fix(i18n): read refusals along a closed list of shapes, report the rest

The scan followed many shapes and called the rest unread, but each shape it learned opened a silent
branch: a parameter rebound in the body, a second visit of a followed helper, an `import app.x`, a name
bound twice or under an `if`, a spread or a store that replaces the code, a detail rebuilt from its
variable, a None reason. It now reads only what the app writes and reports everything else: a
branch the app never takes is gone (an `or`, a default, a base class's attribute, string and union
annotations, a bare module constant), a spread is clean only from an untouched `**kwargs` no call puts
a `code` in, and the detail's variable may only grow by `|=` or `["reason"]`. The real tree reads
exactly as before; every exit and every kept branch has a self-test that a mutant of it fails.

The detail test also fails a dict without a `code`, a JSONResponse sentence and a detail-less
HTTPException; the two that are no refusal (the setup gate's 503, an unmatched api path's 404) are
listed with why. The contract's assertion that compared the reason map with itself is gone.
This commit is contained in:
2026-10-31 18:45:02 +01:00
parent 5e8589bbf8
commit f8a8e905da
5 changed files with 985 additions and 373 deletions
+442 -298
View File
@@ -2,27 +2,37 @@
with every reason its code can be sent with.
The frontend's contract (`i18n/backendIds.contract.test.ts`) runs this and requires a sentence for each
code and reason in both languages. A regex over the source read only the shapes it was written for: a
reason after another key, a status passed by name, a second handler with the same text all went past it.
Here a reason is followed to where it comes from, and what cannot be followed is reported, never guessed:
code and reason in both languages. A regex over the source read only the shapes it was written for, and a
scan that tries to follow every shape opens a silent gap with each one it learns. So this one reads a
CLOSED list, the shapes the app writes, and reports everything else, never guesses:
- a literal; both arms of `a if c else b`, every operand of `a or b`;
- a module constant (`urlguard.REASON_MALFORMED`), a local variable's assignments;
- a parameter or attribute typed with a `Literal` (basedpyright holds every value passed in), the
attribute found on the class an `except … as e` catches, or the class a parameter is typed with;
- a `str` parameter of a helper (`refused`, `not_found`): followed to every call of it, its default
included, and a helper passed around as a value is unread;
- a call: its return annotation; a map lookup (`_SETTING_REASONS[exc.reason]`): the map's values, and
its keys must be exactly what the index can be.
- a reason: a string literal (`None` sends none: the site counts as reasonless); both arms of
`a if c else b`; a local variable bound only by plain `x = …`; a module constant reached as `mod.NAME`;
a call's `Literal` return annotation;
- a parameter typed with a `Literal` (basedpyright holds every value passed in), or a `str` one of a
module-level helper (`refused`, `not_found`), followed to every call, each of which must pass it by
position or by name; a parameter its body binds again is unread;
- an attribute: `e.x` of the exception an `except C as e` catches, or of a parameter typed `C`, when a
method of `C` declares `self.x: <Literal>`;
- a map lookup (`_SETTING_REASONS[exc.reason]`): the map's values, and its keys must be exactly what the
index can be;
- the code: one literal, never changed after. A `**` spread in the detail is the `**kwargs` of a helper or
of an exception's `__init__` that takes `reason` by name, used for nothing but being spread, and no call
of it passes `code=` or a `**` that is not such a spread itself. The detail's variable is changed only by
`|= {…}` or `["reason"] = …`, and goes nowhere but into an `HTTPException` / `JSONResponse`.
Anything else is UNREAD, and the contract fails on it unless no screen shows that code. Stdlib only: it
runs on the host, beside the frontend's tests (`python -I backend/tests/refusal_sites.py` → JSON).
A name bound twice at module level, inside a module-level `if`/`try`/loop, or imported from two places is
not looked up. A reference named like a followed helper that the scan cannot resolve (`import app.x`, a
relative import) is unread. Anything else is UNREAD, or an unread code, and the contract fails on it unless
no screen shows that code. Stdlib only: it runs on the host, beside the frontend's tests
(`python -I backend/tests/refusal_sites.py` → JSON).
"""
import ast
import json
import pathlib
import sys
from collections.abc import Iterator
from dataclasses import dataclass, field
APP = pathlib.Path(__file__).resolve().parents[1] / "app"
