // Takes ownership
) {
for (auto& record : records) {
record->validate();
record->transform();
}
return records; // Transfers ownership to caller
}
C++ Projection: @Deallocate(Explicit) - Manual Control
// Developer explicitly controls allocation and deallocation
// @deallocation-time: end_of_function
// @deallocation-location: caller_responsibility
#include <vector>
std::vector<Record*> process_records(
std::vector<Record*> records, // Raw pointers - caller owns
size_t count
) {
for (size_t i = 0; i < count; ++i) {
records[i]->validate();
records[i]->transform();
}
return records;
// NOTE: No deallocation here - @deallocation-location specifies caller handles it
}
C++ Projection: @Allocate(Static) + @ConstExpr - Immutable
// Immutable data, identity by content hash, can be freely shared
#include <vector>
#include <functional>
struct ImmutableRecord {
const std::string data;
const size_t hash;
ImmutableRecord transform() const {
// Returns NEW record, original unchanged
return ImmutableRecord{transformed_data, new_hash};
}
};
std::vector<ImmutableRecord> process_records(
const std::vector<ImmutableRecord>& records // Immutable reference
) {
std::vector<ImmutableRecord> results;
results.reserve(records.size());
for (const auto& record : records) {
record.validate(); // Throws if invalid, doesn't mutate
results.push_back(record.transform()); // New record
}
return results;
}
Example 2: Building a Cache
Whetstone AST (Canonical Form)
Function: getOrCreate
@{MemoryAnnotation} // One of the canonical types
Parameter: cache -> Map<string, Widget>
Parameter: key -> string
Body:
IfStatement:
condition: key in cache
then: Return cache[key]
else:
Assignment: widget = Widget.create(key)
Assignment: cache[key] = widget
Return: widget
Python Projection
def get_or_create(cache: dict[str, Widget], key: str) -> Widget:
"""
# @Reclaim(Tracing) — Annotation visible but doesn't affect Python code
"""
if key in cache:
return cache[key]
else:
widget = Widget.create(key)
cache[key] = widget
return widget
C++ Projection: @Reclaim(Tracing)
std::shared_ptr<Widget> get_or_create(
std::unordered_map<std::string, std::shared_ptr<Widget>>& cache,
const std::string& key
) {
auto it = cache.find(key);
if (it != cache.end()) {
return it->second;
}
auto widget = std::make_shared<Widget>(Widget::create(key));
cache[key] = widget;
return widget;
}
C++ Projection: @Owner(Single)
// Note: Unique ownership makes caching tricky - must use reference
Widget& get_or_create(
std::unordered_map<std::string, std::unique_ptr<Widget>>& cache,
const std::string& key
) {
auto it = cache.find(key);
if (it != cache.end()) {
return *it->second;
}
auto [inserted_it, _] = cache.emplace(
key,
std::make_unique<Widget>(Widget::create(key))
);
return *inserted_it->second;
}
C++ Projection: @Deallocate(Explicit)
// @deallocation-time: cache_destruction
// @deallocation-location: CacheManager::cleanup()
Widget* get_or_create(
std::unordered_map<std::string, Widget*>& cache,
const std::string& key
) {
auto it = cache.find(key);
if (it != cache.end()) {
return it->second;
}
Widget* widget = new Widget(Widget::create(key)); // Manual allocation
cache[key] = widget;
return widget;
// Caller note: Deallocation handled by CacheManager::cleanup()
}
Warning System Specification
Warning Triggers
A warning appears when a node has an OptimizationLock annotation and the current user's tier is below the lock level.
