Refs https://github.com/astral-sh/ty/issues/544
## Summary
Takes a more incremental approach to PEP 613 type alias support (vs
https://github.com/astral-sh/ruff/pull/20107). Instead of eagerly
inferring the RHS of a PEP 613 type alias as a type expression, infer it
as a value expression, just like we do for implicit type aliases, taking
advantage of the same support for e.g. unions and other type special
forms.
The main reason I'm following this path instead of the one in
https://github.com/astral-sh/ruff/pull/20107 is that we've realized that
people do sometimes use PEP 613 type aliases as values, not just as
types (because they are just a normal runtime assignment, unlike PEP 695
type aliases which create an opaque `TypeAliasType`).
This PR doesn't yet provide full support for recursive type aliases
(they don't panic, but they just fall back to `Unknown` at the recursion
point). This is future work.
## Test Plan
Added mdtests.
Many new ecosystem diagnostics, mostly because we
understand new types in lots of places.
Conformance suite changes are correct.
Performance regression is due to understanding lots of new
types; nothing we do in this PR is inherently expensive.
## Summary
Eagerly evaluate the elements of a PEP 604 union in value position (e.g.
`IntOrStr = int | str`) as type expressions and store the result (the
corresponding `Type::Union` if all elements are valid type expressions,
or the first encountered `InvalidTypeExpressionError`) on the
`UnionTypeInstance`, such that the `Type::Union(…)` does not need to be
recomputed every time the implicit type alias is used in a type
annotation.
This might lead to performance improvements for large unions, but is
also necessary for correctness, because the elements of the union might
refer to type variables that need to be looked up in the scope of the
type alias, not at the usage site.
## Test Plan
New Markdown tests
#21414 added the ability to create a specialization from a constraint
set. It handled mutually constrained typevars just fine, e.g. given `T ≤
int ∧ U = T` we can infer `T = int, U = int`.
But it didn't handle _nested_ constraints correctly, e.g. `T ≤ int ∧ U =
list[T]`. Now we do! This requires doing a fixed-point "apply the
specialization to itself" step to propagate the assignments of any
nested typevars, and then a cycle detection check to make sure we don't
have an infinite expansion in the specialization.
This gets at an interesting nuance in our constraint set structure that
@sharkdp has asked about before. Constraint sets are BDDs, and each
internal node represents an _individual constraint_, of the form `lower
≤ T ≤ upper`. `lower` and `upper` are allowed to be other typevars, but
only if they appear "later" in the arbitary ordering that we establish
over typevars. The main purpose of this is to avoid infinite expansion
for mutually constrained typevars.
However, that restriction doesn't help us here, because only applies
when `lower` and `upper` _are_ typevars, not when they _contain_
typevars. That distinction is important, since it means the restriction
does not affect our expressiveness: we can always rewrite `Never ≤ T ≤
U` (a constraint on `T`) into `T ≤ U ≤ object` (a constraint on `U`).
The same is not true of `Never ≤ T ≤ list[U]` — there is no "inverse" of
`list` that we could apply to both sides to transform this into a
constraint on a bare `U`.
## Summary
Fixes https://github.com/astral-sh/ty/issues/1571.
I realised I was overcomplicating things when I described what we should
do in that issue description. The simplest thing to do here is just to
special-case call expressions and short-circuit the call-binding
machinery entirely if we see it's `NotImplemented` being called. It
doesn't really matter if the subdiagnostic doesn't fire when a union is
called and one element of the union is `NotImplemented` -- the
subdiagnostic doesn't need to be exhaustive; it's just to help people in
some common cases.
## Test Plan
Added snapshots
This patch lets us create specializations from a constraint set. The
constraint encodes the restrictions on which types each typevar can
specialize to. Given a generic context and a constraint set, we iterate
through all of the generic context's typevars. For each typevar, we
abstract the constraint set so that it only mentions the typevar in
question (propagating derived facts if needed). We then find the "best
representative type" for the typevar given the abstracted constraint
set.
When considering the BDD structure of the abstracted constraint set,
each path from the BDD root to the `true` terminal represents one way
that the constraint set can be satisfied. (This is also one of the
clauses in the DNF representation of the constraint set's boolean
formula.) Each of those paths is the conjunction of the individual
constraints of each internal node that we traverse as we walk that path,
giving a single lower/upper bound for the path. We use the upper bound
as the "best" (i.e. "closest to `object`") type for that path.
If there are multiple paths in the BDD, they technically represent
independent possible specializations. If there's a single specialization
that satisfies all of them, we will return that as the specialization.
If not, then the constraint set is ambiguous. (This happens most often
with constrained typevars.) We could in the future turn _each_ of the
paths into separate specializations, but it's not clear what we would do
with that, so instead we just report the ambiguity as a specialization
failure.
## Summary
Add a set of comprehensive tests for generic implicit type aliases to
illustrate the current behavior with many flavors of `@Todo` types and
false positive diagnostics.
The tests are partially based on the typing conformance suite, and the
expected behavior has been checked against other type checkers.
## Summary
Get rid of the catch-all todo type from subscripting a base type we
haven't implemented handling for yet in a type expression, and turn it
into a diagnostic instead.
Handle a few more cases explicitly, to avoid false positives from the
above change:
1. Subscripting any dynamic type (not just a todo type) in a type
expression should just result in that same dynamic type. This is
important for gradual guarantee, and matches other type checkers.
2. Subscripting a generic alias may be an error or not, depending
whether the specialization itself contains typevars. Don't try to handle
this yet (it should be handled in a later PR for specializing generic
non-PEP695 type aliases), just use a dedicated todo type for it.
3. Add a temporary todo branch to avoid false positives from string PEP
613 type aliases. This can be removed in the next PR, with PEP 613 type
alias support.
## Test Plan
Adjusted mdtests, ecosystem.
All new diagnostics in conformance suite are supposed to be diagnostics,
so this PR is a strict improvement there.
New diagnostics in the ecosystem are surfacing cases where we already
don't understand an annotation, but now we emit a diagnostic about it.
They are mostly intentional choices. Analysis of particular cases:
* `attrs`, `bokeh`, `django-stubs`, `dulwich`, `ibis`, `kornia`,
`mitmproxy`, `mongo-python-driver`, `mypy`, `pandas`, `poetry`,
`prefect`, `pydantic`, `pytest`, `scrapy`, `trio`, `werkzeug`, and
`xarray` are all cases where under `from __future__ import annotations`
or Python 3.14 deferred-annotations semantics, we follow normal
name-scoping rules, whereas some other type checkers prefer global names
over local names. This means we don't like it if e.g. you have a class
with a method or attribute named `type` or `tuple`, and you also try to
use `type` or `tuple` in method/attribute annotations of that class.
This PR isn't changing those semantics, just revealing them in more
cases where previously we just silently fell back to `Unknown`. I think
failing with a diagnostic (so authors can alias names as needed to avoid
relying on scoping rules that differ between type checkers) is better
than failing silently here.
* `beartype` assumes we support `TypeForm` (because it only supports
mypy and pyright, it uses `if MYPY:` to hide the `TypeForm` from mypy,
and pyright supports `TypeForm`), and we don't yet.
* `graphql-core` likes to use a `try: ... except ImportError: ...`
pattern for importing special forms from `typing` with fallback to
`typing_extensions`, instead of using `sys.version_info` checks. We
don't handle this well when type checking under an older Python version
(where the import from `typing` is not found); we see the imported name
as of type e.g. `Unknown | SpecialFormType(...)`, and because of the
union with `Unknown` we fail to handle it as the special form type. Mypy
and pyright also don't seem to support this pattern. They don't complain
about subscripting such special forms, but they do silently fail to
treat them as the desired special form. Again here, if we are going to
fail I'd rather fail with a diagnostic rather than silently.
* `ibis` is [trying to
use](https://github.com/ibis-project/ibis/blob/main/ibis/common/collections.py#L372)
`frozendict: type[FrozenDict]` as a way to create a "type alias" to
`FrozenDict`, but this is wrong: that means `frozendict:
type[FrozenDict[Any, Any]]`.
* `mypy` has some errors due to the fact that type-checking `typing.pyi`
itself (without knowing that it's the real `typing.pyi`) doesn't work
very well.
* `mypy-protobuf` imports some types from the protobufs library that end
up unioned with `Unknown` for some reason, and so we don't allow
explicit-specialization of them. Depending on the reason they end up
unioned with `Unknown`, we might want to better support this? But it's
orthogonal to this PR -- we aren't failing any worse here, just alerting
the author that we didn't understand their annotation.
* `pwndbg` has unresolved references due to star-importing from a
dependency that isn't installed, and uses un-imported names like `Dict`
in annotation expressions. Some of the unresolved references were hidden
by
https://github.com/astral-sh/ruff/blob/main/crates/ty_python_semantic/src/types/infer/builder.rs#L7223-L7228
when some annotations previously resolved to a Todo type that no longer
do.