@@ -33,16 +43,34 @@ NOT_REFUSALS = {
"plex/tmdb.py": "a production country's ISO code",
}
# What a detail held in a variable may be handed to: the response it is sent in, nothing that could change it.
SENDERS = {"HTTPException", "JSONResponse"}
READS = {"get", "items", "keys", "values", "copy"}
Node = ast.AST
FUNCS = (ast.FunctionDef, ast.AsyncFunctionDef)
SCOPES = (ast.FunctionDef, ast.AsyncFunctionDef, ast.Lambda, ast.ClassDef)
COMPOUND = (
ast.If,
ast.Try,
ast.TryStar,
ast.For,
ast.AsyncFor,
ast.While,
ast.With,
ast.AsyncWith,
ast.Match,
)
@dataclass
class Res:
values: set[str] = field(default_factory=set)
unread: list[str] = field(default_factory=list)
none: bool = False # an arm sends no reason
def __or__(self, other: "Res") -> "Res":
return Res(self.values | other.values, self.unread + other.unread)
return Res(self.values | other.values, self.unread + other.unread, self.none or other.none)
@dataclass
@@ -51,10 +79,45 @@ class Module:
name: str # dotted, `app.…`
tree: ast.Module
parents: dict[Node, Node]
imports: dict[
str, tuple[str, str | None]
] # local name → (module, symbol or None for the module)
top: dict[str, Node] # module-level bindings: def, class, or the assigned value
# local name → (module, symbol or None for the module)
imports: dict[str, tuple[str, str | None]]
top: dict[str, Node] # module-level bindings: def, class, or the assignment
ambiguous: set[str] # bound more than once, conditionally, or imported from two places
def _own_scope(node: Node) -> Iterator[Node]:
"""`node` and what it holds in its own scope: a nested def, class or lambda is itself, not its body."""
stack = [node]
while stack:
n = stack.pop()
yield n
if n is not node and isinstance(n, SCOPES):
continue
stack.extend(ast.iter_child_nodes(n))
def _stored(target: Node) -> set[str]:
"""The names an assignment target binds, not the ones it reads (`LABELS[k] = v` binds none)."""
return {
n.id for n in ast.walk(target) if isinstance(n, ast.Name) and isinstance(n.ctx, ast.Store)
}
def _binds(node: Node) -> set[str]:
"""The names one node binds in the scope it sits in."""
if isinstance(node, ast.Assign):
return set().union(*(_stored(t) for t in node.targets))
if isinstance(node, (ast.AnnAssign, ast.AugAssign, ast.For, ast.AsyncFor, ast.NamedExpr)):
return _stored(node.target)
if isinstance(node, (ast.With, ast.AsyncWith)):
return set().union(*(_stored(i.optional_vars) for i in node.items if i.optional_vars))
if isinstance(node, (ast.Import, ast.ImportFrom)):
return {(a.asname or a.name).split(".")[0] for a in node.names}
if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
return {node.name}
if isinstance(node, (ast.MatchAs, ast.MatchStar, ast.ExceptHandler)) and node.name:
return {node.name}
return set()
def _module_name(path: str) -> str:
@@ -69,16 +132,18 @@ def _load(sources: dict[str, str]) -> dict[str, Module]:
for path, src in sources.items():
tree = ast.parse(src)
parents = {child: node for node in ast.walk(tree) for child in ast.iter_child_nodes(node)}
imports: dict[str, tuple[str, str | None]] = {}
origins: dict[str, set[tuple[str, str | None]]] = {}
for node in ast.walk(tree):
if isinstance(node, ast.ImportFrom) and node.module and node.level == 0:
for a in node.names:
imports[a.asname or a.name] = (node.module, a.name)
origins.setdefault(a.asname or a.name, set()).add((node.module, a.name))
elif isinstance(node, ast.Import):
for a in node.names:
if a.asname:
imports[a.asname] = (a.name, None)
origins.setdefault(a.asname, set()).add((a.name, None))
top: dict[str, Node] = {}
count: dict[str, int] = {}
ambiguous = {name for name, o in origins.items() if len(o) > 1}
for stmt in tree.body:
if isinstance(stmt, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
top[stmt.name] = stmt
@@ -88,8 +153,15 @@ def _load(sources: dict[str, str]) -> dict[str, Module]:
top[t.id] = stmt
elif isinstance(stmt, ast.AnnAssign) and isinstance(stmt.target, ast.Name):
top[stmt.target.id] = stmt
for n in [stmt] if isinstance(stmt, SCOPES) else _own_scope(stmt):
for name in _binds(n):
count[name] = count.get(name, 0) + 1
if isinstance(stmt, COMPOUND):
ambiguous.add(name)
ambiguous |= {name for name, c in count.items() if c > 1}
imports = {name: next(iter(o)) for name, o in origins.items() if name not in ambiguous}
name = _module_name(path)
mods[name] = Module(path, name, tree, parents, imports, top)
mods[name] = Module(path, name, tree, parents, imports, top, ambiguous)
return mods
@@ -98,11 +170,17 @@ class Scan:
self.mods = _load(sources)
self.errors: list[str] = []
self._following: set[tuple[str, str, str]] = set()
self._followed: dict[tuple[str, str, str], Res] = {}
self._clean: dict[int, bool] = {}
# --- names ---------------------------------------------------------------------------------------
def lookup(self, m: Module, name: str) -> tuple[Module, Node] | tuple[str, None] | None:
def lookup(
self, m: Module, name: str, seen: frozenset[tuple[str, str]] = frozenset()
) -> tuple[Module, Node] | tuple[str, None] | None:
"""What a module-level name is: (its module, its binding), or (a module's name, None)."""
