First working solution of new task

This commit is contained in:
Jan-Niclas Loosen
2026-03-05 18:03:55 +01:00
parent 691d6eba8c
commit de46c67129
11 changed files with 1077 additions and 9 deletions

View File

@@ -0,0 +1,351 @@
"""
live_variables.py — Live Variables backward dataflow analysis for TRIPLA CFGs.
A variable v is *live* at the entry of node n if there exists a path
n → … → use(v) where v is not redefined along the way.
Data structures
---------------
gen dict[int, set[Var]] — GEN(n) = variables *used* at n
kill dict[int, set[Var]] — KILL(n) = variables *defined* at n
in_sets dict[int, set[Var]] — live variables at node *entry*
out_sets dict[int, set[Var]] — live variables at node *exit*
Transfer equations (backward):
OUT(n) = IN(s) for all successors s
IN(n) = (OUT(n) KILL(n)) GEN(n)
Variables are represented in scoped form ``(scope, name)``, e.g. ``("f","x")``.
This avoids collisions between equal variable names in different functions.
This module also exports ``_BackwardAnalysisBase``, the shared base class
that ``ReachedUsesAnalysis`` in reached_uses.py inherits from. The base
provides:
• AST traversal to collect function-nesting and parameter metadata
• Lexical variable resolution (parameter shadowing handled correctly)
• BFS-based CFG-node → owning-function assignment
• Unified uses / defs extraction for all node types
Var = tuple[str, str]
"""
from __future__ import annotations
from collections import deque
from typing import TYPE_CHECKING
import cfg_build
import syntax
from cfg.CFG_Node import CFG_START
if TYPE_CHECKING:
from cfg.CFG import CFG
# ---------------------------------------------------------------------------
# Public type alias (imported by reached_uses.py)
# ---------------------------------------------------------------------------
GLOBAL_SCOPE = ""
Var = tuple[str, str] # (function_name|GLOBAL_SCOPE, variable_name)
# ============================================================================
# Shared base: function metadata, scope assignment, uses/defs extraction
# ============================================================================
class _BackwardAnalysisBase:
"""Infrastructure shared by LiveVariablesAnalysis and ReachedUsesAnalysis.
Calling ``super().__init__(cfg)`` from a subclass:
1. Snapshots cfg_build.FUNCTIONS.
2. Collects AST-level function-nesting and parameter metadata.
3. BFS-assigns every CFG node to its owning function.
4. Extracts uses and defs for every CFG node.
After __init__ the following attributes are available to subclasses:
self.cfg — the CFG object
self._functions — dict[str, tuple]: snapshot of cfg_build.FUNCTIONS
self._func_parent — dict[str, str|None]: lexical parent per function
self._func_params — dict[str, tuple[str,...]]: params per function
self._func_scope — dict[int, str]: node-id → owning function name
self.uses — dict[int, set[Var]]: variables used at each node
self.defs — dict[int, set[Var]]: variables defined at each node
"""
def __init__(self, cfg: "CFG") -> None:
self.cfg = cfg
# Snapshot FUNCTIONS so later global-state resets do not affect us.
self._functions: dict[str, tuple] = dict(cfg_build.FUNCTIONS)
self.uses: dict[int, set[Var]] = {}
self.defs: dict[int, set[Var]] = {}
self._func_parent, self._func_params = self._collect_function_metadata()
self._func_scope: dict[int, str] = self._compute_function_scope()
self._extract_uses_defs()
# ------------------------------------------------------------------
# Step 1a — Walk AST to collect lexical nesting + parameter lists
# ------------------------------------------------------------------
def _collect_function_metadata(
self,
) -> tuple[dict[str, str | None], dict[str, tuple[str, ...]]]:
"""Walk the AST and collect function-parent and parameter information.
Returns
-------
func_parent : dict[str, str | None]
func_parent[f] is the name of the immediately enclosing function
(or None for top-level functions).
func_params : dict[str, tuple[str, ...]]
func_params[f] is the ordered tuple of formal parameter names of f.