This saga began with a regression in how we handle constraint sets where
a typevar is constrained by another typevar, which #21068 first added
support for:
```py
def mutually_constrained[T, U]():
# If [T = U ∧ U ≤ int], then [T ≤ int] must be true as well.
given_int = ConstraintSet.range(U, T, U) & ConstraintSet.range(Never, U, int)
static_assert(given_int.implies_subtype_of(T, int))
```
While working on #21414, I saw a regression in this test, which was
strange, since that PR has nothing to do with this logic! The issue is
that something in that PR made us instantiate the typevars `T` and `U`
in a different order, giving them differently ordered salsa IDs. And
importantly, we use these salsa IDs to define the variable ordering that
is used in our constraint set BDDs. This showed that our "mutually
constrained" logic only worked for one of the two possible orderings.
(We can — and now do — test this in a brute-force way by copy/pasting
the test with both typevar orderings.)
The underlying bug was in our `ConstraintSet::simplify_and_domain`
method. It would correctly detect `(U ≤ T ≤ U) ∧ (U ≤ int)`, because
those two constraints affect different typevars, and from that, infer `T
≤ int`. But it wouldn't detect the equivalent pattern in `(T ≤ U ≤ T) ∧
(U ≤ int)`, since those constraints affect the same typevar. At first I
tried adding that as yet more pattern-match logic in the ever-growing
`simplify_and_domain` method. But doing so caused other tests to start
failing.
At that point, I realized that `simplify_and_domain` had gotten to the
point where it was trying to do too much, and for conflicting consumers.
It was first written as part of our display logic, where the goal is to
remove redundant information from a BDD to make its string rendering
simpler. But we also started using it to add "derived facts" to a BDD. A
derived fact is a constraint that doesn't appear in the BDD directly,
but which we can still infer to be true. Our failing test relies on
derived facts — being able to infer that `T ≤ int` even though that
particular constraint doesn't appear in the original BDD. Before,
`simplify_and_domain` would trace through all of the constraints in a
BDD, figure out the full set of derived facts, and _add those derived
facts_ to the BDD structure. This is brittle, because those derived
facts are not universally true! In our example, `T ≤ int` only holds
along the BDD paths where both `T = U` and `U ≤ int`. Other paths will
test the negations of those constraints, and on those, we _shouldn't_
infer `T ≤ int`. In theory it's possible (and we were trying) to use BDD
operators to express that dependency...but that runs afoul of how we
were simultaneously trying to _remove_ information to make our displays
simpler.
So, I ripped off the band-aid. `simplify_and_domain` is now _only_ used
for display purposes. I have not touched it at all, except to remove
some logic that is definitely not used by our `Display` impl. Otherwise,
I did not want to touch that house of cards for now, since the display
logic is not load-bearing for any type inference logic.
For all non-display callers, we have a new **_sequent map_** data type,
which tracks exactly the same derived information. But it does so (a)
without trying to remove anything from the BDD, and (b) lazily, without
updating the BDD structure.
So the end result is that all of the tests (including the new
regressions) pass, via a more efficient (and hopefully better
structured/documented) implementation, at the cost of hanging onto a
pile of display-related tech debt that we'll want to clean up at some
point.
## Summary
This PR proposes that we add a new `set_concise_message` functionality
to our `Diagnostic` construction API. When used, the concise message
that is otherwise auto-generated from the main diagnostic message and
the primary annotation will be overwritten with the custom message.
To understand why this is desirable, let's look at the `invalid-key`
diagnostic. This is how I *want* the full diagnostic to look like:
<img width="620" height="282" alt="image"
src="https://github.com/user-attachments/assets/3bf70f52-9d9f-4817-bc16-fb0ebf7c2113"
/>
However, without the change in this PR, the concise message would have
the following form:
```
error[invalid-key]: Unknown key "Age" for TypedDict `Person`: Unknown key "Age" - did you mean "age"?
```
This duplication is why the full `invalid-key` diagnostic used a main
diagnostic message that is only "Invalid key for TypedDict `Person`", to
make that bearable:
```
error[invalid-key] Invalid key for TypedDict `Person`: Unknown key "Age" - did you mean "age"?
```
This is still less than ideal, *and* we had to make the "full"
diagnostic worse. With the new API here, we have to make no such
compromises. We need to do slightly more work (provide one additional
custom-designed message), but we get to keep the "full" diagnostic that
we actually want, and we can make the concise message more terse and
readable:
```
error[invalid-key] Unknown key "Age" for TypedDict `Person` - did you mean "age"?
```
Similar problems exist for other diagnostics as well (I really want this
for https://github.com/astral-sh/ruff/pull/21476). In this PR, I only
changed `invalid-key` and `type-assertion-failure`.
The PR here is somewhat related to the discussion in
https://github.com/astral-sh/ty/issues/1418, but note that we are
solving a problem that is unrelated to sub-diagnostics.
## Test Plan
Updated tests
## Summary
Add support for `Callable` special forms in implicit type aliases.
## Typing conformance
Four new tests are passing
## Ecosystem impact
* All of the `invalid-type-form` errors are from libraries that use
`mypy_extensions` and do something like `Callable[[NamedArg("x", str)],
int]`.
* A handful of new false positives because we do not support generic
specializations of implicit type aliases, yet. But other
* Everything else looks like true positives or known limitations
## Test Plan
New Markdown tests.
Constraint sets can now track subtyping/assignability/etc of generic
callables correctly. For instance:
```py
def identity[T](t: T) -> T:
return t
constraints = ConstraintSet.always()
static_assert(constraints.implies_subtype_of(TypeOf[identity], Callable[[int], int]))
static_assert(constraints.implies_subtype_of(TypeOf[identity], Callable[[str], str]))
```
A generic callable can be considered an intersection of all of its
possible specializations, and an assignability check with an
intersection as the lhs side succeeds of _any_ of the intersected types
satisfies the check. Put another way, if someone expects to receive any
function with a signature of `(int) -> int`, we can give them
`identity`.
Note that the corresponding check using `is_subtype_of` directly does
not yet work, since #20093 has not yet hooked up the core typing
relationship logic to use constraint sets:
```py
# These currently fail
static_assert(is_subtype_of(TypeOf[identity], Callable[[int], int]))
static_assert(is_subtype_of(TypeOf[identity], Callable[[str], str]))
```
To do this, we add a new _existential quantification_ operation on
constraint sets. This takes in a list of typevars and _removes_ those
typevars from the constraint set. Conceptually, we return a new
constraint set that evaluates to `true` when there was _any_ assignment
of the removed typevars that caused the old constraint set to evaluate
to `true`.
When comparing a generic constraint set, we add its typevars to the
`inferable` set, and figure out whatever constraints would allow any
specialization to satisfy the check. We then use the new existential
quantification operator to remove those new typevars, since the caller
doesn't (and shouldn't) know anything about them.
---------
Co-authored-by: David Peter <sharkdp@users.noreply.github.com>
## Summary
Follow up from https://github.com/astral-sh/ruff/pull/21411. Again,
there are more things that could be improved here (like the diagnostics
for `lists`, or extending what we have for `dict` to `OrderedDict` etc),
but that will have to be postponed.
## Summary
We previously only allowed models to overwrite the
`{eq,order,kw_only,frozen}_defaults` of the dataclass-transformer, but
all other standard-dataclass parameters should be equally supported with
the same behavior.
## Test Plan
Added regression tests.
## Summary
Not a high-priority task... but it _is_ a weekend :P
This PR improves our diagnostics for invalid exceptions. Specifically:
- We now give a special-cased ``help: Did you mean
`NotImplementedError`` subdiagnostic for `except NotImplemented`, `raise
NotImplemented` and `raise <EXCEPTION> from NotImplemented`
- If the user catches a tuple of exceptions (`except (foo, bar, baz):`)
and multiple elements in the tuple are invalid, we now collect these
into a single diagnostic rather than emitting a separate diagnostic for
each tuple element
- The explanation of why the `except`/`raise` was invalid ("must be a
`BaseException` instance or `BaseException` subclass", etc.) is
relegated to a subdiagnostic. This makes the top-level diagnostic
summary much more concise.
## Test Plan
Lots of snapshots. And here's some screenshots:
<details>
<summary>Screenshots</summary>
<img width="1770" height="1520" alt="image"
src="https://github.com/user-attachments/assets/7f27fd61-c74d-4ddf-ad97-ea4fd24d06fd"
/>
<img width="1916" height="1392" alt="image"
src="https://github.com/user-attachments/assets/83e5027c-8798-48a6-a0ec-1babfc134000"
/>
<img width="1696" height="588" alt="image"
src="https://github.com/user-attachments/assets/1bc16048-6eb4-4dfa-9ace-dd271074530f"
/>
</details>
## Summary
Allow metaclass-based and baseclass-based dataclass-transformers to
overwrite the default behavior using class arguments:
```py
class Person(Model, order=True):
# ...
```
## Conformance tests
Four new tests passing!
## Test Plan
New Markdown tests
This PR updates the constraint implication type relationship to work on
compound types as well. (A compound type is a non-atomic type, like
`list[T]`.)
The goal of constraint implication is to check whether the requirements
of a constraint imply that a particular subtyping relationship holds.
Before, we were only checking atomic typevars. That would let us verify
that the constraint set `T ≤ bool` implies that `T` is always a subtype
of `int`. (In this case, the lhs of the subtyping check, `T`, is an
atomic typevar.)
But we weren't recursing into compound types, to look for nested
occurrences of typevars. That means that we weren't able to see that `T
≤ bool` implies that `Covariant[T]` is always a subtype of
`Covariant[int]`.
