## Summary
Move the `class_member` function to the `member` module. This allows us
to move the `member` module into the `types` module and to reduce the
visibility of its contents to `pub(super)`. The drawback is that we need
to make `place::place_by_id` public.
## Test Plan
Pure refactoring.
## Summary
When accessing an (instance) attribute on a given class, we were
previously traversing its MRO, and building a union of types (if the
attribute was available on multiple classes in the MRO) until we found a
*definitely bound* symbol. The idea was that possibly unbound symbols in
a subclass might only partially shadow the underlying base class
attribute.
This behavior was problematic for two reasons:
* if the attribute was definitely bound on a class (e.g. `self.x =
None`), we would have stopped iterating, even if there might be a `x:
str | None` declaration in a base class (the bug reported in
https://github.com/astral-sh/ty/issues/1067).
* if the attribute originated from an implicit instance attribute
assignment (e.g. `self.x = 1` in method `Sub.foo`), we might stop
looking and miss another implicit instance attribute assignment in a
base class method (e.g. `self.x = 2` in method `Base.bar`).
With this fix, we still iterate the MRO of the class, but we only stop
iterating if we find a *definitely declared* symbol. In this case, we
only return the declared attribute type. Otherwise, we keep building a
union of inferred attribute types.
The implementation here seemed to be the easiest fix for
https://github.com/astral-sh/ty/issues/1067 that also kept the ecosystem
impact low (the changes that I see all look correct). However, as the
Markdown tests show, there are other things to fix in this area. For
example, we should do a similar thing for *class attributes*. This is
more involved, though (affects many different areas and probably
involves a change to our descriptor protocol implementation), so I'd
like to postpone this to a follow-up.
closes https://github.com/astral-sh/ty/issues/1067
## Test Plan
Updated Markdown tests, including a regression test for
https://github.com/astral-sh/ty/issues/1067.
## Summary
Implements bidirectional type inference using function return type
annotations.
This PR was originally proposed to solve astral-sh/ty#1167, but this
does not fully resolve it on its own.
Additionally, I believe we need to allow dataclasses to generate their
own `__new__` methods, [use constructor return types for
inference](5844c0103d/crates/ty_python_semantic/src/types.rs (L5326-L5328)),
and a mechanism to discard type narrowing like `& ~AlwaysFalsy` if
necessary (at a more general level than this PR).
## Test Plan
`mdtest/bidirectional.md` is added.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
Co-authored-by: Ibraheem Ahmed <ibraheem@ibraheem.ca>
## Summary
Adds a set of basic new tests corresponding to open points in
https://github.com/astral-sh/ty/issues/1327, to document the state of
support for `dataclass_transform`.
## Summary
Type annotations are deferred by default starting with Python 3.14. No
`from __future__ import annotations` import is necessary.
## Test Plan
New Markdown test
## Summary
Simplify and fix the implementation of
`ty_extensions.CallableTypeOf[..]`.
closes https://github.com/astral-sh/ty/issues/1331
## Test Plan
Added regression test.
## Summary
This allows us to handle self-referential bounds/constraints/defaults
without panicking.
Handles more cases from https://github.com/astral-sh/ty/issues/256
This also changes the way we infer the types of legacy TypeVars. Rather
than understanding a constructor call to `typing[_extension].TypeVar`
inside of any (arbitrarily nested) expression, and having to use a
special `assigned_to` field of the semantic index to try to best-effort
figure out what name the typevar was assigned to, we instead understand
the creation of a legacy `TypeVar` only in the supported syntactic
position (RHS of a simple un-annotated assignment with one target). In
any other position, we just infer it as creating an opaque instance of
`typing.TypeVar`. (This behavior matches all other type checkers.)
So we now special-case TypeVar creation in `TypeInferenceBuilder`, as a
special case of an assignment definition, rather than deeper inside call
binding. This does mean we re-implement slightly more of
argument-parsing, but in practice this is minimal and easy to handle
correctly.
This is easier to implement if we also make the RHS of a simple (no
unpacking) one-target assignment statement no longer a standalone
expression. Which is fine to do, because simple one-target assignments
don't need to infer the RHS more than once. This is a bonus performance
(0-3% across various projects) and significant memory-usage win, since
most assignment statements are simple one-target assignment statements,
meaning we now create many fewer standalone-expression salsa
ingredients.
