The Binder and related types have been refactored to use the new
`NodeKind` enum instead of the previous `BoundKind`. This commit updates
all references to use the new structure, ensuring consistency across the
compiler's AST representation.
Key changes include:
- Replacing `BoundKind` with `NodeKind` in the Binder's `bind` and
`visit` methods.
- Updating pattern matching and field access to reflect the new enum
variants (e.g., `NodeKind::Def` instead of `BoundKind::Define`).
- Adjusting identifier bindings to use the new `IdentifierBinding` enum.
- Reflecting changes in `DefBinding`, `AssignBinding`, and
`LambdaBinding` structures.
- Ensuring all newly created nodes use `NodeKind` and the appropriate
metadata.
This commit reduces unnecessary cloning for fields like `addr`, `kind`,
and `field` within `BoundKind` and `VM::GetField`. This improves
performance by avoiding redundant memory allocations.
Introduce generic `CompilerPhase` trait to unify different stages of the
bound AST. Rename `LocalSlot` to `VirtualId` and introduce `StackOffset`
for clearer distinction between compile-time and run-time addressing.
Update type aliases and implementations to reflect these changes.
The VM's field access logic has been refactored to clone `Rc` pointers
for addresses and field identifiers. This change aims to improve
efficiency by avoiding unnecessary dereferencing and copying of values
when accessing fields within records and objects.
Additionally, the `Value::Object` handling in `BoundKind::GetField` has
been refined to more explicitly manage `RecordSeries` and `StreamNode`
types, ensuring that field access is handled polymorphically and
efficiently without copying underlying data structures.
This commit makes several changes related to how closures are handled:
- **Dumper:** Removes redundant introspection of closure ASTs, as this
is no longer relevant.
- **Optimizer:** Updates `try_inline` to accept `ExecNode` for
parameters and adds a check to ensure the `function_node` is a
`Lambda` before attempting inlining.
- **Environment:** Simplifies closure creation by passing an `ExecNode`
for parameters and ensuring the correct `stack_size` is used.
- **VM:** Adjusts `Closure` to store an `ExecNode` for parameters and
updates `VM::unpack` to accept an `ExecNode`.
Simplified the AST architecture by removing the overly complex Node<K,
T> structure and replacing it with specialized types for
each phase:
- Introduced UntypedNode in nodes.rs for the Parser and Macro system,
eliminating redundant ty: () initializations.
- Defined a new generic Node<T> in bound_nodes.rs for all compiler
stages, where T represents the phase-specific metadata.
- Updated the Binder to act as the bridge between UntypedNode and
Node<()>.
- Simplified type aliases: TypedNode, AnalyzedNode, and ExecNode now
leverage the streamlined Node<T> structure.
- Updated Parser, Macros, Type-Checker, Optimizer, and VM to reflect
the architectural changes.
- Fixed import chains and resolved needless borrows via clippy.
This change improves type safety, reduces boilerplate code in the
parser, and makes the compiler stages more idiomatic and
readable.
The TCO (Tail Call Optimization) module has been renamed to `lowering`.
This change better reflects the module's broader responsibility, which
includes not only TCO but also general AST transformations and
preparation for VM execution.
The `optimize` function has been renamed to `lower` to align with the
module's new name.
The `calculate_stack_size` method was duplicated in `bound_nodes.rs` and
`tco.rs`. This commit extracts it into a single standalone function in
`tco.rs` to avoid duplication. The stack size is now calculated and
stored in the `ExecNode`'s `ty` field during the TCO optimization phase.
Replaces manual iteration for pushing `Value::Void` with `Vec::resize`
for more efficient stack management. This change improves performance by
reducing redundant operations and simplifies the code.
Store the pre-calculated stack size of lambda bodies in the
RuntimeMetadata. This allows the VM to directly access this information
without recalculating it.
This commit introduces a new mechanism for calculating the required
stack size for closures at bind time, rather than storing it in the AST.
This is achieved by adding a `calculate_stack_size` method to
`BoundNode`.