Doing this recursion means that we have to carry the constraint set
along with us as we recurse into types as part of `has_relation_to`, by
adding constraint implication as a new `TypeRelation` variant. (Before
it was just a method on `ConstraintSet`.)
---------
Co-authored-by: David Peter <sharkdp@users.noreply.github.com>
## Summary
Currently our diagnostic only covers the range of the thing being
subscripted:
<img width="1702" height="312" alt="image"
src="https://github.com/user-attachments/assets/7e630431-e846-46ca-93c1-139f11aaba11"
/>
But it should probably cover the _whole_ subscript expression (arguably
the more "incorrect" bit is the `["foo"]` part of this expression, not
the `x` part of this expression!)
## Test Plan
Added a snapshot
Co-authored-by: Brent Westbrook
<36778786+ntBre@users.noreply.github.com>
## Summary
Extends literal promotion to apply to any generic method, as opposed to
only generic class constructors. This PR also improves our literal
promotion heuristics to only promote literals in non-covariant position
in the return type, and avoid promotion if the literal is present in
non-covariant position in any argument type.
Resolves https://github.com/astral-sh/ty/issues/1357.
## Summary
- Always restore the previous `deferred_state` after parsing a type
expression: we don't want that state leaking out into other contexts
where we shouldn't be deferring expression inference
- Always defer the right-hand-side of a PEP-613 type alias in a stub
file, allowing for forward references on the right-hand side of `T:
TypeAlias = X | Y` in a stub file
Addresses @carljm's review in
https://github.com/astral-sh/ruff/pull/21401#discussion_r2524260153
## Test Plan
I added a regression test for a regression that the first version of
this PR introduced (we need to make sure the r.h.s. of a PEP-613
`TypeAlias`es is always deferred in a stub file)
## Summary
We currently fail to account for the type context when inferring generic
classes constructed with `__new__`, or synthesized `__init__` for
dataclasses.
## Summary
Infer the first argument `type` inside `Annotated[type, …]` as a type
expression. This allows us to support stringified annotations inside
`Annotated`.
## Ecosystem
* The removed diagnostic on `prefect` shows that we now understand the
`State.data` type annotation in
`src/prefect/client/schemas/objects.py:230`, which uses a stringified
annotation in `Annoated`. The other diagnostics are downstream changes
that result from this, it seems to be a commonly used data type.
* `artigraph` does something like `Annotated[cast(Any,
field_info.annotation), *field_info.metadata]` which I'm not sure we
need to allow? It's unfortunate since this is probably supported at
runtime, but it seems reasonable that they need to add a `# type:
ignore` for that.
* `pydantic` uses something like `Annotated[(self.annotation,
*self.metadata)]` but adds a `# type: ignore`
## Test Plan
New Markdown test
## Summary
Typeshed has a (fake) `__getattr__` method on `types.ModuleType` with a
return type of `Any`. We ignore this method when accessing attributes on
module *literals*, but with this PR, we respect this method when dealing
with `ModuleType` itself. That is, we allow arbitrary attribute accesses
on instances of `types.ModuleType`. This is useful because dynamic
import mechanisms such as `importlib.import_module` use `ModuleType` as
a return type.
closes https://github.com/astral-sh/ty/issues/1346
## Ecosystem
Massive reduction in diagnostics. The few new diagnostics are true
positives.
## Test Plan
Added regression test.
## Summary
Add synthetic members to completions on dataclasses and dataclass
instances.
Also, while we're at it, add support for `__weakref__` and
`__match_args__`.
closes https://github.com/astral-sh/ty/issues/1542
## Test Plan
New Markdown tests
## Summary
Support various legacy `typing` special forms (`List`, `Dict`, …) in
implicit type aliases.
## Ecosystem impact
A lot of true positives (e.g. on `alerta`)!
## Test Plan
New Markdown tests
## Summary
Support `type[…]` in implicit type aliases, for example:
```py
SubclassOfInt = type[int]
reveal_type(SubclassOfInt) # GenericAlias
def _(subclass_of_int: SubclassOfInt):
reveal_type(subclass_of_int) # type[int]
```
part of https://github.com/astral-sh/ty/issues/221
## Typing conformance
```diff
-specialtypes_type.py:138:5: error[type-assertion-failure] Argument does not have asserted type `type[Any]`
-specialtypes_type.py:140:5: error[type-assertion-failure] Argument does not have asserted type `type[Any]`
```
Two new tests passing ✔️
```diff
-specialtypes_type.py:146:1: error[unresolved-attribute] Object of type `GenericAlias` has no attribute `unknown`
```
An `TA4.unknown` attribute on a PEP 613 alias (`TA4: TypeAlias =
type[Any]`) is being accessed, and the conformance suite expects this to
be an error. Since we currently use the inferred type for these type
aliases (and possibly in the future as well), we treat this as a direct
access of the attribute on `type[Any]`, which falls back to an access on
`Any` itself, which succeeds. 🔴
```
+specialtypes_type.py:152:16: error[invalid-type-form] `typing.TypeVar` is not a generic class
+specialtypes_type.py:156:16: error[invalid-type-form] `typing.TypeVar` is not a generic class
```
New errors because we don't handle `T = TypeVar("T"); MyType = type[T];
MyType[T]` yet. Support for this is being tracked in
https://github.com/astral-sh/ty/issues/221🔴
## Ecosystem impact
Looks mostly good, a few known problems.
## Test Plan
New Markdown tests
## Summary
Further improve subscript assignment diagnostics, especially for
`dict`s:
```py
config: dict[str, int] = {}
config["retries"] = "three"
```
<img width="1276" height="274" alt="image"
src="https://github.com/user-attachments/assets/9762c733-8d1c-4a57-8c8a-99825071dc7d"
/>
I have many more ideas, but this looks like a reasonable first step.
Thank you @AlexWaygood for some of the suggestions here.
## Test Plan
Update tests
## Summary
We synthesize a (potentially large) set of `__setitem__` overloads for
every item in a `TypedDict`. Previously, validation of subscript
assignments on `TypedDict`s relied on actually calling `__setitem__`
with the provided key and value types, which implied that we needed to
do the full overload call evaluation for this large set of overloads.
This PR improves the performance of subscript assignment checks on
`TypedDict`s by validating the assignment directly instead of calling
`__setitem__`.
This PR also adds better handling for assignments to subscripts on union
and intersection types (but does not attempt to make it perfect). It
achieves this by distributing the check over unions and intersections,
instead of calling `__setitem__` on the union/intersection directly. We
already do something similar when validating *attribute* assignments.
## Ecosystem impact
* A lot of diagnostics change their rule type, and/or split into
multiple diagnostics. The new version is more verbose, but easier to
understand, in my opinion
* Almost all of the invalid-key diagnostics come from pydantic, and they
should all go away (including many more) when we implement
https://github.com/astral-sh/ty/issues/1479
* Everything else looks correct to me. There may be some new diagnostics
due to the fact that we now check intersections.
## Test Plan
New Markdown tests.
## Summary
cf. https://github.com/astral-sh/ruff/pull/20962
In the following code, `foo` in the comprehension was not reported as
unresolved:
```python
# error: [unresolved-reference] "Name `foo` used when not defined"
foo
foo = [
# no error!
# revealed: Divergent
reveal_type(x) for _ in () for x in [foo]
]
baz = [
# error: [unresolved-reference] "Name `baz` used when not defined"
# revealed: Unknown
reveal_type(x) for _ in () for x in [baz]
]
```
In fact, this is a more serious bug than it looks: for `foo`,
[`explicit_global_symbol` is
called](6cc3393ccd/crates/ty_python_semantic/src/types/infer/builder.rs (L8052)),
causing a symbol that should actually be `Undefined` to be reported as
being of type `Divergent`.
This PR fixes this bug. As a result, the code in
`mdtest/regression/pr_20962_comprehension_panics.md` no longer panics.
## Test Plan
`corpus\cyclic_symbol_in_comprehension.py` is added.
New tests are added in `mdtest/comprehensions/basic.md`.
---------
Co-authored-by: Micha Reiser <micha@reiser.io>
Co-authored-by: Carl Meyer <carl@astral.sh>
## Summary
Add (snapshot) tests for subscript assignment diagnostics. This is
mainly intended to establish a baseline before I hope to improve some of
these messages.
## Summary
Add support for `typing.Union` in implicit type aliases / in value
position.
## Typing conformance tests
Two new tests are passing
## Ecosystem impact
* The 2k new `invalid-key` diagnostics on pydantic are caused by
https://github.com/astral-sh/ty/issues/1479#issuecomment-3513854645.
* Everything else I've checked is either a known limitation (often
related to type narrowing, because union types are often narrowed down
to a subset of options), or a true positive.
## Test Plan
New Markdown tests
## Summary
Fix https://github.com/astral-sh/ty/issues/664
This PR adds support for storing attributes in comprehension scopes (any
eager scope.)
For example in the following code we infer type of `z` correctly:
```py
class C:
def __init__(self):
[None for self.z in range(1)]
reveal_type(C().z) # previously [unresolved-attribute] but now shows Unknown | int
```
The fix works by adjusting the following logics:
To identify if an attriute is an assignment to self or cls we need to
check the scope is a method. To allow comprehension scopes here we skip
any eager scope in the check.