This change does mean that inference of manually-constructed
`TypeAliasType` instances can no longer find its Definition in
`assigned_to`, which regresses go-to-definition for these aliases. In a
future PR, `TypeAliasType` will receive the same treatment that
`TypeVar` did in this PR (moving its special-case inference into
`TypeInferenceBuilder` and supporting it only in the correct syntactic
position, and lazily inferring its value type to support recursion),
which will also fix the go-to-definition regression. (I decided a
temporary edge-case regression is better in this case than doubling the
size of this PR.)
This PR also tightens up and fixes various aspects of the validation of
`TypeVar` creation, as seen in the tests.
We still (for now) treat all typevars as instances of `typing.TypeVar`,
even if they were created using `typing_extensions.TypeVar`. This means
we'll wrongly error on e.g. `T.__default__` on Python 3.11, even if `T`
is a `typing_extensions.TypeVar` instance at runtime. We share this
wrong behavior with both mypy and pyrefly. It will be easier to fix
after we pull in https://github.com/python/typeshed/pull/14840.
There are some issues that showed up here with typevar identity and
`MarkTypeVarsInferable`; the fix here (using the new `original` field
and `is_identical_to` methods on `BoundTypeVarInstance` and
`TypeVarInstance`) is a bit kludgy, but it can go away when we eliminate
`MarkTypeVarsInferable`.
## Test Plan
Added and updated mdtests.
### Conformance suite impact
The impact here is all positive:
* We now correctly error on a legacy TypeVar with exactly one constraint
type given.
* We now correctly error on a legacy TypeVar with both an upper bound
and constraints specified.
### Ecosystem impact
Basically none; in the setuptools case we just issue slightly different
errors on an invalid TypeVar definition, due to the modified validation
code.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
## Summary
Avoid literal promotion when a literal type annotation is provided, e.g.,
```py
x: list[Literal[1]] = [1]
```
Resolves https://github.com/astral-sh/ty/issues/1198. This does not fix
issue https://github.com/astral-sh/ty/issues/1284, but it does make it
more relevant because after this change, it is possible to directly
instantiate a generic type with a literal specialization.
## Summary
Respect parameters such as `frozen_default` for metaclass-based
`@dataclass_transformer` models.
Related to: https://github.com/astral-sh/ty/issues/1260
## Typing conformance changes
Those are all correct (new true positives)
## Test Plan
New Markdown tests
## Summary
- Add support for eq, kw_only, and frozen parameter overrides in
@dataclass_transform
- Previously only order parameter override was supported
- Update test documentation to reflect fixed behavior
- Resolves issue where kw_only_default and frozen_default could not be
overridden
closes https://github.com/astral-sh/ty/issues/1260
## Test Plan
New Markdown tests
---------
Co-authored-by: David Peter <mail@david-peter.de>
## Summary
Two stable property tests are currently failing on `main`, following
f054b8a55e
(of course, I only thought to run the property tests again around 30
minutes _after_ landing that PR...). The issue is quite subtle, and took
me an annoying amount of time to pin down: we're matching over `(self,
other)` in `Type::is_disjoint_from_impl`, but `other` here is shadowed
by the binding in the `match` branch, which means that the wrong key is
inserted into the cache of the `IsDisjointFrom` cycle detector:
f054b8a55e/crates/ty_python_semantic/src/types.rs (L2408-L2435)
This PR fixes that issue, and also adds a few `Debug` implementations to
our cycle detectors, so that issues like this are easier to debug in the
future.
I'm adding the `internal` label, as this fixes a bug that hasn't yet
appeared in any released version of ty, so it doesn't deserve its own
changelog entry.
## Test Plan
`QUICKCHECK_TESTS=1000000 cargo test --release -p ty_python_semantic --
--ignored types::property_tests::stable` now once again passes on `main`
I considered adding new mdtests as well, but the examples that the
property tests were throwing at me all seemed _quite_ obscure and
somewhat unlikely to occur in the real world. I don't think it's worth
it.
## Summary
Even disambiguating classes using their fully qualified names is not
enough for some diagnostics. We've seen real-world examples in the
ecosystem (and https://github.com/astral-sh/ruff/pull/20368 introduces
some more!) where two types can be different, but can still have the
same fully qualified name. In these cases, our disambiguation machinery
needs to print the file path and line number of the class in order to
disambiguate classes with similar names in our diagnostics.