if name in m.ambiguous:
return None
if name in m.top:
return m, m.top[name]
if name in m.imports:
@@ -111,19 +189,33 @@ class Scan:
return (mod, None) if mod in self.mods else None
if f"{mod}.{sym}" in self.mods:
return f"{mod}.{sym}", None
if mod in self.mods:
return self.lookup(self.mods[mod], sym)
if mod in self.mods and (mod, sym) not in seen:
return self.lookup(self.mods[mod], sym, seen | {(mod, sym)})
return None
def lookup_expr(self, m: Module, e: Node) -> tuple[Module, Node] | tuple[str, None] | None:
if isinstance(e, ast.Name):
return self.lookup(m, e.id)
return None if self.local(m, e) else self.lookup(m, e.id)
if isinstance(e, ast.Attribute):
base = self.lookup_expr(m, e.value)
if base and base[1] is None and isinstance(base[0], str):
return self.lookup(self.mods[base[0]], e.attr)
return None
def local(self, m: Module, e: ast.Name) -> bool:
"""`e` names a parameter or a variable of a function it sits in."""
fn = self.enclosing(m, e, FUNCS)
while fn is not None:
a = fn.args
if e.id in {
p.arg for p in [*a.posonlyargs, *a.args, *a.kwonlyargs, a.vararg, a.kwarg] if p
}:
return True
if self.assignments(fn, e.id):
return True
fn = self.enclosing(m, fn, FUNCS)
return False
def at(self, m: Module, e: Node) -> str:
return f"{m.path}:{getattr(e, 'lineno', '?')}: {ast.unparse(e)}"
@@ -133,60 +225,106 @@ class Scan:
node = m.parents.get(node)
return node
def assignments(self, fn: Node, name: str) -> list[Node]:
"""What binds `name` in this function's own body (not in a nested one)."""
return [n for n in _own_scope(fn) if n is not fn and name in _binds(n)]
def caught_as(self, m: Module, e: Node, name: str) -> ast.ExceptHandler | None:
node = m.parents.get(e)
while node is not None and not isinstance(node, FUNCS):
if isinstance(node, ast.ExceptHandler) and node.name == name:
return node
node = m.parents.get(node)
return None
def references(
self, m: Module, target: Node
) -> Iterator[tuple[Module, Node, ast.Call | None, bool]]:
"""Every reference to a module-level def or class: the call it makes (None: not a call), and
whether it resolved. A reference named like it that the scan cannot resolve is yielded too."""
name = getattr(target, "name", None)
for cm in self.mods.values():
for node in ast.walk(cm.tree):
if not isinstance(node, (ast.Name, ast.Attribute)) or not isinstance(
node.ctx, ast.Load
):
continue
if (node.id if isinstance(node, ast.Name) else node.attr) != name:
continue
parent = cm.parents.get(node)
if isinstance(parent, ast.Attribute) and parent.value is node:
continue
found = self.lookup_expr(cm, node)
if found:
if not (found[0] is m and found[1] is target):
continue
elif not self.unresolved(cm, node):
continue
call = parent if isinstance(parent, ast.Call) and parent.func is node else None
yield cm, node, call, found is not None
def unresolved(self, m: Module, node: Node) -> bool:
"""A name, or a dotted path, the app could mean that the scan cannot resolve: not a local, nor an
outside module's (an attribute of an object the function holds is not one)."""
root = node
while isinstance(root, ast.Attribute):
root = root.value
if not isinstance(root, ast.Name) or self.local(m, root):
return False
if root.id in m.ambiguous or (root.id not in m.top and root.id not in m.imports):
return True
if root is node or root.id in m.top:
return False
found = self.lookup(m, root.id)
return (
found is not None and found[1] is None
) # an app module whose dotted path went nowhere
# --- types ---------------------------------------------------------------------------------------
def ann_values(self, m: Module, ann: Node | None) -> set[str] | None:
"""The values a `Literal` annotation allows, or None when it is not one (`str`, a class, …)."""
if ann is None:
return None
if isinstance(ann, ast.Constant):
if ann.value is None:
return set()
if isinstance(ann.value, str):
return self.ann_values(m, ast.parse(ann.value, mode="eval").body)
return None
def ann_values(self, m: Module, ann: Node | None, depth: int = 0) -> set[str] | None:
"""The values a `Literal[…]` annotation, or an alias of one, allows; None when it is not one."""
if isinstance(ann, ast.Subscript) and getattr(ann.value, "id", None) == "Literal":
items = ann.slice.elts if isinstance(ann.slice, ast.Tuple) else [ann.slice]
if all(isinstance(i, ast.Constant) and isinstance(i.value, str) for i in items):
return {i.value for i in items}
return None
if isinstance(ann, ast.BinOp) and isinstance(ann.op, ast.BitOr):
left, right = self.ann_values(m, ann.left), self.ann_values(m, ann.right)
return None if left is None or right is None else left | right
found = self.lookup_expr(m, ann)
if found and isinstance(found[1], ast.Assign) and isinstance(found[0], Module):
if isinstance(found[1].value, (ast.Subscript, ast.BinOp, ast.Name, ast.Attribute)):
return self.ann_values(found[0], found[1].value)
if isinstance(ann, (ast.Name, ast.Attribute)) and depth < 8:
found = self.lookup_expr(m, ann)
if found and isinstance(found[0], Module) and isinstance(found[1], ast.Assign):
return self.ann_values(found[0], found[1].value, depth + 1)
return None
def classes(self, m: Module, e: Node | None) -> list[tuple[Module, ast.ClassDef]] | None:
if e is None:
return None
if isinstance(e, ast.Tuple):
out = [self.classes(m, x) for x in e.elts]
return None if any(o is None for o in out) else [c for o in out for c in o]
found = self.lookup_expr(m, e)
found = self.lookup_expr(m, e) if e is not None else None
if found and isinstance(found[0], Module) and isinstance(found[1], ast.ClassDef):
return [(found[0], found[1])]
return None
def typed(self, m: Module, at: Node, name: str) -> list[tuple[Module, ast.ClassDef]] | None:
"""The class `name` holds where `at` sits: the one an `except C as name` catches, or a parameter's
annotation (a parameter its body binds again holds no known class)."""
