## Summary
Regression test(s) for something that broken while implementing #14759.
We have similar tests for other control flow elements, but feel free to
let me know if this seems superfluous.
## Test Plan
New mdtests
## Summary
Using `typing.LiteralString` breaks as soon as we understand
`sys.version_info` branches, as it's only available in 3.11 and later.
## Test Plan
Made sure it didn't fail on my #14759 branch anymore.
We support using `typing.Type[]` as a base class (and we have tests for
it), but not yet `builtins.type[]`. At some point we should fix that,
but I don't think it';s worth spending much time on now (and it might be
easier once we've implemented generics?). This PR just adds a failing
test with a TODO.
## Summary
This is the third and last PR in this stack that adds support for
toggling lints at a per-rule level.
This PR introduces a new `LintRegistry`, a central index of known lints.
The registry is required because we want to support lint rules from many
different crates but need a way to look them up by name, e.g., when
resolving a lint from a name in the configuration or analyzing a
suppression comment.
Adding a lint now requires two steps:
1. Declare the lint with `declare_lint`
2. Register the lint in the registry inside the `register_lints`
function.
I considered some more involved macros to avoid changes in two places.
Still, I ultimately decided against it because a) it's just two places
and b) I'd expect that registering a type checker lint will differ from
registering a lint that runs as a rule in the linter. I worry that any
more opinionated design could limit our options when working on the
linter, so I kept it simple.
The second part of this PR is the `RuleSelection`. It stores which lints
are enabled and what severity they should use for created diagnostics.
For now, the `RuleSelection` always gets initialized with all known
lints and it uses their default level.
## Linter crates
Each crate that defines lints should export a `register_lints` function
that accepts a `&mut LintRegistryBuilder` to register all its known
lints in the registry. This should make registering all known lints in a
top-level crate easy: Just call `register_lints` of every crate that
defines lint rules.
I considered defining a `LintCollection` trait and even some fancy
macros to accomplish the same but decided to go for this very simplistic
approach for now. We can add more abstraction once needed.
## Lint rules
This is a bit hand-wavy. I don't have a good sense for how our linter
infrastructure will look like, but I expect we'll need a way to register
the rules that should run as part of the red knot linter. One way is to
keep doing what Ruff does by having one massive `checker` and each lint
rule adds a call to itself in the relevant AST visitor methods. An
alternative is that we have a `LintRule` trait that provides common
hooks and implementations will be called at the "right time". Such a
design would need a way to register all known lint implementations,
possibly with the lint. This is where we'd probably want a dedicated
`register_rule` method. A third option is that lint rules are handled
separately from the `LintRegistry` and are specific to the linter crate.
The current design should be flexible enough to support the three
options.
## Documentation generation
The documentation for all known lints can be generated by creating a
factory, registering all lints by calling the `register_lints` methods,
and then querying the registry for the metadata.
## Deserialization and Schema generation
I haven't fully decided what the best approach is when it comes to
deserializing lint rule names:
* Reject invalid names in the deserializer. This gives us error messages
with line and column numbers (by serde)
* Don't validate lint rule names during deserialization; defer the
validation until the configuration is resolved. This gives us more
control over handling the error, e.g. emit a warning diagnostic instead
of aborting when a rule isn't known.
One technical challenge for both deserialization and schema generation
is that the `Deserialize` and `JSONSchema` traits do not allow passing
the `LintRegistry`, which is required to look up the lints by name. I
suggest that we either rely on the salsa db being set for the current
thread (`salsa::Attach`) or build our own thread-local storage for the
`LintRegistry`. It's the caller's responsibility to make the lint
registry available before calling `Deserialize` or `JSONSchema`.
## CLI support
I prefer deferring adding support for enabling and disabling lints from
the CLI for now because I think it will be easier
to add once I've figured out how to handle configurations.
## Bitset optimization
Ruff tracks the enabled rules using a cheap copyable `Bitset` instead of
a hash map. This helped improve performance by a few percent (see
https://github.com/astral-sh/ruff/pull/3606). However, this approach is
no longer possible because lints have no "cheap" way to compute their
index inside the registry (other than using a hash map).
We could consider doing something similar to Salsa where each
`LintMetadata` stores a `LazyLintIndex`.