Helps with https://github.com/astral-sh/ty/issues/1306
## Test Plan
Mdtests
## Summary
This adds a couple of new test cases related to
https://github.com/astral-sh/ty/issues/1067 and beyond that. For now,
they are just documenting the current (problematic) behavior. Since the
topic has some subtleties, I'd like to merge this prior to the actual
bugfix(es) in order to evaluate the changes in an easier way.
## Summary
The `types` module currently re-exports a lot of functions and data
types from `types::ide_support`. One of these is called `Member`, a name
that is overloaded several times already. And I'd like to add one more
`Member` struct soon. Making the whole `ide_support` module public seems
cleaner to me, anyway.
## Test Plan
Pure refactoring.
## Summary
We have the following test in `protocols.md`:
```py
class HasX(Protocol):
x: int
# […]
class Foo:
x: int
# […]
class FooBool(Foo):
x: bool
static_assert(not is_subtype_of(FooBool, HasX))
static_assert(not is_assignable_to(FooBool, HasX))
```
If `Foo` was indeed intended to be a base class of `FooBool`, then `x:
bool` should be reported as a Liskov violation. And then it's a matter
of definition whether or not these assertions should hold true or not
(should the incorrect override take precedence or not?). So it looks to
me like this is just an oversight, probably a copy-paste error from
another test right before it, where `FooSub` is indeed intended to be a
subclass of `Foo`.
I am fixing this because this test started to fail on a branch of mine
that changes how attribute lookup in inheritance chains works.
## Summary
Fixes [astral-sh/ty#1307](https://github.com/astral-sh/ty/issues/1307)
Unions with length <= 5 are unaffected to minimize test churn
Unions with length > 5 will only display the first 3 elements + "...
omitted x union elements"
Here "length" is defined as the number of elements after condensation to
literals
Edit: we no longer truncate in revel case.
Before:
> info: Attempted to call union type `(def f1() -> int) | (def f2(name:
str) -> int) | (def f3(a: int, b: int) -> int) | (def f4[T](x: T@f4) ->
int) | Literal[5] | (Overload[() -> None, (x: str) -> str]) |
(Overload[() -> None, (x: str, y: str) -> str]) | PossiblyNotCallable`
After:
> info: Attempted to call union type `(def f1() -> int) | (def f2(name:
str) -> int) | (def f3(a: int, b: int) -> int) | ... omitted 5 union
elements`
The below comparisons are outdated, but left here as a reference.
Before:
```reveal_type(x) # revealed: Literal[1, 2] | A | B | C | D | E | F | G```
```reveal_type(x) # revealed: Result1A | Result1B | Result2A | Result2B
| Result3 | Result4```
After:
```reveal_type(x) # revealed: Literal[1, 2] | A | B | ... omitted 5 union elements```
```reveal_type(x) # revealed: Result1A | Result1B | Result2A | ...
omitted 3 union elements```
This formatting is consistent with
`crates/ty_python_semantic/src/types/call/bind.rs` line 2992
## Test Plan
Cosmetic only, covered and verified by changes in mdtest
## Summary
Bump the latest supported Python version of ty to 3.14 and updates some
references from 3.13 to 3.14.
This also fixes a bug with `dataclasses.field` on 3.14 (which adds a new
keyword-only parameter to that function, breaking our previously naive
matching on the parameter structure of that function).
## Test Plan
A `ty check` on a file with template strings (without any further
configuration) doesn't raise errors anymore.
## Summary
Typevar attributes (bound/constraints/default) can be either lazily
evaluated or eagerly evaluated. Currently they are lazily evaluated for
PEP 695 typevars, and eager for legacy and synthetic typevars.
https://github.com/astral-sh/ruff/pull/20598 will make them lazy also
for legacy typevars, and the ecosystem report on that PR surfaced the
issue fixed here (because legacy typevars are much more common in the
ecosystem than PEP 695 typevars.)
Applying a transform to a typevar (normalization, materialization, or
mark-inferable) will reify all lazy attributes and create a new typevar
with eager attributes. In terms of Salsa identity, this transformed
typevar will be considered different from the original typevar, whether
or not the attributes were actually transformed.
In general, this is not a problem, since all typevars in a given generic
context will be transformed, or not, together.
The exception to this was implicit-self vs explicit Self annotations.