handler = self.caught_as(m, at, name)
if handler is not None:
return self.classes(m, handler.type)
fn = self.enclosing(m, at, FUNCS)
if fn is None:
return None
params = {a.arg: a for a in [*fn.args.posonlyargs, *fn.args.args, *fn.args.kwonlyargs]}
if name in params and not self.assignments(fn, name):
return self.classes(m, params[name].annotation)
return None
def attr_values(self, cm: Module, cls: ast.ClassDef, attr: str) -> set[str] | None:
"""An attribute's `Literal` values, from its annotation on the class or `self.attr: T` in a method,
else from a base class."""
"""An attribute's `Literal` values, from `self.attr: T` in a method of the class."""
for node in ast.walk(cls):
if isinstance(node, ast.AnnAssign):
t = node.target
if (isinstance(t, ast.Name) and t.id == attr and node in cls.body) or (
isinstance(t, ast.Attribute)
and t.attr == attr
and getattr(t.value, "id", None) == "self"
):
return self.ann_values(cm, node.annotation)
for base in cls.bases:
for bm, bc in self.classes(cm, base) or []:
v = self.attr_values(bm, bc, attr)
if v is not None:
return v
if (
isinstance(node, ast.AnnAssign)
and isinstance(node.target, ast.Attribute)
and node.target.attr == attr
and getattr(node.target.value, "id", None) == "self"
):
return self.ann_values(cm, node.annotation)
return None
# --- reasons -------------------------------------------------------------------------------------
@@ -194,27 +332,18 @@ class Scan:
def resolve(self, m: Module, e: Node) -> Res:
if isinstance(e, ast.Constant):
if e.value is None:
return Res()
return Res(none=True)
if isinstance(e.value, str):
return Res({e.value})
if isinstance(e, ast.IfExp):
return self.resolve(m, e.body) | self.resolve(m, e.orelse)
if isinstance(e, ast.BoolOp) and isinstance(e.op, ast.Or):
out = Res()
for v in e.values:
out = out | self.resolve(m, v)
return out
if isinstance(e, ast.Name):
return self.resolve_name(m, e)
if isinstance(e, ast.Attribute):
return self.resolve_attr(m, e)
if isinstance(e, ast.Call):
found = self.lookup_expr(m, e.func)
if (
found
and isinstance(found[0], Module)
and isinstance(found[1], (ast.FunctionDef, ast.AsyncFunctionDef))
):
if found and isinstance(found[0], Module) and isinstance(found[1], FUNCS):
v = self.ann_values(found[0], found[1].returns)
if v is not None:
return Res(v)
@@ -223,95 +352,31 @@ class Scan:
return Res(unread=[self.at(m, e)])
def resolve_name(self, m: Module, e: ast.Name) -> Res:
fn = self.enclosing(m, e, (ast.FunctionDef, ast.AsyncFunctionDef))
fn = self.enclosing(m, e, FUNCS)
while fn is not None:
assert isinstance(fn, (ast.FunctionDef, ast.AsyncFunctionDef))
assert isinstance(fn, FUNCS)
if self.caught_as(m, e, e.id) is not None:
return Res(unread=[self.at(m, e)]) # the exception itself, not a reason
params = {a.arg: a for a in [*fn.args.posonlyargs, *fn.args.args, *fn.args.kwonlyargs]}
if e.id in params:
return self.resolve_param(m, fn, params[e.id], e)
if (
fn.args.vararg
and fn.args.vararg.arg == e.id
or fn.args.kwarg
and fn.args.kwarg.arg == e.id
):
return Res(unread=[self.at(m, e)])
a = fn.args
params = {p.arg: p for p in [*a.posonlyargs, *a.args, *a.kwonlyargs]}
assigned = self.assignments(fn, e.id)
if e.id in params:
if assigned:
return Res(unread=[f"{self.at(m, e)} (a parameter its body binds again)"])
return self.resolve_param(m, fn, params[e.id], e)
if assigned:
out = Res()
for a in assigned:
if (
isinstance(a, ast.AnnAssign)
and (v := self.ann_values(m, a.annotation)) is not None
for s in assigned:
if isinstance(s, ast.Assign) and all(
isinstance(t, ast.Name) for t in s.targets
):
out = out | Res(v)
elif (
isinstance(a, (ast.Assign, ast.AnnAssign))
and a.value is not None
and not any(
isinstance(t, (ast.Tuple, ast.List)) for t in getattr(a, "targets", [])
)
):
out = out | self.resolve(m, a.value)
out = out | self.resolve(m, s.value)
else:
out = out | Res(unread=[self.at(m, e)])
return out
fn = self.enclosing(m, fn, (ast.FunctionDef, ast.AsyncFunctionDef))
found = self.lookup(m, e.id)
if (
found
and isinstance(found[0], Module)
and isinstance(found[1], (ast.Assign, ast.AnnAssign))
):
stmt = found[1]
if stmt.value is not None and isinstance(stmt.value, ast.Constant):
return self.resolve(found[0], stmt.value)
fn = self.enclosing(m, fn, FUNCS)
return Res(unread=[self.at(m, e)])
def assignments(self, fn: Node, name: str) -> list[Node]:
"""The statements that bind `name` in this function's own body (not in a nested one)."""