Also at this stage the code checks if self or the first method argument
is shadowed by another binding that eager scope to prevent this:
```py
class D:
g: int
class C:
def __init__(self):
[[None for self.g in range(1)] for self in [D()]]
reveal_type(C().g) # [unresolved-attribute]
```
When determining scopes that attributes might be defined after
collecting all the methods of the class the code also returns any
decendant scope that is eager and only has eager parents until the
method scope.
When checking reachability of a attribute definition if the attribute is
defined in an eager scope we use the reachability of the first non eager
scope which must be a method. This allows attributes to be marked as
reachable and be seen.
There are also which I didn't add support for:
```py
class C:
def __init__(self):
def f():
[None for self.z in range(1)]
f()
reveal_type(C().z) # [unresolved-attribute]
```
In the above example we will not even return the comprehension scope as
an attribute scope because there is a non eager scope (`f` function)
between the comprehension and the `__init__` method
---------
Co-authored-by: Carl Meyer <carl@astral.sh>
## Summary
Fixes https://github.com/astral-sh/ty/issues/1409
This PR allows `Final` instance attributes to be initialized in
`__init__` methods, as mandated by the Python typing specification (PEP
591). Previously, ty incorrectly prevented this initialization, causing
false positive errors.
The fix checks if we're inside an `__init__` method before rejecting
Final attribute assignments, allowing assignments during
instance initialization while still preventing reassignment elsewhere.
## Test Plan
- Added new test coverage in `final.md` for the reported issue with
`Self` annotations
- Updated existing tests that were incorrectly expecting errors
- All 278 mdtest tests pass
- Manually tested with real-world code examples
---------
Co-authored-by: Carl Meyer <carl@astral.sh>
Fixes https://github.com/astral-sh/ty/issues/1487
This one is a true extension of non-standard semantics, and is therefore
a certified Hot Take we might conclude is simply a Bad Take (let's see
what ecosystem tests say...).
By resolving `.` and the LHS of the from import during semantic
indexing, we can check if the LHS is a submodule of `.`, and handle
`from whatever.thispackage.x.y import z` exactly like we do `from .x.y
import z`.
Fixes https://github.com/astral-sh/ty/issues/1484
This manifested as an error when inferring the type of a PEP-695 generic
class via its constructor parameters:
```py
class D[T, U]:
@overload
def __init__(self: "D[str, U]", u: U) -> None: ...
@overload
def __init__(self, t: T, u: U) -> None: ...
def __init__(self, *args) -> None: ...
# revealed: D[Unknown, str]
# SHOULD BE: D[str, str]
reveal_type(D("string"))
```
This manifested because `D` is inferred to be bivariant in both `T` and
`U`. We weren't seeing this in the equivalent example for legacy
typevars, since those default to invariant. (This issue also showed up
for _covariant_ typevars, so this issue was not limited to bivariance.)
The underlying cause was because of a heuristic that we have in our
current constraint solver, which attempts to handle situations like
this:
```py
def f[T](t: T | None): ...
f(None)
```
Here, the `None` argument matches the non-typevar union element, so this
argument should not add any constraints on what `T` can specialize to.
Our previous heuristic would check for this by seeing if the argument
type is a subtype of the parameter annotation as a whole — even if it
isn't a union! That would cause us to erroneously ignore the `self`
parameter in our constructor call, since bivariant classes are
equivalent to each other, regardless of their specializations.
The quick fix is to move this heuristic "down a level", so that we only
apply it when the parameter annotation is a union. This heuristic should
go away completely 🤞 with the new constraint solver.
This loses any ability to have "per-function" implicit submodule
imports, to avoid the "ok but now we need per-scope imports" and "ok but
this should actually introduce a global that only exists during this
function" problems. A simple and clean implementation with no weird
corners.
Fixes https://github.com/astral-sh/ty/issues/1482
This rips out the previous implementation in favour of a new
implementation with 3 rules:
- **froms are locals**: a `from..import` can only define locals, it does
not have global
side-effects. Specifically any submodule attribute `a` that's implicitly
introduced by either
`from .a import b` or `from . import a as b` (in an `__init__.py(i)`) is
a local and not a
global. If you do such an import at the top of a file you won't notice
this. However if you do
such an import in a function, that means it will only be function-scoped
(so you'll need to do
it in every function that wants to access it, making your code less
sensitive to execution
order).
- **first from first serve**: only the *first* `from..import` in an
`__init__.py(i)` that imports a
particular direct submodule of the current package introduces that
submodule as a local.
Subsequent imports of the submodule will not introduce that local. This
reflects the fact that
in actual python only the first import of a submodule (in the entire
execution of the program)
introduces it as an attribute of the package. By "first" we mean "the
first time in this scope
(or any parent scope)". This pairs well with the fact that we are
specifically introducing a
local (as long as you don't accidentally shadow or overwrite the local).
- **dot re-exports**: `from . import a` in an `__init__.pyi` is
considered a re-export of `a`
(equivalent to `from . import a as a`). This is required to properly
handle many stubs in the
wild. Currently it must be *exactly* `from . import ...`.
This implementation is intentionally limited/conservative (notably,
often requiring a from import to be relative). I'm going to file a ton
of followups for improvements so that their impact can be evaluated
separately.
Fixes https://github.com/astral-sh/ty/issues/133
## Summary
Detect usages of implicit `self` in property getters, which allows us to
treat their signature as being generic.
closes https://github.com/astral-sh/ty/issues/1502
## Typing conformance
Two new type assertions that are succeeding.
## Ecosystem results
Mostly look good. There are a few new false positives related to a bug
with constrained typevars that is unrelated to the work here. I reported
this as https://github.com/astral-sh/ty/issues/1503.
## Test Plan
Added regression tests.
## Summary
Add support for `Optional` and `Annotated` in implicit type aliases
part of https://github.com/astral-sh/ty/issues/221
## Typing conformance changes
New expected diagnostics.
## Ecosystem
A lot of true positives, some known limitations unrelated to this PR.
## Test Plan
New Markdown tests
## Summary
This PR adds extra validation for `isinstance()` and `issubclass()`
calls that use `UnionType` instances for their second argument.
According to typeshed's annotations, any `UnionType` is accepted for the
second argument, but this isn't true at runtime: at runtime, all
elements in the `UnionType` must either be class objects or be `None` in
order for the `isinstance()` or `issubclass()` call to reliably succeed:
```pycon
% uvx python3.14
Python 3.14.0 (main, Oct 10 2025, 12:54:13) [Clang 20.1.4 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from typing import LiteralString
>>> import types
>>> type(LiteralString | int) is types.UnionType
True
>>> isinstance(42, LiteralString | int)
Traceback (most recent call last):
File "<python-input-5>", line 1, in <module>
isinstance(42, LiteralString | int)
~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Users/alexw/Library/Application Support/uv/python/cpython-3.14.0-macos-aarch64-none/lib/python3.14/typing.py", line 559, in __instancecheck__
raise TypeError(f"{self} cannot be used with isinstance()")
TypeError: typing.LiteralString cannot be used with isinstance()
```
## Test Plan
Added mdtests/snapshots
When checking whether a constraint set is satisfied, if a typevar has a
non-fully-static upper bound or constraint, we are free to choose any
materialization that makes the check succeed.
In non-inferable positions, we have to show that the constraint set is
satisfied for all valid specializations, so it's best to choose the most
restrictive materialization, since that minimizes the set of valid
specializations that have to pass.
In inferable positions, we only have to show that the constraint set is
satisfied for _some_ valid specializations, so it's best to choose the
most permissive materialization, since that maximizes our chances of
finding a specialization that passes.
## Summary
Add support for `Literal` types in implicit type aliases.
part of https://github.com/astral-sh/ty/issues/221
## Ecosystem analysis
This looks good to me, true positives and known problems.
## Test Plan
New Markdown tests.
## Summary
This PR adds support for understanding the legacy definition and PEP 695
definition for `ParamSpec`.
This is still very initial and doesn't really implement any of the
semantics.
Part of https://github.com/astral-sh/ty/issues/157
## Test Plan
Add mdtest cases.
## Ecosystem analysis
Most of the diagnostics in `starlette` are due to the fact that ty now
understands `ParamSpec` is not a `Todo` type, so the assignability check
fails. The code looks something like:
```py
class _MiddlewareFactory(Protocol[P]):
def __call__(self, app: ASGIApp, /, *args: P.args, **kwargs: P.kwargs) -> ASGIApp: ... # pragma: no cover
class Middleware:
def __init__(
self,
cls: _MiddlewareFactory[P],
*args: P.args,
**kwargs: P.kwargs,
) -> None:
self.cls = cls
self.args = args
self.kwargs = kwargs
# ty complains that `ServerErrorMiddleware` is not assignable to `_MiddlewareFactory[P]`
Middleware(ServerErrorMiddleware, handler=error_handler, debug=debug)
```
There are multiple diagnostics where there's an attribute access on the
`Wrapped` object of `functools` which Pyright also raises:
```py
from functools import wraps
def my_decorator(f):
@wraps(f)
def wrapper(*args, **kwds):
return f(*args, **kwds)
# Pyright: Cannot access attribute "__signature__" for class "_Wrapped[..., Unknown, ..., Unknown]"
Attribute "__signature__" is unknown [reportAttributeAccessIssue]
# ty: Object of type `_Wrapped[Unknown, Unknown, Unknown, Unknown]` has no attribute `__signature__` [unresolved-attribute]
wrapper.__signature__
return wrapper
```
There are additional diagnostics that is due to the assignability checks
failing because ty now infers the `ParamSpec` instead of using the
`Todo` type which would always succeed. This results in a few
`no-matching-overload` diagnostics because the assignability checks
fail.