"""
func_parent: dict[str, str | None] = {}
func_params: dict[str, tuple[str, ...]] = {}
def visit(expr: syntax.EXPRESSION | None, current_func: str | None) -> None:
if expr is None:
return
if isinstance(expr, syntax.LET):
decls = expr.decl if isinstance(expr.decl, list) else [expr.decl]
# Register metadata for each declared function.
for d in decls:
if isinstance(d, syntax.DECL):
# Use assignment (last-seen wins) to stay consistent
# with cfg_build.FUNCTIONS, which also overwrites on
# duplicate names. setdefault (first-seen wins) would
# disagree when a nested function shadows a top-level
# one with the same name, causing wrong scope resolution.
func_parent[d.f_name] = current_func
func_params[d.f_name] = tuple(d.params)
# Recurse into function bodies and the in-expression.
for d in decls:
if isinstance(d, syntax.DECL):
visit(d.body, d.f_name)
else:
visit(d, current_func)
visit(expr.body, current_func)
return
for _, child in expr.children():
visit(child, current_func)
visit(self.cfg.ast, None)
return func_parent, func_params
# ------------------------------------------------------------------
# Step 1b — Resolve a variable name through the lexical scope chain
# ------------------------------------------------------------------
def _resolve_var(self, func: str | None, name: str) -> Var:
"""Resolve a variable name via lexical scope chain."""
if func is None:
return (GLOBAL_SCOPE, name)
cur: str | None = func
seen: set[str] = set()
while cur is not None and cur not in seen:
seen.add(cur)
if name in self._func_params.get(cur, ()):
return (cur, name)
cur = self._func_parent.get(cur)
# Fallback: local variable in current function scope.
return (func, name)
# ------------------------------------------------------------------
# Step 2 — BFS-assign every CFG node to its owning function
# ------------------------------------------------------------------
def _compute_function_scope(self) -> dict[int, str]:
"""BFS from each function's START node; return node-id → function-name.
Two stopping conditions keep attribution strictly inside each function:
1. Do not follow into a *different* function's CFG_START (prevents
attributing callee body nodes to the caller, and vice-versa).
2. Do not follow *past* the function's own CFG_END (prevents
following CFG_END → CFG_RETURN → continuation nodes that belong
to the *caller* context, which caused variables used there to be
resolved in the wrong scope).
The first function whose BFS claims a node wins.
"""
functions = self._functions
func_scope: dict[int, str] = {}
all_f_start_ids: set[int] = {fs.id for _, (fs, _) in functions.items()}
for f_name, (f_start, f_end) in functions.items():
queue: deque = deque([f_start])
while queue:
node = queue.popleft()
if node.id in func_scope:
continue # already claimed by an earlier function
func_scope[node.id] = f_name
# Stop here — do not follow CFG_END into caller context.
if node.id == f_end.id:
continue
for child in node.children:
# Do not follow into a different function's START.
if (
isinstance(child, CFG_START)
and child.id in all_f_start_ids
and child.id != f_start.id
):
continue
queue.append(child)
return func_scope
# ------------------------------------------------------------------
# Step 3 — Extract uses / defs for every CFG node
# ------------------------------------------------------------------
def _extract_uses_defs(self) -> None:
"""Populate ``self.uses`` and ``self.defs`` for every node in the CFG.
Extraction rules:
• CFG_START(DECL f(p1,…,pk)) → defs = {(f,p1), …, (f,pk)}
• Node wrapping ID(x) → uses = {lexical_resolve(func, x)}
• Node wrapping ASSIGN(x = e) → defs = {lexical_resolve(func, x)}
• Everything else → uses = {}, defs = {}
Sub-expressions already have their own CFG nodes and are not
re-inspected here; each node is responsible only for its own ast_node.