out: list[Node] = []
stack = list(ast.iter_child_nodes(fn))
while stack:
node = stack.pop()
if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.Lambda, ast.ClassDef)):
continue
if isinstance(node, ast.Assign):
if any(
name in {n.id for n in ast.walk(t) if isinstance(n, ast.Name)}
for t in node.targets
):
out.append(node)
elif isinstance(node, ast.AnnAssign) and getattr(node.target, "id", None) == name:
out.append(node)
elif isinstance(
node, (ast.For, ast.With, ast.AugAssign, ast.NamedExpr, ast.comprehension)
):
targets = [getattr(node, "target", None)] + [
i.optional_vars
for i in getattr(node, "items", [])
if i.optional_vars is not None
]
if any(
t is not None and name in {n.id for n in ast.walk(t) if isinstance(n, ast.Name)}
for t in targets
):
out.append(
node
) # bound by a loop, a `with`, an augmented or a walrus: not followed
stack.extend(ast.iter_child_nodes(node))
return out
def caught_as(self, m: Module, e: Node, name: str) -> ast.ExceptHandler | None:
node = m.parents.get(e)
while node is not None and not isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
if isinstance(node, ast.ExceptHandler) and node.name == name:
return node
node = m.parents.get(node)
return None
def resolve_param(self, m: Module, fn: Node, param: ast.arg, e: Node) -> Res:
v = self.ann_values(m, param.annotation)
if v is not None:
@@ -322,70 +387,45 @@ class Scan:
def follow(self, m: Module, fn: Node, param: ast.arg) -> Res:
"""A `str` parameter of a helper: what every call of the helper passes it."""
assert isinstance(fn, (ast.FunctionDef, ast.AsyncFunctionDef))
assert isinstance(fn, FUNCS)
key = (m.name, fn.name, param.arg)
if key in self._followed:
return self._followed[key]
where = f"{m.path}:{fn.lineno}: {fn.name}({param.arg})"
if key in self._following:
return Res()
if m.top.get(fn.name) is not fn:
return Res(
unread=[
f"{m.path}:{fn.lineno}: {fn.name}({param.arg}) is not a module-level helper"
]
)
return Res(unread=[f"{where} reached again while it is followed"])
if m.top.get(fn.name) is not fn or fn.name in m.ambiguous:
return Res(unread=[f"{where} is not a module-level helper"])
self._following.add(key)
positional = [a.arg for a in [*fn.args.posonlyargs, *fn.args.args]]
defaults = dict(
zip(
positional[len(positional) - len(fn.args.defaults) :], fn.args.defaults, strict=True
)
)
defaults |= {
a.arg: d
for a, d in zip(fn.args.kwonlyargs, fn.args.kw_defaults, strict=True)
if d is not None
}
by_name = param not in fn.args.posonlyargs
out = Res()
for cm in self.mods.values():
for node in ast.walk(cm.tree):
if not isinstance(node, (ast.Name, ast.Attribute)) or isinstance(
cm.parents.get(node), ast.Attribute
):
continue
found = self.lookup_expr(cm, node)
if not (found and found[0] is m and found[1] is fn):
continue
call = cm.parents.get(node)
if not (isinstance(call, ast.Call) and call.func is node):
out = out | Res(unread=[f"{self.at(cm, node)} (the helper used as a value)"])
continue
kw = {k.arg: k.value for k in call.keywords}
given = param.arg in positional and positional.index(param.arg) < len(call.args)
if param.arg in kw:
out = out | self.resolve(cm, kw[param.arg])
elif any(isinstance(a, ast.Starred) for a in call.args):
out = out | Res(unread=[self.at(cm, call)])
elif given: # given by position, a `**` spread cannot pass it again (a TypeError)
out = out | self.resolve(cm, call.args[positional.index(param.arg)])
elif None in kw:
out = out | Res(unread=[self.at(cm, call)])
elif param.arg in defaults:
out = out | self.resolve(m, defaults[param.arg])
for cm, node, call, resolved in self.references(m, fn):
if not resolved:
out = out | Res(
unread=[f"{self.at(cm, node)} (a reference the scan cannot resolve)"]
)
continue
if call is None:
out = out | Res(unread=[f"{self.at(cm, node)} (the helper used as a value)"])
continue
kw = {k.arg: k.value for k in call.keywords}
given = param.arg in positional and positional.index(param.arg) < len(call.args)
if by_name and param.arg in kw:
out = out | self.resolve(cm, kw[param.arg])
elif any(isinstance(a, ast.Starred) for a in call.args):
out = out | Res(unread=[self.at(cm, call)])