```
pub struct LazyLintIndex {
cached: OnceLock<(Nonce, LintIndex)>
}
impl LazyLintIndex {
pub fn get(registry: &LintRegistry, lint: &'static LintMetadata) {
let (nonce, index) = self.cached.get_or_init(|| registry.lint_index(lint));
if registry.nonce() == nonce {
index
} else {
registry.lint_index(lint)
}
}
```
Each registry keeps a map from `LintId` to `LintIndex` where `LintIndex`
is in the range of `0...registry.len()`. The `LazyLintIndex` is based on
the assumption that every program has exactly **one** registry. This
assumption allows to cache the `LintIndex` directly on the
`LintMetadata`. The implementation falls back to the "slow" path if
there is more than one registry at runtime.
I was very close to implementing this optimization because it's kind of
fun to implement. I ultimately decided against it because it adds
complexity and I don't think it's worth doing in Red Knot today:
* Red Knot only queries the rule selection when deciding whether or not
to emit a diagnostic. It is rarely used to detect if a certain code
block should run. This is different from Ruff where the rule selection
is queried many times for every single AST node to determine which rules
*should* run.
* I'm not sure if a 2-3% performance improvement is worth the complexity
I suggest revisiting this decision when working on the linter where a
fast path for deciding if a rule is enabled might be more important (but
that depends on how lint rules are implemented)
## Test Plan
I removed a lint from the default rule registry, and the MD tests
started failing because the diagnostics were no longer emitted.
This adds support for `type[Any]`, which represents an unknown type (not
an instance of an unknown type), and `type`, which we are choosing to
interpret as `type[object]`.
Closes#14546
## Summary
This is already several hundred lines of code, and it will get more
complex with call-signature checking.
## Test Plan
This is a pure code move; the moved code wasn't changed, just imports.
Existing tests pass.
## Summary
Add a `is_fully_static` premise to the equivalence on subtyping property tests.
## Test Plan
```
cargo test -p red_knot_python_semantic -- --ignored types::property_tests::stable
```
## Summary
This is the second PR out of three that adds support for
enabling/disabling lint rules in Red Knot. You may want to take a look
at the [first PR](https://github.com/astral-sh/ruff/pull/14869) in this
stack to familiarize yourself with the used terminology.
This PR adds a new syntax to define a lint:
```rust
declare_lint! {
/// ## What it does
/// Checks for references to names that are not defined.
///
/// ## Why is this bad?
/// Using an undefined variable will raise a `NameError` at runtime.
///
/// ## Example
///
/// ```python
/// print(x) # NameError: name 'x' is not defined
/// ```
pub(crate) static UNRESOLVED_REFERENCE = {
summary: "detects references to names that are not defined",
status: LintStatus::preview("1.0.0"),
default_level: Level::Warn,
}
}
```
A lint has a name and metadata about its status (preview, stable,
removed, deprecated), the default diagnostic level (unless the
configuration changes), and documentation. I use a macro here to derive
the kebab-case name and extract the documentation automatically.
This PR doesn't yet add any mechanism to discover all known lints. This
will be added in the next and last PR in this stack.
## Documentation
I documented some rules but then decided that it's probably not my best
use of time if I document all of them now (it also means that I play
catch-up with all of you forever). That's why I left some rules
undocumented (marked with TODO)
## Where is the best place to define all lints?
I'm not sure. I think what I have in this PR is fine but I also don't
love it because most lints are in a single place but not all of them. If
you have ideas, let me know.
## Why is the message not part of the lint, unlike Ruff's `Violation`
I understand that the main motivation for defining `message` on
`Violation` in Ruff is to remove the need to repeat the same message
over and over again. I'm not sure if this is an actual problem. Most
rules only emit a diagnostic in a single place and they commonly use
different messages if they emit diagnostics in different code paths,
requiring extra fields on the `Violation` struct.
That's why I'm not convinced that there's an actual need for it and
there are alternatives that can reduce the repetition when creating a
diagnostic:
* Create a helper function. We already do this in red knot with the
`add_xy` methods
* Create a custom `Diagnostic` implementation that tailors the entire
diagnostic and pre-codes e.g. the message
Avoiding an extra field on the `Violation` also removes the need to
allocate intermediate strings as it is commonly the place in Ruff.
Instead, Red Knot can use a borrowed string with `format_args`
## Test Plan
`cargo test`
## Summary
This PR introduces a structured `DiagnosticId` instead of using a plain
`&'static str`. It is the first of three in a stack that implements a
basic rules infrastructure for Red Knot.