The typevar we created for implicit self was created initially using
inferable typevars, whereas an explicit Self annotation is initially
non-inferable, then transformed via mark-inferable when accessed as part
of a function signature. If the containing class (which becomes the
upper bound of `Self`) is generic, and has e.g. a lazily-evaluated
default, then the explicit-Self annotation will reify that default in
the upper bound, and the implicit-self would not, leading them to be
treated as different typevars, and causing us to fail to solve a call to
a method such as `def method(self) -> Self` correctly.
The fix here is to treat implicit-self more like explicit-Self,
initially creating it as non-inferable and then using the mark-inferable
transform on it. This is less efficient, but restores the invariant that
all typevars in a given generic context are transformed together, or
not, fixing the bug.
In the improved-constraint-solver work, the separation of typevars into
"inferable" and "non-inferable" is expected to disappear, along with the
mark-inferable transform, which would render both this bug and the fix
moot. So this fix is really just temporary until that lands.
There is a performance regression, but not a huge one: 1-2% on most
projects, 5% on one outlier. This seems acceptable, given that it should
be fully recovered by removing the mark-inferable transform.
## Test Plan
Added mdtests that failed before this change.
This PR adds a specialization inference special case that lets us handle
the following examples better:
```py
def f[T](t: T | None) -> T: ...
def g[T](t: T | int | None) -> T | int: ...
def _(x: str | None):
reveal_type(f(x)) # revealed: str (previously str | None)
def _(y: str | int | None):
reveal_type(g(x)) # revealed: str | int (previously str | int | None)
```
We already have a special case for when the formal is a union where one
element is a typevar, but it maps the entire actual type to the typevar
(as you can see in the "previously" results above).
The new special case kicks in when the actual is also a union. Now, we
filter out any actual union elements that are already subtypes of the
formal, and only bind whatever types remain to the typevar. (The `|
None` pattern appears quite often in the ecosystem results, but it's
more general and works with any number of non-typevar union elements.)
The new constraint solver should handle this case as well, but it's
worth adding this heuristic now with the old solver because it
eliminates some false positives from the ecosystem report, and makes the
ecosystem report less noisy on the other constraint solver PRs.
Summary
--
Closes#19467 and also removes the warning about using Python 3.14
without
preview enabled.
I also bumped `PythonVersion::default` to 3.9 because it reaches EOL
this month,
but we could also defer that for now if we wanted.
The first three commits are related to the `latest` bump to 3.14; the
fourth commit
bumps the default to 3.10.
Note that this PR also bumps the default Python version for ty to 3.10
because
there was a test asserting that it stays in sync with
`ast::PythonVersion`.
Test Plan
--
Existing tests
I spot-checked the ecosystem report, and I believe these are all
expected. Inbits doesn't specify a target Python version, so I guess
we're applying the default. UP007, UP035, and UP045 all use the new
default value to emit new diagnostics.
## Summary
`infer_method_information` was previously calling
`ClassLiteral::to_class_type`, which uses the default-specialization of
a generic class. This specialized `ClassType` was later only used if the
class was non-generic, making the specialization irrelevant. The
implementation was still a bit confusing, so this PR proposes a way to
avoid turning the class literal into a `ClassType`.
## Summary
Use the type annotation of function parameters as bidirectional type
context when inferring the argument expression. For example, the
following example now type-checks:
```py
class TD(TypedDict):
x: int
def f(_: TD): ...
f({ "x": 1 })
```
Part of https://github.com/astral-sh/ty/issues/168.
We add an `inherited_generic_context` to the constructors of a generic
class. That lets us infer specializations of the class when invoking the
constructor. The constructor might itself be generic, in which case we
have to merge the list of typevars that we are willing to infer in the
constructor call.
Before we did that by tracking the two (and their specializations)
separately, with distinct `Option` fields/parameters. This PR updates
our call binding logic such that any given function call has _one_
optional generic context that we're willing to infer a specialization
for. If needed, we use the existing `GenericContext::merge` method to
create a new combined generic context for when the class and constructor
are both generic. This simplifies the call binding code considerably,
and is no more complex in the constructor call logic.
We also have a heuristic that we will promote any literals in the
specialized types of a generic class, but we don't promote literals in
the specialized types of the function itself. To handle this, we now
track this `should_promote_literals` property within `GenericContext`.
And moreover, we track this separately for each typevar, instead of a
single property for the generic context as a whole, so that we can
correctly merge the generic context of a constructor method (where the
option should be `false`) with the inherited generic context of its
containing class (where the option should be `true`).