elif given:
out = out | self.resolve(cm, call.args[positional.index(param.arg)])
else:
out = out | Res(unread=[f"{self.at(cm, call)} (does not pass {param.arg})"])
self._followed[key] = out # once followed, its key answers from here: never reached again
return out
def resolve_attr(self, m: Module, e: ast.Attribute) -> Res:
base = e.value
if isinstance(base, ast.Name):
handler = self.caught_as(m, e, base.id)
fn = self.enclosing(m, e, (ast.FunctionDef, ast.AsyncFunctionDef))
typed = None
if handler is not None:
typed = self.classes(m, handler.type)
elif fn is not None:
params = {
a.arg: a for a in [*fn.args.posonlyargs, *fn.args.args, *fn.args.kwonlyargs]
}
if base.id in params:
typed = self.classes(m, params[base.id].annotation)
typed = self.typed(m, e, base.id)
if typed is not None:
out = Res()
for cm, cls in typed:
@@ -426,144 +466,248 @@ class Scan:
for v in table.values:
out = out | self.resolve(found[0], v)
index = self.resolve(m, e.slice)
keys = {k.value for k in table.keys}
if index.unread or index.values != keys:
keys = {k.value for k in table.keys if isinstance(k, ast.Constant)}
if index.unread or index.none or index.values != keys:
self.errors.append(
f"{self.at(m, e)}: the map's keys {sorted(keys)} are not what its index can be "
f"{sorted(index.values)}{' + unread ' + str(index.unread) if index.unread else ''}"
)
return out
# --- sites ---------------------------------------------------------------------------------------
def spread_is_safe(self, m: Module, e: Node, src: Node) -> bool:
"""A `**x` in a detail, or `detail |= {…}` from it, cannot carry a reason when `x` is the `**kwargs`
of a function that takes `reason` by name: Python binds that one to the parameter."""
if isinstance(src, ast.Attribute):
return self.spread_is_safe_attr(m, e, src)
fn = self.enclosing(m, e, (ast.FunctionDef, ast.AsyncFunctionDef))
if fn is None or fn.args.kwarg is None:
return False
names = {a.arg for a in [*fn.args.posonlyargs, *fn.args.args, *fn.args.kwonlyargs]}
kwarg = fn.args.kwarg.arg
if isinstance(src, ast.Name):
return src.id == kwarg and "reason" in names
# --- spreads -------------------------------------------------------------------------------------
def clean_spread(self, m: Module, at: Node, src: Node) -> bool:
"""`**src` cannot carry a `reason` or a `code`: it is a function's `**kwargs` (or a caught
exception's `e.params` kept from its `__init__`'s), or `{k: … for k, v in that.items()}`: a
value may change, a key may not."""
if isinstance(src, ast.DictComp):
it = src.generators[0].iter
g = src.generators
if len(g) != 1 or not isinstance(g[0].target, ast.Tuple) or len(g[0].target.elts) != 2:
return False
k, v = g[0].target.elts
it = g[0].iter
return (
isinstance(it, ast.Call)
isinstance(k, ast.Name)
and isinstance(v, ast.Name)
and getattr(src.key, "id", None) == k.id
and isinstance(it, ast.Call)
and isinstance(it.func, ast.Attribute)
and it.func.attr == "items"
and getattr(it.func.value, "id", None) == kwarg
and "reason" in names
and not it.args
and self.clean_spread(m, at, it.func.value)
)
if isinstance(src, ast.Attribute) and isinstance(src.value, ast.Name):
typed = self.typed(m, at, src.value.id)
return bool(typed) and all(self.kept(cm, cls, src.attr) for cm, cls in typed or [])
if isinstance(src, ast.Name):
fn = self.enclosing(m, at, FUNCS)
return (
fn is not None
and fn.args.kwarg is not None
and src.id == fn.args.kwarg.arg
and self.kwargs_clean(m, fn, fn)
)
return False
def spread_is_safe_attr(self, m: Module, e: Node, src: ast.Attribute) -> bool:
"""`e.params` of a caught exception whose `__init__(self, reason, **params)` keeps `self.params = params`."""
if not isinstance(src.value, ast.Name):
return False
handler = self.caught_as(m, e, src.value.id)
typed = self.classes(m, handler.type) if handler else None
return bool(typed) and all(self.keeps_kwargs(cls, src.attr) for _, cls in typed or [])
def keeps_kwargs(self, cls: ast.ClassDef, attr: str) -> bool:
def kept(self, cm: Module, cls: ast.ClassDef, attr: str) -> bool:
"""`self.attr` is only ever the `**kwargs` of the class's `__init__`, and that one is clean."""