Key changes include:
- `BoundNode::calculate_stack_size`: Recursively traverses the bound AST
to find the maximum `LocalSlot` index used.
- Recursion blocker for lambdas: Ensures that nested lambdas are not
visited during stack size calculation for the outer closure,
preventing incorrect stack size estimations.
- VM update: The `VM::run` method now accepts and uses the
pre-calculated stack size for setting up call frames.
- `Closure` struct update: Stores the `stack_size` directly within the
`Closure` struct.
- Binder and TypeChecker updates: Minor adjustments to accommodate the
new strategy and error reporting.
- Optimizer enhancements: Constant and pure-lambda propagation for
`Define` statements within blocks to ensure inlinability.
This refactoring addresses issues related to accurate stack size
management and improves the overall robustness of the binder and VM.
The `PipelineNode` has been refactored into `StreamNode`. This commit
updates the example files to reflect this change and also updates the
`series` constructor to accept a lookback parameter.
The `PipelineNode` was an artifact of a previous implementation and is
no longer needed. The `StreamNode` is the appropriate type for
representing the output of a pipe operation.
The `series` function now requires a `lookback` argument to specify the
maximum number of items to store. This makes the behavior of series more
explicit and prevents accidental unbounded memory growth.
The type checker now correctly handles unwrapping `Stream(T)` types for
lambda arguments.
A new `build_map_stream` function has been added to `rtl/streams.rs` to
facilitate field access on streams.
The `create-random-ohlc` function now returns a `Stream` instead of a
`Series`.
Examples have been updated to reflect these changes and demonstrate
stream usage.
The type checker's fallback logic for non-lambda nodes was unnecessarily
complex. This commit simplifies it by directly checking the node within
a new, basic type context.
Additionally, the environment now explicitly wraps non-lambda AST nodes
in a lambda before type checking, ensuring consistent handling. This
aligns the type checking process for all top-level expressions.
The `resolve_tail_calls` function is being called twice in some
execution paths, which is unnecessary and can lead to incorrect
behavior. This commit removes the redundant calls and simplifies the
execution flow.
The tail call resolution logic was duplicated in `Environment::call` and
`VM::run`. This commit extracts the tail call resolution logic into a
single method `VM::resolve_tail_calls` and uses it in both places.
Additionally, this commit adds support for series indexing as a form of
tail call, allowing for direct access to series elements through the
`series(index)` syntax. This is useful for back-referencing in
time-series data.
A new example `err.myc` is added to demonstrate basic series usage and
error handling.
The specializer now correctly identifies when a compiled value is a
scalar and avoids folding it into a constant node. Instead, it updates
the callee to the specialized version if it's an object. This ensures
that scalar folding does not interfere with the program's execution.
Introduce new indicators for Simple Moving Average (SMA) and Weighted
Moving Average (WMA), along with their Hull Moving Average (HMA)
derivative.
Also refactors series implementation to use `RefCell` for interior
mutability and adds `SeriesMember` trait for better dynamic series
access. Updates `soa_series.myc` example to reflect new series creation
and push syntax.
This commit refactors various AST node types to use `Rc` (Reference
Counting) instead of `Box`. This change is primarily driven by the need
for shared ownership and more efficient handling of potentially
recursive or shared data structures within the AST.
The main benefits of this change include:
- **Reduced cloning:** `Rc` allows multiple parts of the AST to refer to
the same node without needing to clone the entire node. This is
particularly beneficial for optimization passes where nodes might be
visited multiple times.
- **Enabling sharing:** It makes it easier to represent scenarios where
a single AST node might be a child of multiple parent nodes.
- **Performance improvements:** By avoiding unnecessary deep copies of
AST nodes, this change can lead to performance gains, especially
during complex compilation phases.
The Binder and its helper functions have been updated to accept and
propagate a `Diagnostics` struct. This allows for better error reporting
during the binding phase, enabling the compiler to continue processing
even after encountering errors and collect all issues before halting.
The `BoundKind::Error` node and `StaticType::Error` are introduced as
"poison" nodes/types. These nodes indicate that an error occurred during
compilation, preventing further valid processing of that specific AST
fragment but allowing the compiler to continue with other parts of the
code.