There are a few diagnostics related to
https://github.com/astral-sh/ty/issues/491 where there's a variable
which is either a bound method or a variable that's annotated with
`Callable` that doesn't contain the instance as the first parameter.
Another set of (valid) diagnostics are where the code hasn't provided
all the type variables. ty is now raising diagnostics for these because
we include `ParamSpec` type variable in the signature. For example,
`staticmethod[Any]` which contains two type variables.
This PR carries over some of the `has_relation_to` logic for comparing a
typevar with itself. A typevar will specialize to the same type if it's
mentioned multiple times, so it is always assignable to and a subtype of
itself. (Note that typevars can only specialize to fully static types.)
This is also true when the typevar appears in a union on the right-hand
side, or in an intersection on the left-hand side. Similarly, a typevar
is always disjoint from its negation, so when a negated typevar appears
on the left-hand side, the constraint set is never satisfiable.
(Eventually this will allow us to remove the corresponding clauses from
`has_relation_to`, but that can't happen until more of #20093 lands.)
## Summary
Splitting this one out from https://github.com/astral-sh/ruff/pull/21210. This is also something that should be made obselete by the new constraint solver, but is easy enough to fix now.
## Summary
Allow values of type `None` in type expressions. The [typing
spec](https://typing.python.org/en/latest/spec/annotations.html#type-and-annotation-expressions)
could be more explicit on whether this is actually allowed or not, but
it seems relatively harmless and does help in some use cases like:
```py
try:
from module import MyClass
except ImportError:
MyClass = None # ty: ignore
def f(m: MyClass):
pass
```
## Test Plan
Updated tests, ecosystem check.
## Summary
A lot of the bidirectional inference work relies on `dict` not being
assignable to `TypedDict`, so I think it makes sense to add this before
fully implementing https://github.com/astral-sh/ty/issues/1387.
## Summary
Add support for implicit type aliases that use PEP 604 unions:
```py
IntOrStr = int | str
reveal_type(IntOrStr) # UnionType
def _(int_or_str: IntOrStr):
reveal_type(int_or_str) # int | str
```
## Typing conformance
The changes are either removed false positives, or new diagnostics due
to known limitations unrelated to this PR.
## Ecosystem impact
Spot checked, a mix of true positives and known limitations.
## Test Plan
New Markdown tests.
Fixes https://github.com/astral-sh/ty/issues/1053
## Summary
Other type checkers prioritize a submodule over a package `__getattr__`
in `from mod import sub`, even though the runtime precedence is the
other direction. In effect, this is making an implicit assumption that a
module `__getattr__` will not handle (that is, will raise
`AttributeError`) for names that are also actual submodules, rather than
shadowing them. In practice this seems like a realistic assumption in
the ecosystem? Or at least the ecosystem has adapted to it, and we need
to adapt this precedence also, for ecosystem compatibility.
The implementation is a bit ugly, precisely because it departs from the
runtime semantics, and our implementation is oriented toward modeling
runtime semantics accurately. That is, `__getattr__` is modeled within
the member-lookup code, so it's hard to split "member lookup result from
module `__getattr__`" apart from other member lookup results. I did this
via a synthetic `TypeQualifier::FROM_MODULE_GETATTR` that we attach to a
type resulting from a member lookup, which isn't beautiful but it works
well and doesn't introduce inefficiency (e.g. redundant member lookups).
## Test Plan
Updated mdtests.
Also added a related mdtest formalizing our support for a module
`__getattr__` that is explicitly annotated to accept a limited set of
names. In principle this could be an alternative (more explicit) way to
handle the precedence problem without departing from runtime semantics,
if the ecosystem would adopt it.
### Ecosystem analysis
Lots of removed diagnostics which are an improvement because we now
infer the expected submodule.
Added diagnostics are mostly unrelated issues surfaced now because we
previously had an earlier attribute error resulting in `Unknown`; now we
correctly resolve the module so that earlier attribute error goes away,
we get an actual type instead of `Unknown`, and that triggers a new
error.
In scipy and sklearn, the module `__getattr__` which we were respecting
previously is un-annotated so returned a forgiving `Unknown`; now we
correctly see the actual module, which reveals some cases of
https://github.com/astral-sh/ty/issues/133 that were previously hidden
(`scipy/optimize/__init__.py` [imports `from
._tnc`](eff82ca575/scipy/optimize/__init__.py (L429)).)
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Fixes https://github.com/astral-sh/ty/issues/1368
## Summary
Add support for patterns like this, where a type alias to a literal type
(or union of literal types) is used to subscript `typing.Literal`:
```py
type MyAlias = Literal[1]
def _(x: Literal[MyAlias]): ...
```
This shows up in the ecosystem report for PEP 613 type alias support.
One interesting case is an alias to `bool` or an enum type. `bool` is an
equivalent type to `Literal[True, False]`, which is a union of literal
types. Similarly an enum type `E` is also equivalent to a union of its
member literal types. Since (for explicit type aliases) we infer the RHS
directly as a type expression, this makes it difficult for us to
distinguish between `bool` and `Literal[True, False]`, so we allow
either one to (or an alias to either one) to appear inside `Literal`,
where other type checkers allow only the latter.
I think for implicit type aliases it may be simpler to support only
types derived from actually subscripting `typing.Literal`, though, so I
didn't make a TODO-comment commitment here.
## Test Plan
Added mdtests, including TODO-filled tests for PEP 613 and implicit type
aliases.
### Conformance suite
All changes here are positive -- we now emit errors on lines that should
be errors. This is a side effect of the new implementation, not the
primary purpose of this PR, but it's still a positive change.
### Ecosystem
Eliminates one ecosystem false positive, where a PEP 695 type alias for
a union of literal types is used to subscript `typing.Literal`.
## Summary
Adds type inference for list/dict/set comprehensions, including
bidirectional inference:
```py
reveal_type({k: v for k, v in [("a", 1), ("b", 2)]}) # dict[Unknown | str, Unknown | int]
squares: list[int | None] = [x for x in range(10)]
reveal_type(squares) # list[int | None]
```
## Ecosystem impact
I did spot check the changes and most of them seem like known
limitations or true positives. Without proper bidirectional inference,
we saw a lot of false positives.
## Test Plan
New Markdown tests
## Summary
Discussion with @ibraheemdev clarified that
https://github.com/astral-sh/ruff/pull/21168 was incorrect. In a case of
failed inference of a dict literal as a `TypedDict`, we should store the
context-less inferred type of the dict literal as the type of the dict
literal expression itself; the fallback to declared type should happen
at the level of the overall assignment definition.
The reason the latter isn't working yet is because currently we
(wrongly) consider a homogeneous dict type as assignable to a
`TypedDict`, so we don't actually consider the assignment itself as
failed. So the "bug" I observed (and tried to fix) will naturally be
fixed by implementing TypedDict assignability rules.
Rollback https://github.com/astral-sh/ruff/pull/21168 except for the
tests, and modify the tests to include TODOs as needed.
## Test Plan
Updated mdtests.
The parser currently uses single quotes to wrap tokens. This is
inconsistent with the rest of ruff/ty, which use backticks.
For example, see the inconsistent diagnostics produced in this simple
example: https://play.ty.dev/0a9d6eab-6599-4a1d-8e40-032091f7f50f
Consistently wrapping tokens in backticks produces uniform diagnostics.
Following the style decision of #723, in #2889 some quotes were already
switched into backticks.
This is also in line with Rust's guide on diagnostics
(https://rustc-dev-guide.rust-lang.org/diagnostics.html#diagnostic-structure):
> When code or an identifier must appear in a message or label, it
should be surrounded with backticks
## Summary
In general, when we have an invalid assignment (inferred assigned type
is not assignable to declared type), we fall back to inferring the
declared type, since the declared type is a more explicit declaration of
the programmer's intent. This also maintains the invariant that our
inferred type for a name is always assignable to the declared type for
that same name. For example:
```py
x: str = 1
reveal_type(x) # revealed: str
```
We weren't following this pattern for dictionary literals inferred (via
type context) as a typed dictionary; if the literal was not valid for
the annotated TypedDict type, we would just fall back to the normal
inferred type of the dict literal, effectively ignoring the annotation,
and resulting in inferred type not assignable to declared type.
## Test Plan
Added mdtest assertions.
## Summary
The solver is currently order-dependent, and will choose a supertype
over the exact type if it appears earlier in the list of constraints. We
could be smarter and try to choose the most precise subtype, but I
imagine this is something the new constraint solver will fix anyways,
and this fixes the issue showing up on
https://github.com/astral-sh/ruff/pull/21070.
This PR adds a new `satisfied_by_all_typevar` method, which implements
one of the final steps of actually using these dang constraint sets.
Constraint sets exist to help us check assignability and subtyping of
types in the presence of typevars. We construct a constraint set
describing the conditions under which assignability holds between the
two types. Then we check whether that constraint set is satisfied for
the valid specializations of the relevant typevars (which is this new
method).