init = next(
(s for s in cls.body if isinstance(s, ast.FunctionDef) and s.name == "__init__"), None
)
if init is None or init.args.kwarg is None:
return False
names = {a.arg for a in [*init.args.posonlyargs, *init.args.args, *init.args.kwonlyargs]}
kwarg = init.args.kwarg.arg
return "reason" in names and any(
isinstance(n, ast.Assign)
and any(
isinstance(t, ast.Attribute)
and t.attr == attr
and getattr(t.value, "id", None) == "self"
for t in n.targets
)
and getattr(n.value, "id", None) == kwarg
for n in ast.walk(init)
stores = [
n
for n in ast.walk(cls)
if isinstance(n, ast.Attribute)
and n.attr == attr
and getattr(n.value, "id", None) == "self"
and not isinstance(n.ctx, ast.Load)
]
kept = [
s
for s in stores
if isinstance(p := cm.parents.get(s), ast.Assign)
and getattr(p.value, "id", None) == init.args.kwarg.arg
and self.enclosing(cm, p, FUNCS) is init
]
return len(stores) == 1 and len(kept) == 1 and self.kwargs_clean(cm, init, cls)
def kwargs_clean(self, m: Module, fn: Node, target: Node) -> bool:
"""`fn`'s `**kwargs` holds no `reason` (`fn` takes it by name) and no `code`: it is used for
nothing but `**kwargs`, `kwargs.items()`, `self.x = kwargs` or an f-string, and every call of
`target` (the helper, or the class `fn` is the `__init__` of) passes no `code=` and no unclean
`**`."""
assert isinstance(fn, FUNCS) and fn.args.kwarg is not None
if id(fn) in self._clean:
return self._clean[id(fn)]
self._clean[id(fn)] = False # a call that reaches it again while it is checked is not clean
kwarg = fn.args.kwarg.arg
clean = "reason" in {a.arg for a in [*fn.args.args, *fn.args.kwonlyargs]} and not (
self.assignments(fn, kwarg)
)
for n in _own_scope(fn):
if not clean:
break
if isinstance(n, ast.Name) and n.id == kwarg:
p = m.parents.get(n)
clean = (
isinstance(p, ast.FormattedValue) # printed in a message
or isinstance(p, ast.keyword)
and p.arg is None
or isinstance(p, ast.Dict)
and any(k is None and v is n for k, v in zip(p.keys, p.values, strict=True))
or isinstance(p, ast.Attribute)
and p.attr == "items"
and isinstance(m.parents.get(p), ast.Call)
or isinstance(p, ast.Assign)
and p.value is n
and all(
isinstance(t, ast.Attribute) and getattr(t.value, "id", None) == "self"
for t in p.targets
)
)
refs = list(self.references(m, target)) if clean else []
for cm, node, call, resolved in refs:
if call is not None:
clean = not any(
k.arg == "code" or k.arg is None and not self.clean_spread(cm, call, k.value)
for k in call.keywords
)
elif not resolved or not isinstance(target, ast.ClassDef):
clean = False # a helper passed around, or a name the scan cannot resolve
else:
# `except C`, an annotation: not a construction; a subclass is built some other way
cls = self.enclosing(cm, node, ast.ClassDef)
clean = not (cls is not None and any(node in ast.walk(b) for b in cls.bases))
if not clean:
break
self._clean[id(fn)] = clean
return clean
# --- sites ---------------------------------------------------------------------------------------
def entries(self, d: Node) -> list[tuple[str | None, Node]]:
"""A dict's (key, value) pairs: None for a `**` spread, "?" for a key that is not a literal."""
if isinstance(d, ast.Dict):
return [
(k.value if isinstance(k, ast.Constant) else None if k is None else "?", v)
(
k.value
if isinstance(k, ast.Constant) and isinstance(k.value, str)
else None
if k is None
else "?",
v,
)
for k, v in zip(d.keys, d.values, strict=True)
]
assert isinstance(d, ast.Call)
return [(k.arg, k.value) for k in d.keywords]
return [(None, a) for a in d.args] + [(k.arg, k.value) for k in d.keywords]
def site(self, m: Module, d: Node, out: dict) -> None:
items = self.entries(d)
code_expr = next(v for k, v in items if k == "code")
code = self.resolve(m, code_expr)
if code.unread or len(code.values) != 1:
out["unreadCodes"].append(self.at(m, code_expr))
code = Res()
for v in (v for k, v in items if k == "code"):
code = code | self.resolve(m, v)
problems = [
f"{self.at(m, v)} (a key the scan cannot read)" for k, v in items if k == "?"
] + [
f"{self.at(m, v)} (a spread that may carry the code)"
for k, v in items
if k is None and not self.clean_spread(m, d, v)
]
res, has_reason, changed = self.mutations(m, d)
problems += changed
if code.unread or code.none or len(code.values) != 1 or problems:
out["unreadCodes"] += problems or [self.at(m, d)]
return
entry = out["codes"].setdefault(
next(iter(code.values)), {"reasons": set(), "unread": [], "sites": 0, "reasonless": []}
)
entry["sites"] += 1
res, has_reason = Res(), False
for k, v in items:
if k == "reason":
res, has_reason = res | self.resolve(m, v), True
elif k is None and not self.spread_is_safe(m, d, v):
res = res | Res(unread=[f"{self.at(m, v)} (a spread that may carry a reason)"])
res, more = self.mutations(m, d, res)
has_reason = has_reason or more
if not has_reason:
if not has_reason or res.none:
entry["reasonless"].append(f"{m.path}:{d.lineno}")
entry["reasons"] |= res.values
entry["unread"] += res.unread
def mutations(self, m: Module, d: Node, res: Res) -> tuple[Res, bool]:
"""`detail |= {…}`, `detail["reason"] = …`, `detail.update(…)` on the variable a detail is put in."""