Warning Display
┌─────────────────────────────────────────────────────────────┐
│ ⚠ OPTIMIZATION WARNING │
├─────────────────────────────────────────────────────────────┤
│ This node was optimized by: senior_dev_alice │
│ Optimization: @Deallocate(Explicit) with SIMD vectorization │
│ │
│ Modifying this code will: │
│ • Invalidate the manual memory management strategy │
│ • Disable the 4x SIMD optimization │
│ │
│ [Proceed Anyway] [View C++ Projection] [Cancel] │
└─────────────────────────────────────────────────────────────┘
Warning Metadata
When user proceeds despite warning:
- Original optimization is shadowed (preserved but inactive)
@provenanceupdated with modification chain- Notification queued for original optimizer
File Structure for Sprint 1
The SemAnno language is built incrementally across the sprint:
languages/
└── SemAnno/
├── models/
│ ├── SemAnno.structure.mps # Core AST + annotation concepts
│ ├── SemAnno.editor.mps # Python & C++ projections
│ ├── SemAnno.behavior.mps # Concept methods & helpers
│ ├── SemAnno.typesystem.mps # Type rules & inference
│ ├── SemAnno.textGen.mps # Python & C++ code generators
│ └── SemAnno.constraints.mps # Validation constraints
├── tests/
│ └── SemAnno.tests.mps # Language feature tests
└── SemAnno.mpl
Acceptance Criteria
AC-1: Round-Trip Parsing
- Parse Python file with tree-sitter → Whetstone AST
- Parse C++ file with tree-sitter → Whetstone AST
- Generate Python from AST that is functionally equivalent
- Generate C++ from AST that compiles and runs
AC-2: Memory Strategy Application
- Apply
@Reclaim(Tracing)to a function → C++ uses shared_ptr - Apply
@Owner(Single)or@Lifetime(RAII)to a function → C++ uses unique_ptr - Apply
@Deallocate(Explicit)to a function → C++ uses raw pointers - Python projection shows annotation as comment but code unchanged
AC-3: Warning System
- Add optimization lock to node
- Junior user attempts edit → warning appears
- Warning shows what will be invalidated
- User can proceed (warning only, not blocking)
- Provenance updated after edit
AC-4: Dual Projection Editing
- Open AST in Python projection → edit logic
- Switch to C++ projection → same logic, different syntax
- Edit in C++ projection → Python projection updates
- Annotations visible in both projections
AC-5: Language-Specific Idiom Preservation
- Import Python file with decorators →
@lang_specificannotations created - Import C++ file with templates →
@lang_specificannotations created - View Python decorator in C++ projection → shows as comment with hint
- View C++ template in Python projection → shows as comment with hint
- Re-export to original language → idiom syntax restored exactly
- Semantic hints populated for known patterns (lru_cache → "memoization")
Implementation Order
Build SemAnno incrementally, testing each phase before moving forward:
-
Core AST Structure
- Add core concepts to SemAnno.structure.mps (Module, Function, Variable, Statement, Expression, Type nodes)
- Create basic editors in SemAnno.editor.mps for each node type
- Manual AST creation works in MPS
- Test: Create a simple function with statements and expressions
-
Python Projection & Generator
- Extend SemAnno.editor.mps with Python-syntax projections
- Implement Python generator in SemAnno.textGen.mps
- Manual round-trip: type Python-like code → see generated .py file
- Test: Generate valid, runnable Python code
-
C++ Projection & Generator
- Extend SemAnno.editor.mps with C++ syntax projections
- Implement C++ generator in SemAnno.textGen.mps
- Add memory annotation translation to generator
- Test: Generate valid, compilable C++ code for each memory annotation type
-
Tree-sitter Import
- Add import behavior to SemAnno.behavior.mps
- Integrate tree-sitter-python for Python parsing
- Integrate tree-sitter-cpp for C++ parsing
- Test: Parse Python/C++ files → populate SemAnno AST
-
Warning System & Annotations
- Add OptimizationLock and memory strategy annotations to SemAnno.structure.mps
- Implement warning logic in SemAnno.behavior.mps
- Add warning UI to SemAnno.editor.mps
- Implement provenance tracking
- Test: Locked nodes show warnings, allow edits with provenance updates
Open Questions
-
Strategy inference: Should the system suggest memory annotations based on usage patterns, or always require explicit annotation?
-
Partial optimization: Can a senior optimize just one function while leaving others with default
@Reclaim(Tracing)? -
Conflict granularity: If a junior modifies a loop inside an optimized function, does that invalidate the whole function or just the loop?
-
Tree-sitter fidelity: How do we handle Python/C++ features that don't map cleanly to each other (e.g., Python decorators, C++ templates)?
Dependencies
| Dependency | Purpose | Source |
|---|---|---|
| JetBrains MPS 2023.2+ | Language workbench | jetbrains.com/mps |
| tree-sitter | Parsing | github.com/tree-sitter |
| tree-sitter-python | Python grammar | github.com/tree-sitter/tree-sitter-python |
| tree-sitter-cpp | C++ grammar | github.com/tree-sitter/tree-sitter-cpp |
Success Metrics
| Metric | Target |
|---|---|
| Parse success rate (Python) | >95% of valid Python files |
| Parse success rate (C++) | >90% of valid C++ files |
| Round-trip fidelity | Semantically equivalent output |
| Warning accuracy | 100% of locked nodes trigger warnings |
| Generator correctness | Generated C++ compiles without errors |