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
## Summary
The union `T | U` can be validly simplified to `U` iff:
1. `T` is a subtype of `U` OR
2. `T` is equivalent to `U` OR
3. `U` is a union and contains a type that is equivalent to `T` OR
4. `T` is an intersection and contains a type that is equivalent to `U`
(In practice, the only situation in which 2, 3 or 4 would be true when
(1) was not true would be if `T` or `U` is a dynamic type.)
Currently we achieve these simplifications in the union builder by doing
something along the lines of `t.is_subtype_of(db, u) ||
t.is_equivalent_to_(db, u) ||
t.into_intersection().is_some_and(|intersection|
intersection.positive(db).contains(&u)) ||
u.into_union().is_some_and(|union| union.elements(db).contains(&t))`.
But this is both slow and misses some cases (it doesn't simplify the
union `Any | (Unknown & ~None)` to `Any`, for example). We can improve
the consistency and performance of our union simplifications by adding a
third type relation that sits in between `TypeRelation::Subtyping` and
`TypeRelation::Assignability`: `TypeRelation::UnionSimplification`.
This change leads to simpler, more user-friendly types due to the more
consistent simplification. It also lead to a pretty huge performance
improvement!
## Test Plan
Existing tests, plus some new ones.
<!--
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
#19990 didn't completely fix the base vs. child conda environment
distinction, since it detected slightly different behavior than what I
usually see in conda. E.g., I see something like the following:
```
(didn't yet activate conda, but base is active)
➜ printenv | grep CONDA
CONDA_PYTHON_EXE=/opt/anaconda3/bin/python
CONDA_PREFIX=/opt/anaconda3
CONDA_DEFAULT_ENV=base
CONDA_EXE=/opt/anaconda3/bin/conda
CONDA_SHLVL=1
CONDA_PROMPT_MODIFIER=(base)
(activating conda)
➜ conda activate test
(test is an active conda environment)
❯ printenv | grep CONDA
CONDA_PREFIX=/opt/anaconda3/envs/test
CONDA_PYTHON_EXE=/opt/anaconda3/bin/python
CONDA_SHLVL=2
CONDA_PREFIX_1=/opt/anaconda3
CONDA_DEFAULT_ENV=test
CONDA_PROMPT_MODIFIER=(test)
CONDA_EXE=/opt/anaconda3/bin/conda
```
But the current behavior looks for `CONDA_DEFAULT_ENV =
basename(CONDA_PREFIX)` for the base environment instead of the child
environment, where we actually see this equality.
This pull request fixes that and updates the tests correspondingly.
## Test Plan
I updated the existing tests with the new behavior. Let me know if you
want more tests. Note: It shouldn't be necessary to test for the case
where we have `conda/envs/base`, since one should not be able to create
such an environment (one with the name of `CONDA_DEFAULT_ENV`).
---------
Co-authored-by: Aria Desires <aria.desires@gmail.com>
Previously, we would always add `/{*filepath}` as our wildcard to match
descendant paths. But when the root is just `/` (as it can be in tests,
weird environments or in the ty playground), this causes a double `/`
and inhibits most descendant matches.
The regression test added in this commit fails without this fix.
Specifically, it panics because it can't find a file root for
`/project`.
Fixes#1277
## Summary
closes: https://github.com/astral-sh/ty/issues/247
This PR adds support for variadic arguments to overload call evaluation.
This basically boils down to making sure that the overloads are not
filtered out incorrectly during the step 5 in the overload call
evaluation algorithm. For context, the step 5 tries to filter out the
remaining overloads after finding an overload where the materialization
of argument types are assignable to the parameter types.
The issue with the previous implementation was that it wouldn't unpack
the variadic argument and wouldn't consider the many-to-one (multiple
arguments mapping to a single variadic parameter) correctly. This PR
fixes that.
## Test Plan
Update existing test cases and resolve the TODOs.
## Summary
Currently we do not emit an error on this code:
```py
from ty_extensions import Not
def f[T](x: T, y: Not[T]) -> T:
x = y
return x
```
But we should do! `~T` should never be assignable to `T`.