stmt = m.parents.get(d)
if not (isinstance(stmt, (ast.Assign, ast.AnnAssign)) and stmt.value is d):
return res, False
target = stmt.targets[0] if isinstance(stmt, ast.Assign) else stmt.target
fn = self.enclosing(m, d, (ast.FunctionDef, ast.AsyncFunctionDef))
if not isinstance(target, ast.Name) or fn is None:
return res, False
found = False
for node in ast.walk(fn):
if isinstance(node, ast.AugAssign) and getattr(node.target, "id", None) == target.id:
v = node.value
if isinstance(v, ast.Dict):
for k, val in self.entries(v):
def mutations(self, m: Module, d: Node) -> tuple[Res, bool, list[str]]:
"""A detail put in a variable: what `|= {…}` and `["reason"] = …` add, and every other use that
could change it or send it elsewhere."""
holder, stmt = d, m.parents.get(d)
while isinstance(stmt, ast.IfExp) and holder is not stmt.test:
holder, stmt = stmt, m.parents.get(stmt)
if not (isinstance(stmt, (ast.Assign, ast.AnnAssign)) and stmt.value is holder):
return Res(), False, []
targets = stmt.targets if isinstance(stmt, ast.Assign) else [stmt.target]
fn = self.enclosing(m, d, FUNCS)
if fn is None or len(targets) != 1 or not isinstance(targets[0], ast.Name):
return (
Res(),
False,
[f"{self.at(m, stmt)} (a detail put where the scan cannot follow it)"],
)
name = targets[0].id
res, found, problems = Res(), False, []
for n in _own_scope(fn):
if n is stmt:
continue
if name in _binds(n):
if not (isinstance(n, ast.AugAssign) and isinstance(n.op, ast.BitOr)):
problems.append(f"{self.at(m, n)} (binds the detail's variable again)")
elif isinstance(n.value, ast.Dict):
for k, v in self.entries(n.value):
if k == "reason":
res, found = res | self.resolve(m, val), True
elif k is None and not self.spread_is_safe(m, node, val):
res = res | Res(
unread=[f"{self.at(m, val)} (a spread that may carry a reason)"]
)
elif not self.spread_is_safe(m, node, v):
res = res | Res(unread=[f"{self.at(m, node)} (may carry a reason)"])
elif (
isinstance(node, ast.Subscript)
and getattr(node.value, "id", None) == target.id
and isinstance(node.ctx, ast.Store)
):
if isinstance(node.slice, ast.Constant) and node.slice.value != "reason":
res, found = res | self.resolve(m, v), True
elif k in ("code", "?") or k is None and not self.clean_spread(m, n, v):
problems.append(f"{self.at(m, v)} (may change the code)")
elif not self.clean_spread(m, n, n.value):
problems.append(f"{self.at(m, n)} (may change the code)")
continue
if not (isinstance(n, ast.Name) and n.id == name and isinstance(n.ctx, ast.Load)):
continue
p = m.parents.get(n)
if isinstance(p, ast.AugAssign) and p.target is n:
continue
if isinstance(p, ast.Subscript) and p.value is n:
key = p.slice.value if isinstance(p.slice, ast.Constant) else None
assign = m.parents.get(p)
if isinstance(p.ctx, ast.Load):
continue
assign = m.parents.get(node)
if isinstance(node.slice, ast.Constant) and isinstance(assign, ast.Assign):
if (
key == "reason"
and isinstance(p.ctx, ast.Store)
and isinstance(assign, ast.Assign)
and assign.targets == [p]
):
res, found = res | self.resolve(m, assign.value), True
else:
res = res | Res(unread=[f"{self.at(m, node)} (may carry a reason)"])
elif (
isinstance(node, ast.Call)
and isinstance(node.func, ast.Attribute)
and getattr(node.func.value, "id", None) == target.id
and node.func.attr in {"update", "setdefault"}
elif not (
isinstance(key, str)
and key not in ("code", "reason")
and isinstance(p.ctx, ast.Store)
):
problems.append(f"{self.at(m, p)} (may change the code or drop the reason)")
continue
if isinstance(p, ast.Attribute) and p.attr in READS:
continue
sender = (
p
if isinstance(p, ast.Call)
else m.parents.get(p)
if isinstance(p, ast.keyword)
else None
)
func = getattr(sender, "func", None)
if (
isinstance(sender, ast.Call)
and (getattr(func, "id", None) or getattr(func, "attr", None)) in SENDERS
and (n in sender.args or isinstance(p, ast.keyword) and p.arg == "detail")
):
res = res | Res(unread=[f"{self.at(m, node)} (may carry a reason)"])
return res, found
continue
problems.append(f"{self.at(m, n)} (the detail goes where the scan cannot follow it)")
return res, found, problems
def run(self) -> dict:
out: dict = {