The `Parser` has also been updated to return a `Diagnostics` struct,
enabling it to report errors during the initial parsing stage while
still attempting to build a partial AST. This adheres to the same error
recovery strategy.
Introduces type specialization for reactive pipeline nodes and their
data buffers. This eliminates the overhead of generic `Value` enums and
`SharedValueSeries` when dealing with known scalar or record types.
The architecture shifts buffer instantiation from the VM to the runtime
(RTL), leveraging type information from the AST. A new `out_type` field
is added to `BoundKind::Pipe` to store the static type of the pipeline's
output, determined by the type checker.
This enables the creation of specialized `RingBuffer<T>` and
`SharedRecordSeries` (for Struct-of-Arrays layout) when the output type
is known, significantly improving memory usage and processing speed for
time-series data, especially in financial analysis.
Adds support for pipelines, allowing streams to be chained together and
transformed.
This includes new types for `PipeStream` and `SourceAdapter`, along with
modifications
to the `Environment` to manage pipeline execution. A new example
`pipeline.myc`
demonstrates its usage.
Introduces `SharedValueSeries` for read-only access to generic `Value`
buffers, and `PipelineNode` which combines an `ObservableStream` with a
`Series` view. This is a foundational step for implementing the `pipe`
operator.
Introduces a new `Pipe` node to the Abstract Syntax Tree (AST).
This node represents a functional composition pattern, allowing
for chaining of operations.
The changes include:
- Defining the `Pipe` variant in `BoundKind` and `UntypedKind`.
- Implementing parsing logic for the `pipe` keyword.
- Adding handling for the `Pipe` node in various compiler passes
(analyzer, binder, captures, dumper, macros, optimizer, TCO,
type checker).
- Including a placeholder implementation for `Pipe` execution in the
VM.
- Updating integration tests to use the new `pipe` syntax.
This commit introduces the `Series` trait to provide a unified interface
for accessing indexed data across different series types (RecordSeries,
SeriesView, and future SharedSeries). This simplifies the VM's indexer
logic and lays the groundwork for the dual series architecture.
This commit introduces the `get_record` method to `RecordSeries`,
allowing for the dynamic reconstruction of a full record from its
constituent fields at a given index. This enables idiomatic access to
series data using indexing like `(my_ticks 0)`.
The `push-series` function has also been updated to correctly handle the
structure of records when pushing data into a `RecordSeries`.
Introduces a new RTL module for handling series data, enabling efficient
storage and retrieval of time-series data in a Struct-of-Arrays (SoA)
format.
Includes `SeriesView`, a zero-copy wrapper that provides a fast,
read-only interface to individual columns within a `RecordSeries`,
optimizing performance for data access in the VM.
Registers new native functions `create-series` and `push-series` for
managing series data.
This commit refactors the internal representation of `BoundKind::Record`
to store a `RecordLayout` and a `Vec` of values, rather than a `Vec` of
key-value pairs. This change simplifies the representation and improves
efficiency by decoupling the record's structure from its specific values
during compilation and analysis.
The `RecordLayout` now defines the structure of the record, and the
values are stored in a separate vector, ordered according to the layout.
This allows for better optimization and type checking, as the record's
shape is explicitly defined and immutable once created.
The `CallFrame` struct and various VM methods now use `Rc<dyn Object>`
to hold closures, allowing for more flexibility and avoiding unnecessary
cloning.
This change also addresses the performance concern regarding deep copies
of closures, preferring `Rc<dyn Trait>` with local `downcast_ref` where
appropriate.
Additionally, a documentation note has been added to `gemini.md`
regarding this performance preference in Rust.
Introduces a new `RecordLayout` system for efficient and type-safe
record handling.
Key changes:
- `RecordLayout` struct with `O(1)` field lookup using FMap
optimization.
- Field accessors (`.name`) are now first-class callable values.
- Parser recognizes dot-prefixed identifiers as field accessors.
- Binder and TypeChecker handle field accessors.