We also add a new `ty_extensions.ConstraintSet` method so that we can
test this method's behavior in mdtests, before hooking it up to the rest
of the specialization inference machinery.
## Summary
We currently perform a subtyping check instead of the intended subclass
check (and the subtyping check is confusingly named `is_subclass_of`).
This showed up in https://github.com/astral-sh/ruff/pull/21070.
## Summary
Before this PR, we would emit diagnostics like "Invalid key access" for
a TypedDict literal with invalid key, which doesn't make sense since
there's no "access" in that case. This PR just adjusts the wording to be
more general, and adjusts the documentation of the lint rule too.
I noticed this in the playground and thought it would be a quick fix. As
usual, it turned out to be a bit more subtle than I expected, but for
now I chose to punt on the complexity. We may ultimately want to have
different rules for invalid subscript vs invalid TypedDict literal,
because an invalid key in a TypedDict literal is low severity: it's a
typo detector, but not actually a type error. But then there's another
wrinkle there: if the TypedDict is `closed=True`, then it _is_ a type
error. So would we want to separate the open and closed cases into
separate rules, too? I decided to leave this as a question for future.
If we wanted to use separate rules, or use specific wording for each
case instead of the generalized wording I chose here, that would also
involve a bit of extra work to distinguish the cases, since we use a
generic set of functions for reporting these errors.
## Test Plan
Added and updated mdtests.
This is a second take at the implicit imports approach, allowing `from .
import submodule` in an `__init__.pyi` to create the
`mypackage.submodule` attribute everyhere.
This implementation operates inside of the
available_submodule_attributes subsystem instead of as a re-export rule.
The upside of this is we are no longer purely syntactic, and absolute
from imports that happen to target submodules work (an intentional
discussed deviation from pyright which demands a relative from import).
Also we don't re-export functions or classes.
The downside(?) of this is star imports no longer see these attributes
(this may be either good or bad. I believe it's not a huge lift to make
it work with star imports but it's some non-trivial reworking).
I've also intentionally made `import mypackage.submodule` not trigger
this rule although it's trivial to change that.
I've tried to cover as many relevant cases as possible for discussion in
the new test file I've added (there are some random overlaps with
existing tests but trying to add them piecemeal felt confusing and
weird, so I just made a dedicated file for this extension to the rules).
Fixes https://github.com/astral-sh/ty/issues/133
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
<!-- What's the purpose of the change? What does it do, and why? -->
## Test Plan
<!-- How was it tested? -->
## Summary
Fixes https://github.com/astral-sh/ty/issues/1427
This PR fixes a regression introduced in alpha.24 where non-dataclass
children of generic dataclasses lost generic type parameter information
during `__init__` synthesis.
The issue occurred because when looking up inherited members in the MRO,
the child class's `inherited_generic_context` was correctly passed down,
but `own_synthesized_member()` (which synthesizes dataclass `__init__`
methods) didn't accept this parameter. It only used
`self.inherited_generic_context(db)`, which returned the parent's
context instead of the child's.
The fix threads the child's generic context through to the synthesis
logic, allowing proper generic type inference for inherited dataclass
constructors.
## Test Plan
- Added regression test for non-dataclass inheriting from generic
dataclass
- Verified the exact repro case from the issue now works
- All 277 mdtest tests passing
- Clippy clean
- Manually verified with Python runtime, mypy, and pyright - all accept
this code pattern
## Verification
Tested against multiple type checkers:
- ✅ Python runtime: Code works correctly
- ✅ mypy: No issues found
- ✅ pyright: 0 errors, 0 warnings
- ✅ ty alpha.23: Worked (before regression)
- ❌ ty alpha.24: Regression
- ✅ ty with this fix: Works correctly
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: David Peter <mail@david-peter.de>
It's possible for a constraint to mention two typevars. For instance, in
the body of
```py
def f[S: int, T: S](): ...
```
the baseline constraint set would be `(T ≤ S) ∧ (S ≤ int)`. That is, `S`
must specialize to some subtype of `int`, and `T` must specialize to a
subtype of the type that `S` specializes to.
This PR updates the new "constraint implication" relationship from
#21010 to work on these kinds of constraint sets. For instance, in the
example above, we should be able to see that `T ≤ int` must always hold:
```py
def f[S, T]():
constraints = ConstraintSet.range(Never, S, int) & ConstraintSet.range(Never, T, S)
static_assert(constraints.implies_subtype_of(T, int)) # now succeeds!
```
This did not require major changes to the implementation of
`implies_subtype_of`. That method already relies on how our `simplify`
and `domain` methods expand a constraint set to include the transitive
closure of the constraints that it mentions, and to mark certain
combinations of constraints as impossible. Previously, that transitive
closure logic only looked at pairs of constraints that constrain the
same typevar. (For instance, to notice that `(T ≤ bool) ∧ ¬(T ≤ int)` is
impossible.)
Now we also look at pairs of constraints that constraint different
typevars, if one of the constraints is bound by the other — that is,
pairs of the form `T ≤ S` and `S ≤ something`, or `S ≤ T` and `something
≤ S`. In those cases, transitivity lets us add a new derived constraint
that `T ≤ something` or `something ≤ T`, respectively. Having done that,
our existing `implies_subtype_of` logic finds and takes into account
that derived constraint.
## Summary
We weren't correctly modeling it as a `staticmethod` in all cases,
leading us to incorrectly infer that the `cls` argument would be bound
if it was accessed on an instance (rather than the class object).
## Test Plan
Added mdtests that fail on `main`. The primer output also looks good!
## Summary
Adds proper type narrowing and reachability analysis for matching on
non-inferable type variables bound to enums. For example:
```py
from enum import Enum
class Answer(Enum):
NO = 0
YES = 1
def is_yes(self) -> bool: # no error here!
match self:
case Answer.YES:
return True
case Answer.NO:
return False
```
closes https://github.com/astral-sh/ty/issues/1404
## Test Plan
Added regression tests
## Summary
We previously didn't understand `range` and wrote these custom
`IntIterable`/`IntIterator` classes for tests. We can now remove them
and make the tests shorter in some places.
## Summary
Infer a type of unannotated `self` parameters in decorated methods /
properties.
closes https://github.com/astral-sh/ty/issues/1448
## Test Plan
Existing tests, some new tests.
This PR updates the mdtests that test how our generics solver interacts
with our new constraint set implementation. Because the rendering of a
constraint set can get long, this standardizes on putting the `revealed`
assertion on a separate line. We also add a `static_assert` test for
each constraint set to verify that they are all coerced into simple
`bool`s correctly.
This is a pure reformatting (not even a refactoring!) that changes no
behavior. I've pulled it out of #20093 to reduce the amount of effort
that will be required to review that PR.
We have several functions in `ty_extensions` for testing our constraint
set implementation. This PR refactors those functions so that they are
all methods of the `ConstraintSet` class, rather than being standalone
top-level functions. 🎩 to @sharkdp for pointing out that
`KnownBoundMethod` gives us what we need to implement that!
This PR adds the new **_constraint implication_** relationship between
types, aka `is_subtype_of_given`, which tests whether one type is a
subtype of another _assuming that the constraints in a particular
constraint set hold_.
For concrete types, constraint implication is exactly the same as
subtyping. (A concrete type is any fully static type that is not a
typevar. It can _contain_ a typevar, though — `list[T]` is considered
concrete.)
The interesting case is typevars. The other typing relationships (TODO:
will) all "punt" on the question when considering a typevar, by
translating the desired relationship into a constraint set. At some
point, though, we need to resolve a constraint set; at that point, we
can no longer punt on the question. Unlike with concrete types, the
answer will depend on the constraint set that we are considering.
That PR title might be a bit inscrutable.
Consider the two constraints `T ≤ bool` and `T ≤ int`. Since `bool ≤
int`, by transitivity `T ≤ bool` implies `T ≤ int`. (Every type that is
a subtype of `bool` is necessarily also a subtype of `int`.) That means
that `T ≤ bool ∧ T ≰ int` is an impossible combination of constraints,
and is therefore not a valid input to any BDD. We say that that
assignment is not in the _domain_ of the BDD.
The implication `T ≤ bool → T ≤ int` can be rewritten as `T ≰ bool ∨ T ≤
int`. (That's the definition of implication.) If we construct that
constraint set in an mdtest, we should get a constraint set that is
always satisfiable. Previously, that constraint set would correctly
_display_ as `always`, but a `static_assert` on it would fail.
The underlying cause is that our `is_always_satisfied` method would only
test if the BDD was the `AlwaysTrue` terminal node. `T ≰ bool ∨ T ≤ int`
does not simplify that far, because we purposefully keep around those
constraints in the BDD structure so that it's easier to compare against
other BDDs that reference those constraints.
To fix this, we need a more nuanced definition of "always satisfied".
Instead of evaluating to `true` for _every_ input, we only need it to
evaluate to `true` for every _valid_ input — that is, every input in its
domain.