This fixes a small regression introduced in
14fe1228e7 (diff-8049ab5af787dba29daa389bbe2b691560c15461ef536f122b1beab112a4b48aR1443-R1446),
where a branch that previously returned `false` was replaced with a
branch that returns `C::always_satisfiable` -- the opposite of what it
used to be! The regression occurred because we didn't have any tests for
this -- so I added some tests in this PR that fail on `main`. I only
spotted the problem because I was going through the code of
`has_relation_to_impl` with a fine toothcomb for
https://github.com/astral-sh/ruff/pull/20602😄
## Summary
Quoting from the newly added comment:
Module-level globals can be mutated externally. A `MY_CONSTANT = 1`
global might be changed to `"some string"` from code outside of the
module that we're looking at, and so from a gradual-guarantee
perspective, it makes sense to infer a type of `Literal[1] | Unknown`
for global symbols. This allows the code that does the mutation to type
check correctly, and for code that uses the global, it accurately
reflects the lack of knowledge about the type.
External modifications (or modifications through `global` statements)
that would require a wider type are relatively rare. From a practical
perspective, we can therefore achieve a better user experience by
trusting the inferred type. Users who need the external mutation to work
can always annotate the global with the wider type. And everyone else
benefits from more precise type inference.
I initially implemented this by applying literal promotion to the type
of the unannotated module globals (as suggested in
https://github.com/astral-sh/ty/issues/1069), but the ecosystem impact
showed a lot of problems (https://github.com/astral-sh/ruff/pull/20643).
I fixed/patched some of these problems, but this PR seems like a good
first step, and it seems sensible to apply the literal promotion change
in a second step that can be evaluated separately.
closes https://github.com/astral-sh/ty/issues/1069
## Ecosystem impact
This seems like an (unexpectedly large) net positive with 650 fewer
diagnostics overall.. even though this change will certainly catch more
true positives.
* There are 666 removed `type-assertion-failure` diagnostics, where we
were previously used the correct type already, but removing the
`Unknown` now leads to an "exact" match.
* 1464 of the 1805 total new diagnostics are `unresolved-attribute`
errors, most (1365) of which were previously
`possibly-missing-attribute` errors. So they could also be counted as
"changed" diagnostics.
* For code that uses constants like
```py
IS_PYTHON_AT_LEAST_3_10 = sys.version_info >= (3, 10)
```
where we would have previously inferred a type of `Literal[True/False] |
Unknown`, removing the `Unknown` now allows us to do reachability
analysis on branches that use these constants, and so we get a lot of
favorable ecosystem changes because of that.
* There is code like the following, where we previously emitted
`conflicting-argument-forms` diagnostics on calls to the aliased
`assert_type`, because its type was `Unknown | def …` (and the call to
`Unknown` "used" the type form argument in a non type-form way):
```py
if sys.version_info >= (3, 11):
import typing
assert_type = typing.assert_type
else:
import typing_extensions
assert_type = typing_extensions.assert_type
```
* ~100 new `invalid-argument-type` false positives, due to missing
`**kwargs` support (https://github.com/astral-sh/ty/issues/247)
## Typing conformance
```diff
+protocols_modules.py:25:1: error[invalid-assignment] Object of type `<module '_protocols_modules1'>` is not assignable to `Options1`
```
This diagnostic should apparently not be there, but it looks like we
also fail other tests in that file, so it seems to be a limitation that
was previously hidden by `Unknown` somehow.
## Test Plan
Updated tests and relatively thorough ecosystem analysis.
## Summary
Reformulation of the public symbol type inference test suite to use
class scopes instead of module scopes. This is in preparation for an
upcoming change to module-global scopes (#20664).
## Test Plan
Updated tests
This doesn't seem to be flaky in the sense of tests failing
non-deterministically, but they are flaky in the sense of unrelated
changes causing testing failures from the clauses of a constraint set
being rendered in different orders. This flakiness is because we're
using Salsa IDs to determine the order in which typevars appear in a
constraint set BDD, and those IDs are assigned non-deterministically.
The fix is ham-fisted but effective: sort the constraints in each
clause, and the clauses in each set, as part of the rendering process.
Constraint sets are only rendered in our test cases, so we don't need to
over-optimize this.
## Summary
Not sure if this was the original intention, but it looks to me like the
previous `Type::literal_promotion_type` was more of an implementation
detail for the actual operation of promoting all literals in a
possibly-nested position of a type.
This is not a pure refactor, as I'm technically changing the behavior
for that protocols diagnostic message suggestion.