- Optimizer transforms field accessor calls into specialized `GetField`
nodes.
- VM executes `GetField` efficiently by directly accessing record
values.
- Adds a new `records.myc` example showcasing record features.
- Improves comparison logic for records to use pointer equality for
layout.
This commit introduces newtype wrappers for `LocalSlot`, `UpvalueIdx`,
and `GlobalIdx` to improve type safety and clarity. These wrappers
replace direct use of `u32` for indices, making the code more robust and
easier to understand.
The changes include:
- Defining `LocalSlot`, `UpvalueIdx`, and `GlobalIdx` structs.
- Implementing `Display` for these new types to provide user-friendly
output.
- Updating the `Address` enum to use these new types.
- Modifying various compiler components (analyzer, binder, optimizer,
type checker, environment, VM) to use the new types.
- Adjusting tests to accommodate the changes.
This commit consolidates `DefLocal` and `DefGlobal` into a single
`Define` bound kind. This simplifies the AST and makes it more
consistent.
It also introduces `DeclarationKind` to differentiate between variable
and parameter definitions.
The `flatten_tuple` function was unnecessarily recursive. It can now
simply return the elements of the tuple directly. The VM's destructuring
logic has been updated to handle nested destructuring more efficiently.
A new integration test case for multi-level destructuring has been
added.
Renames `DefDestructure` to `Destructure` to better reflect its use in
both definitions and assignments.
Introduces `bind_assign_pattern` to handle assignment destructuring in
the binder.
Adds `test_assign_destructuring` to verify assignment destructuring
functionality.
This commit introduces the `DefDestructure` bound kind and modifies the
binder, analyzer, type checker, and VM to support destructuring in `def`
statements. This allows for pattern matching on the right-hand side of a
`def` to bind multiple variables.
The parser has been updated to accept patterns in `def` statements. The
binder now handles `UntypedKind::Def` with a `target` pattern, rather
than a simple `name`. This enables destructuring.
The `Gemini.md` documentation has been updated to include a new rule for
incremental development.
The `again` keyword is introduced to facilitate explicit recursive
function calls.
It is restricted to tail-call positions to prevent dead code and ensure
TCO
optimization. Type checking is enhanced to validate argument types
against
function parameters.
The `dispatch_eval` macro has been replaced with explicit
`eval_internal` and `eval_core` methods. This change aims to streamline
the evaluation process and better support the observer pattern by
providing clearer hooks for observing VM execution.
The Analyzer has been refactored to decorate `TypedNode`s with their
purity and recursion status. This involves creating a new `AnalyzedNode`
type and a `NodeMetrics` struct to hold this information. The `Analyzer`
now returns an `AnalyzedNode` instead of a separate `Analysis` struct.
This change lays the groundwork for future optimizations and analysis
passes.
Introduces `register_native` for direct registration of `NativeFunction`
and `register_native_fn` for convenience from closures. The
`Environment::run`
method is removed, and its functionality is now handled by
`Environment::instantiate`,
which packages the linked AST into an invokable `NativeFunction`. This
streamlines
the execution path and better separates compilation/linking from runtime
execution.
Move `Purity` enum definition from `optimizer.rs` to `types.rs` and
create a `NativeFunction` struct to hold the function and its purity.
Update `Value::Function` to store `Rc<NativeFunction>` and
`Value::make_function`
helper to simplify creation.
This change allows tracking the purity of native functions, which is
useful
for optimization and static analysis.
The `is_tail` field on the `BoundKind::Call` node was an intermediate
representation for tail-call optimization and is no longer needed as a
distinct field. The TCO pass now embeds this information directly into
the `RuntimeMetadata` of the `ExecNode`, which is the VM's internal AST.
This simplifies the `BoundKind::Call` structure and removes redundant
information.
The `TailCall` variant has been merged into `Call` by adding an
`is_tail` boolean field. This simplifies the AST by reducing the number
of node variants and makes it easier to handle tail call optimization.
The `tco.rs` module is responsible for setting this flag when a call
occurs in tail position.
Brummel
Michael Schimmel