## Summary
Infer a type of `Self` for unannotated `self` parameters in methods of
classes.
part of https://github.com/astral-sh/ty/issues/159
closes https://github.com/astral-sh/ty/issues/1081
## Conformance tests changes
```diff
+enums_member_values.py:85:9: error[invalid-assignment] Object of type `int` is not assignable to attribute `_value_` of type `str`
```
A true positive ✔️
```diff
-generics_self_advanced.py:35:9: error[type-assertion-failure] Argument does not have asserted type `Self@method2`
-generics_self_basic.py:14:9: error[type-assertion-failure] Argument does not have asserted type `Self@set_scale
```
Two false positives going away ✔️
```diff
+generics_syntax_infer_variance.py:82:9: error[invalid-assignment] Cannot assign to final attribute `x` on type `Self@__init__`
```
This looks like a true positive to me, even if it's not marked with `#
E` ✔️
```diff
+protocols_explicit.py:56:9: error[invalid-assignment] Object of type `tuple[int, int, str]` is not assignable to attribute `rgb` of type `tuple[int, int, int]`
```
True positive ✔️
```
+protocols_explicit.py:85:9: error[invalid-attribute-access] Cannot assign to ClassVar `cm1` from an instance of type `Self@__init__`
```
This looks like a true positive to me, even if it's not marked with `#
E`. But this is consistent with our understanding of `ClassVar`, I
think. ✔️
```py
+qualifiers_final_annotation.py:52:9: error[invalid-assignment] Cannot assign to final attribute `ID4` on type `Self@__init__`
+qualifiers_final_annotation.py:65:9: error[invalid-assignment] Cannot assign to final attribute `ID7` on type `Self@method1`
```
New true positives ✔️
```py
+qualifiers_final_annotation.py:52:9: error[invalid-assignment] Cannot assign to final attribute `ID4` on type `Self@__init__`
+qualifiers_final_annotation.py:57:13: error[invalid-assignment] Cannot assign to final attribute `ID6` on type `Self@__init__`
+qualifiers_final_annotation.py:59:13: error[invalid-assignment] Cannot assign to final attribute `ID6` on type `Self@__init__`
```
This is a new false positive, but that's a pre-existing issue on main
(if you annotate with `Self`):
https://play.ty.dev/3ee1c56d-7e13-43bb-811a-7a81e236e6ab❌ => reported
as https://github.com/astral-sh/ty/issues/1409
## Ecosystem
* There are 5931 new `unresolved-attribute` and 3292 new
`possibly-missing-attribute` attribute errors, way too many to look at
all of them. I randomly sampled 15 of these errors and found:
* 13 instances where there was simply no such attribute that we could
plausibly see. Sometimes [I didn't find it
anywhere](8644d886c6/openlibrary/plugins/openlibrary/tests/test_listapi.py (L33)).
Sometimes it was set externally on the object. Sometimes there was some
[`setattr` dynamicness going
on](a49f6b927d/setuptools/wheel.py (L88-L94)).
I would consider all of them to be true positives.
* 1 instance where [attribute was set on `obj` in
`__new__`](9e87b44fd4/sympy/tensor/array/array_comprehension.py (L45C1-L45C36)),
which we don't support yet
* 1 instance [where the attribute was defined via `__slots__`
](e250ec0fc8/lib/spack/spack/vendor/pyrsistent/_pdeque.py (L48C5-L48C14))
* I see 44 instances [of the false positive
above](https://github.com/astral-sh/ty/issues/1409) with `Final`
instance attributes being set in `__init__`. I don't think this should
block this PR.
## Test Plan
New Markdown tests.
---------
Co-authored-by: Shaygan Hooshyari <sh.hooshyari@gmail.com>
This PR adds another useful simplification when rendering constraint
sets: `T = int` instead of `T = int ∧ T ≠ str`. (The "smaller"
constraint `T = int` implies the "larger" constraint `T ≠ str`.
Constraint set clauses are intersections, and if one constraint in a
clause implies another, we can throw away the "larger" constraint.)
While we're here, we also normalize the bounds of a constraint, so that
we equate e.g. `T ≤ int | str` with `T ≤ str | int`, and change the
ordering of BDD variables so that all constraints with the same typevar
are ordered adjacent to each other.
Lastly, we also add a new `display_graph` helper method that prints out
the full graph structure of a BDD.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
## Summary
Fall back to `C[Divergent]` if we are trying to specialize `C[T]` with a
type that itself already contains deeply nested specialized generic
classes. This is a way to prevent infinite recursion for cases like
`self.x = [self.x]` where type inference for the implicit instance
attribute would not converge.
closes https://github.com/astral-sh/ty/issues/1383
closes https://github.com/astral-sh/ty/issues/837
## Test Plan
Regression tests.
## Summary
We currently panic in the seemingly rare case where the type of a
default value of a parameter depends on the callable itself:
```py
class C:
def f(self: C):
self.x = lambda a=self.x: a
```
Types of default values are only used for display reasons, and it's
unclear if we even want to track them (or if we should rather track the
actual value). So it didn't seem to me that we should spend a lot of
effort (and runtime) trying to achieve a theoretically correct type here
(which would be infinite).
Instead, we simply replace *nested* default types with `Unknown`, i.e.
only if the type of the default value is a callable itself.
closes https://github.com/astral-sh/ty/issues/1402
## Test Plan
Regression tests
## Summary
Only run the "pull types" test after performing the "actual" mdtest. We
observed that the order matters. There is currently one mdtest which
panics when checked in the CLI or the playground. With this change, it
also panics in the mdtest suite.
reopens https://github.com/astral-sh/ty/issues/837?
## Summary
- Type checkers (and type-checker authors) think in terms of types, but
I think most Python users think in terms of values. Rather than saying
that a _type_ `X` "has no attribute `foo`" (which I think sounds strange
to many users), say that "an object of type `X` has no attribute `foo`"
- Special-case certain types so that the diagnostic messages read more
like normal English: rather than saying "Type `<class 'Foo'>` has no
attribute `bar`" or "Object of type `<class 'Foo'>` has no attribute
`bar`", just say "Class `Foo` has no attribute `bar`"
## Test Plan
Mdtests and snapshots updated
<!--
Thank you for contributing to Ruff/ty! To help us out with reviewing,
please consider the following:
- Does this pull request include a summary of the change? (See below.)
- Does this pull request include a descriptive title? (Please prefix
with `[ty]` for ty pull
requests.)
- Does this pull request include references to any relevant issues?
-->
## Summary
<!-- What's the purpose of the change? What does it do, and why? -->
This PR implements semantic syntax error where alternative patterns bind
different names
## Test Plan
<!-- How was it tested? -->
I have written inline tests as directed in #17412
---------
Signed-off-by: 11happy <soni5happy@gmail.com>
Co-authored-by: Brent Westbrook <brentrwestbrook@gmail.com>
## Summary
Derived from #20900
Implement `VarianceInferable` for `KnownInstanceType` (especially for
`KnownInstanceType::TypeAliasType`).
The variance of a type alias matches its value type. In normal usage,
type aliases are expanded to value types, so the variance of a type
alias can be obtained without implementing this. However, for example,
if we want to display the variance when hovering over a type alias, we
need to be able to obtain the variance of the type alias itself (cf.
#20900).
## Test Plan
I couldn't come up with a way to test this in mdtest, so I'm testing it
in a test submodule at the end of `types.rs`.
I also added a test to `mdtest/generics/pep695/variance.md`, but it
passes without the changes in this PR.
## Summary
Support `dataclass_transform` when used on a (base) class.
## Typing conformance
* The changes in `dataclasses_transform_class.py` look good, just a few
mistakes due to missing `alias` support.
* I didn't look closely at the changes in
`dataclasses_transform_converter.py` since we don't support `converter`
yet.
## Ecosystem impact
The impact looks huge, but it's concentrated on a single project (ibis).
Their setup looks more or less like this:
* the real `Annotatable`:
d7083c2c96/ibis/common/grounds.py (L100-L101)
* the real `DataType`:
d7083c2c96/ibis/expr/datatypes/core.py (L161-L179)
* the real `Array`:
d7083c2c96/ibis/expr/datatypes/core.py (L1003-L1006)
```py
from typing import dataclass_transform
@dataclass_transform()
class Annotatable:
pass
class DataType(Annotatable):
nullable: bool = True
class Array[T](DataType):
value_type: T
```
They expect something like `Array([1, 2])` to work, but ty, pyright,
mypy, and pyrefly would all expect there to be a first argument for the
`nullable` field on `DataType`. I don't really understand on what
grounds they expect the `nullable` field to be excluded from the
signature, but this seems to be the main reason for the new diagnostics
here. Not sure if related, but it looks like their typing setup is not
really complete
(https://github.com/ibis-project/ibis/issues/6844#issuecomment-1868274770,
this thread also mentions `dataclass_transform`).
## Test Plan
Update pre-existing tests.
Detect legacy namespace packages and treat them like namespace packages
when looking them up as the *parent* of the module we're interested in.
In all other cases treat them like a regular package.
(This PR is coauthored by @MichaReiser in a shared coding session)
Fixes https://github.com/astral-sh/ty/issues/838
---------
Co-authored-by: Micha Reiser <micha@reiser.io>
## Summary
Prefer the declared type for collection literals, e.g.,
```py
x: list[Any] = [1, "2", (3,)]
reveal_type(x) # list[Any]
```
This solves a large part of https://github.com/astral-sh/ty/issues/136
for invariant generics, where respecting the declared type is a lot more
important. It also means that annotated dict literals with `dict[_,
Any]` is a way out of https://github.com/astral-sh/ty/issues/1248.