## Test Plan
New Markdown test
## Summary
Add two simple tests that we recently discussed with @dcreager. They
demonstrate that the `TypeMapping::MarkTypeVarsInferable` operation
really does need to keep track of the binding context.
## Test Plan
Made sure that those tests fail if we create
`TypeMapping::MarkTypeVarsInferable(None)`s everywhere.
## Summary
Modify the (external) signature of instance methods such that the first
parameter uses `Self` unless it is explicitly annotated. This allows us
to correctly type-check more code, and allows us to infer correct return
types for many functions that return `Self`. For example:
```py
from pathlib import Path
from datetime import datetime, timedelta
reveal_type(Path(".config") / ".ty") # now Path, previously Unknown
def _(dt: datetime, delta: timedelta):
reveal_type(dt - delta) # now datetime, previously Unknown
```
part of https://github.com/astral-sh/ty/issues/159
## Performance
I ran benchmarks locally on `attrs`, `freqtrade` and `colour`, the
projects with the largest regressions on CodSpeed. I see much smaller
effects locally, but can definitely reproduce the regression on `attrs`.
From looking at the profiling results (on Codspeed), it seems that we
simply do more type inference work, which seems plausible, given that we
now understand much more return types (of many stdlib functions). In
particular, whenever a function uses an implicit `self` and returns
`Self` (without mentioning `Self` anywhere else in its signature), we
will now infer the correct type, whereas we would previously return
`Unknown`. This also means that we need to invoke the generics solver in
more cases. Comparing half a million lines of log output on attrs, I can
see that we do 5% more "work" (number of lines in the log), and have a
lot more `apply_specialization` events (7108 vs 4304). On freqtrade, I
see similar numbers for `apply_specialization` (11360 vs 5138 calls).
Given these results, I'm not sure if it's generally worth doing more
performance work, especially since none of the code modifications
themselves seem to be likely candidates for regressions.
| Command | Mean [ms] | Min [ms] | Max [ms] | Relative |
|:---|---:|---:|---:|---:|
| `./ty_main check /home/shark/ecosystem/attrs` | 92.6 ± 3.6 | 85.9 |
102.6 | 1.00 |
| `./ty_self check /home/shark/ecosystem/attrs` | 101.7 ± 3.5 | 96.9 |
113.8 | 1.10 ± 0.06 |
| Command | Mean [ms] | Min [ms] | Max [ms] | Relative |
|:---|---:|---:|---:|---:|
| `./ty_main check /home/shark/ecosystem/freqtrade` | 599.0 ± 20.2 |
568.2 | 627.5 | 1.00 |
| `./ty_self check /home/shark/ecosystem/freqtrade` | 607.9 ± 11.5 |
594.9 | 626.4 | 1.01 ± 0.04 |
| Command | Mean [ms] | Min [ms] | Max [ms] | Relative |
|:---|---:|---:|---:|---:|
| `./ty_main check /home/shark/ecosystem/colour` | 423.9 ± 17.9 | 394.6
| 447.4 | 1.00 |
| `./ty_self check /home/shark/ecosystem/colour` | 426.9 ± 24.9 | 373.8
| 456.6 | 1.01 ± 0.07 |
## Test Plan
New Markdown tests
## Ecosystem report
* apprise: ~300 new diagnostics related to problematic stubs in apprise
😩
* attrs: a new true positive, since [this
function](4e2c89c823/tests/test_make.py (L2135))
is missing a `@staticmethod`?
* Some legitimate true positives
* sympy: lots of new `invalid-operator` false positives in [matrix
multiplication](cf9f4b6805/sympy/matrices/matrixbase.py (L3267-L3269))
due to our limited understanding of [generic `Callable[[Callable[[T1,
T2], T3]], Callable[[T1, T2], T3]]` "identity"
types](cf9f4b6805/sympy/core/decorators.py (L83-L84))
of decorators. This is not related to type-of-self.
## Typing conformance results
The changes are all correct, except for
```diff
+generics_self_usage.py:50:5: error[invalid-assignment] Object of type `def foo(self) -> int` is not assignable to `(typing.Self, /) -> int`
```
which is related to an assignability problem involving type variables on
both sides:
```py
class CallableAttribute:
def foo(self) -> int:
return 0
bar: Callable[[Self], int] = foo # <- we currently error on this assignment
```
---------
Co-authored-by: Shaygan Hooshyari <sh.hooshyari@gmail.com>