"""
for node in self.cfg.nodes():
nid = node.id
func = self._func_scope.get(nid) # None → outer / global scope
ast = node.ast_node
uses: set[Var] = set()
defs: set[Var] = set()
if isinstance(node, CFG_START) and isinstance(ast, syntax.DECL):
# Function entry defines each formal parameter.
for param in ast.params:
defs.add((ast.f_name, param))
elif ast is not None:
if isinstance(ast, syntax.ID):
resolved = self._resolve_var(func, ast.name)
uses.add(resolved)
elif isinstance(ast, syntax.ASSIGN):
resolved = self._resolve_var(func, ast.var.name)
defs.add(resolved)
self.uses[nid] = uses
self.defs[nid] = defs
# ============================================================================
# Live Variables Analysis
# ============================================================================
class LiveVariablesAnalysis(_BackwardAnalysisBase):
"""Backward dataflow analysis: Live Variables.
A variable (f, x) is *live* at the entry of node n if there is a path
from n to some use of (f, x) along which (f, x) is not redefined.
This is the simpler predecessor to ReachedUsesAnalysis (reached_uses.py):
it tracks which variables are live, not *where* they are used.
Attributes
----------
gen dict[int, set[Var]] GEN(n) = uses(n) — vars used at n
kill dict[int, set[Var]] KILL(n) = defs(n) — vars defined at n
in_sets dict[int, set[Var]] live variables at n's *entry*
out_sets dict[int, set[Var]] live variables at n's *exit*
(uses and defs are identical to gen / kill and are inherited from the
base class.)
Transfer equations (backward):
OUT(n) = IN(s) for all successors s
IN(n) = (OUT(n) KILL(n)) GEN(n)
"""
def __init__(self, cfg: "CFG") -> None:
# Base populates uses, defs, _func_scope, etc.
super().__init__(cfg)
self.gen: dict[int, set[Var]] = {}
self.kill: dict[int, set[Var]] = {}
self.in_sets: dict[int, set[Var]] = {}
self.out_sets: dict[int, set[Var]] = {}
self._build_gen_kill()
self.solve()
# ------------------------------------------------------------------
# Build gen / kill; initialise in / out to ∅
# ------------------------------------------------------------------
def _build_gen_kill(self) -> None:
"""GEN(n) = uses(n), KILL(n) = defs(n); initialise in/out sets."""
for node in self.cfg.nodes():
nid = node.id
self.gen[nid] = set(self.uses[nid])
self.kill[nid] = set(self.defs[nid])
self.in_sets[nid] = set()
self.out_sets[nid] = set()
# ------------------------------------------------------------------
# Backward worklist fixpoint
# ------------------------------------------------------------------
def solve(self) -> None:
"""Backward worklist until fixpoint.
Transfer:
OUT(n) = IN(s) for all successors s
IN(n) = (OUT(n) KILL(n)) GEN(n)
Only nodes reachable from cfg.START are processed (guard against
propagate=False parent references from CFG.__remove_and_rewire).
"""
nodes = list(self.cfg.nodes())
known: set[int] = set(self.gen.keys())
id_to_node = {n.id: n for n in nodes}
worklist: deque = deque(nodes)
# Build predecessor relation from children edges. This is more reliable
# than node.parents because CFG rewiring may add edges with
# propagate=False, leaving parent links stale.
preds: dict[int, set[int]] = {nid: set() for nid in known}
for node in nodes:
for child in node.children:
if child.id in known:
preds[child.id].add(node.id)
while worklist:
node = worklist.popleft()
nid = node.id
new_out: set[Var] = set()
for child in node.children:
if child.id in known:
new_out |= self.in_sets[child.id]
new_in: set[Var] = (new_out - self.kill[nid]) | self.gen[nid]
if new_out != self.out_sets[nid] or new_in != self.in_sets[nid]:
self.out_sets[nid] = new_out
self.in_sets[nid] = new_in
for pred_id in preds[nid]:
worklist.append(id_to_node[pred_id])
# ------------------------------------------------------------------
# Result
# ------------------------------------------------------------------
def live_vars_by_node(self) -> dict[int, set[Var]]:
"""Return the live-variable set at the *entry* of each node.
Returns
-------
dict[int, set[Var]]
Keys: CFG node ids whose in_set is non-empty.
Values: copy of the live-variable set at that node's entry.
"""
return {nid: set(vs) for nid, vs in self.in_sets.items() if vs}