## Summary
Use the declared type of variables as type context for the RHS of assignment expressions, e.g.,
```py
x: list[int | str]
x = [1]
reveal_type(x) # revealed: list[int | str]
```
## Summary
Ignore the type context when specializing a generic call if it leads to
an unnecessarily wide return type. For example, [the example mentioned
here](https://github.com/astral-sh/ruff/pull/20796#issuecomment-3403319536)
works as expected after this change:
```py
def id[T](x: T) -> T:
return x
def _(i: int):
x: int | None = id(i)
y: int | None = i
reveal_type(x) # revealed: int
reveal_type(y) # revealed: int
```
I also added extended our usage of `filter_disjoint_elements` to tuple
and typed-dict inference, which resolves
https://github.com/astral-sh/ty/issues/1266.
## Summary
Add support for the `field_specifiers` parameter on
`dataclass_transform` decorator calls.
closes https://github.com/astral-sh/ty/issues/1068
## Conformance test results
All true positives ✔️
## Ecosystem analysis
* `trio`: this is the kind of change that I would expect from this PR.
The code makes use of a dataclass `Outcome` with a `_unwrapped: bool =
attr.ib(default=False, eq=False, init=False)` field that is excluded
from the `__init__` signature, so we now see a bunch of
constructor-call-related errors going away.
* `home-assistant/core`: They have a `domain: str = attr.ib(init=False,
repr=False)` field and then use
```py
@domain.default
def _domain_default(self) -> str:
# …
```
This accesses the `default` attribute on `dataclasses.Field[…]` with a
type of `default: _T | Literal[_MISSING_TYPE.MISSING]`, so we get those
"Object of type `_MISSING_TYPE` is not callable" errors. I don't really
understand how that is supposed to work. Even if `_MISSING_TYPE` would
be absent from that union, what does this try to call? pyright also
issues an error and it doesn't seem to work at runtime? So this looks
like a true positive?
* `attrs`: Similar here. There are some new diagnostics on code that
tries to access `.validator` on a field. This *does* work at runtime,
but I'm not sure how that is supposed to type-check (without a [custom
plugin](2c6c395935/mypy/plugins/attrs.py (L575-L602))).
pyright errors on this as well.
* A handful of new false positives because we don't support `alias` yet
## Test Plan
Updated tests.
Summary
--
This PR unifies the two different ways Ruff and ty construct syntax
errors. Ruff has been storing the primary message in the diagnostic
itself, while ty attached the message to the primary annotation:
```
> ruff check try.py
invalid-syntax: name capture `x` makes remaining patterns unreachable
--> try.py:2:10
|
1 | match 42:
2 | case x: ...
| ^
3 | case y: ...
|
Found 1 error.
> uvx ty check try.py
WARN ty is pre-release software and not ready for production use. Expect to encounter bugs, missing features, and fatal errors.
Checking ------------------------------------------------------------ 1/1 files
error[invalid-syntax]
--> try.py:2:10
|
1 | match 42:
2 | case x: ...
| ^ name capture `x` makes remaining patterns unreachable
3 | case y: ...
|
Found 1 diagnostic
```
I think there are benefits to both approaches, and I do like ty's
version, but I feel like we should pick one (and it might help with
#20901 eventually). I slightly prefer Ruff's version, so I went with
that. Hopefully this isn't too controversial, but I'm happy to close
this if it is.
Note that this shouldn't change any other diagnostic formats in ty
because
[`Diagnostic::primary_message`](98d27c4128/crates/ruff_db/src/diagnostic/mod.rs (L177))
was already falling back to the primary annotation message if the
diagnostic message was empty. As a result, I think this change will
partially resolve the FIXME therein.
Test Plan
--
Existing tests with updated snapshots
This is the ultra-minimal implementation of
* https://github.com/astral-sh/ty/issues/296
that was previously discussed as a good starting point. In particular we
don't actually bother trying to figure out the exact python versions,
but we still mention "hey btw for No Reason At All... you're on python
3.10" when you try to access something that has a definition rooted in
the stdlib that we believe exists sometimes.
This is a drive-by improvement that I stumbled backwards into while
looking into
* https://github.com/astral-sh/ty/issues/296
I was writing some simple tests for "thing not in old version of stdlib"
diagnostics and checked what was added in 3.14, and saw
`compression.zstd` and to my surprise discovered that `import
compression.zstd` and `from compression import zstd` had completely
different quality diagnostics.
This is because `compression` and `compression.zstd` were *both*
introduced in 3.14, and so per VERSIONS policy only an entry for
`compression` was added, and so we don't actually have any definite info
on `compression.zstd` and give up on producing a diagnostic. However the
`from compression import zstd` form fails on looking up `compression`
and we *do* have an exact match for that, so it gets a better
diagnostic!
(aside: I have now learned about the VERSIONS format and I *really* wish
they would just enumerate all the submodules but, oh well!)
The fix is, when handling an import failure, if we fail to find an exact
match *we requery with the parent module*. In cases like
`compression.zstd` this lets us at least identify that, hey, not even
`compression` exists, and luckily that fixes the whole issue. In cases
where the parent module and submodule were introduced at different times
then we may discover that the parent module is in-range and that's fine,
we don't produce the richer stdlib diagnostic.
## Summary
`dataclasses.field` and field-specifier functions of commonly used
libraries like `pydantic`, `attrs`, and `SQLAlchemy` all return the
default type for the field (or `Any`) instead of an actual `Field`
instance, even if this is not what happens at runtime. Let's make use of
this fact and assume that *all* field specifiers return the type of the
default value of the field.
For standard dataclasses, this leads to more or less the same outcome
(see test diff for details), but this change is important for 3rd party
dataclass-transformers.
## Test Plan
Tested the consequences of this change on the field-specifiers branch as
well.
## Summary
Resolves https://github.com/astral-sh/ty/issues/1349.
Fix match statement value patterns to use equality comparison semantics
instead of incorrectly narrowing to literal types directly. Value
patterns use equality for matching, and equality can be overridden, so
we can't always narrow to the matched literal.
## Test Plan
Updated match.md with corrected expected types and an additional example
with explanation
---------
Co-authored-by: David Peter <mail@david-peter.de>
## Summary
If a function is decorated with a decorator that returns a union of
`Callable`s, also treat it as a union of function-like `Callable`s.
Labeling as `internal`, since the previous change has not been released
yet.
## Test Plan
New regression test.
Generic classes are not allowed to bind or reference a typevar from an
enclosing scope:
```py
def f[T](x: T, y: T) -> None:
class Ok[S]: ...
# error: [invalid-generic-class]
class Bad1[T]: ...
# error: [invalid-generic-class]
class Bad2(Iterable[T]): ...
class C[T]:
class Ok1[S]: ...
# error: [invalid-generic-class]
class Bad1[T]: ...
# error: [invalid-generic-class]
class Bad2(Iterable[T]): ...
```
It does not matter if the class uses PEP 695 or legacy syntax. It does
not matter if the enclosing scope is a generic class or function. The
generic class cannot even _reference_ an enclosing typevar in its base
class list.
This PR adds diagnostics for these cases.
In addition, the PR adds better fallback behavior for generic classes
that violate this rule: any enclosing typevars are not included in the
class's generic context. (That ensures that we don't inadvertently try
to infer specializations for those typevars in places where we
shouldn't.) The `dulwich` ecosystem project has [examples of
this](d912eaaffd/dulwich/config.py (L251))
that were causing new false positives on #20677.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
## Summary
Treat `Callable`s as bound-method descriptors if `Callable` is the
return type of a decorator that is applied to a function definition. See
the [rendered version of the new test
file](https://github.com/astral-sh/ruff/blob/david/callables-as-descriptors/crates/ty_python_semantic/resources/mdtest/call/callables_as_descriptors.md)
for the full description of this new heuristic.
I could imagine that we want to treat `Callable`s as bound-method
descriptors in other cases as well, but this seems like a step in the
right direction. I am planning to add other "use cases" from
https://github.com/astral-sh/ty/issues/491 to this test suite.
partially addresses https://github.com/astral-sh/ty/issues/491
closes https://github.com/astral-sh/ty/issues/1333
## Ecosystem impact
All positive
* 2961 removed `unsupported-operator` diagnostics on `sympy`, which was
one of the main motivations for implementing this change
* 37 removed `missing-argument` diagnostics, and no added call-error
diagnostics, which is an indicator that this heuristic shouldn't cause
many false positives
* A few removed `possibly-missing-attribute` diagnostics when accessing
attributes like `__name__` on decorated functions. The two added
`unused-ignore-comment` diagnostics are also cases of this.
* One new `invalid-assignment` diagnostic on `dd-trace-py`, which looks
suspicious, but only because our `invalid-assignment` diagnostics are
not great. This is actually a "Implicit shadowing of function"
diagnostic that hides behind the `invalid-assignment` diagnostic,
because a module-global function is being patched through a
`module.func` attribute assignment.
## Test Plan
New Markdown tests.
Carl Meyer
Jack O'Connor
Alex Waygood
David Peter
Douglas Creager
github-actions[bot]
Ibraheem Ahmed
Bhuminjay Soni
Shunsuke Shibayama
Shaygan Hooshyari
Mahmoud Saada
Aria Desires
justin
Dhruv Manilawala
Luca Chiodini
Micha Reiser
Mark Z. Ding
Brent Westbrook
Eric Mark Martin