`DiagnosticId` is an enum over all known diagnostic codes. A closed enum
reduces the risk of accidentally introducing two identical diagnostic
codes. It also opens the possibility of generating reference
documentation from the enum in the future (not part of this PR).
The enum isn't *fully closed* because it uses a `&'static str` for lint
names. This is because we want the flexibility to define lints in
different crates, and all names are only known in `red_knot_linter` or
above. Still, lower-level crates must already reference the lint names
to emit diagnostics. We could define all lint-names in `DiagnosticId`
but I decided against it because:
* We probably want to share the `DiagnosticId` type between Ruff and Red
Knot to avoid extra complexity in the diagnostic crate, and both tools
use different lint names.
* Lints require a lot of extra metadata beyond just the name. That's why
I think defining them close to their implementation is important.
In the long term, we may also want to support plugins, which would make
it impossible to know all lint names at compile time. The next PR in the
stack introduces extra syntax for defining lints.
A closed enum does have a few disadvantages:
* rustc can't help us detect unused diagnostic codes because the enum is
public
* Adding a new diagnostic in the workspace crate now requires changes to
at least two crates: It requires changing the workspace crate to add the
diagnostic and the `ruff_db` crate to define the diagnostic ID. I
consider this an acceptable trade. We may want to move `DiagnosticId` to
its own crate or into a shared `red_knot_diagnostic` crate.
## Preventing duplicate diagnostic identifiers
One goal of this PR is to make it harder to introduce ambiguous
diagnostic IDs, which is achieved by defining a closed enum. However,
the enum isn't fully "closed" because it doesn't explicitly list the IDs
for all lint rules. That leaves the possibility that a lint rule and a
diagnostic ID share the same name.
I made the names unambiguous in this PR by separating them into
different namespaces by using `lint/<rule>` for lint rule codes. I don't
mind the `lint` prefix in a *Ruff next* context, but it is a bit weird
for a standalone type checker. I'd like to not overfocus on this for now
because I see a few different options:
* We remove the `lint` prefix and add a unit test in a top-level crate
that iterates over all known lint rules and diagnostic IDs to ensure the
names are non-overlapping.
* We only render `[lint]` as the error code and add a note to the
diagnostic mentioning the lint rule. This is similar to clippy and has
the advantage that the header line remains short
(`lint/some-long-rule-name` is very long ;))
* Any other form of adjusting the diagnostic rendering to make the
distinction clear
I think we can defer this decision for now because the `DiagnosticId`
contains all the relevant information to change the rendering
accordingly.
## Why `Lint` and not `LintRule`
I see three kinds of diagnostics in Red Knot:
* Non-suppressable: Reveal type, IO errors, configuration errors, etc.
(any `DiagnosticId`)
* Lints: code-related diagnostics that are suppressable.
* Lint rules: The same as lints, but they can be enabled or disabled in
the configuration. The majority of lints in Red Knot and the Ruff
linter.
Our current implementation doesn't distinguish between lints and Lint
rules because we aren't aware of a suppressible code-related lint that
can't be configured in the configuration. The only lint that comes to my
mind is maybe `division-by-zero` if we're 99.99% sure that it is always
right. However, I want to keep the door open to making this distinction
in the future if it proves useful.
Another reason why I chose lint over lint rule (or just rule) is that I
want to leave room for a future lint rule and lint phase concept:
* lint is the *what*: a specific code smell, pattern, or violation
* the lint rule is the *how*: I could see a future `LintRule` trait in
`red_knot_python_linter` that provides the necessary hooks to run as
part of the linter. A lint rule produces diagnostics for exactly one
lint. A lint rule differs from all lints in `red_knot_python_semantic`
because they don't run as "rules" in the Ruff sense. Instead, they're a
side-product of type inference.
* the lint phase is a different form of *how*: A lint phase can produce
many different lints in a single pass. This is a somewhat common pattern
in Ruff where running one analysis collects the necessary information
for finding many different lints
* diagnostic is the *presentation*: Unlike a lint, the diagnostic isn't
the what, but how a specific lint gets presented. I expect that many
lints can use one generic `LintDiagnostic`, but a few lints might need
more flexibility and implement their custom diagnostic rendering (at
least custom `Diagnostic` implementation).
## Test Plan
`cargo test`
## Summary
Per suggestion in
https://github.com/astral-sh/ruff/pull/14802#discussion_r1875455417
This is a bit less error-prone and allows us to handle both expressions
in the current scope or a different scope. Also, there's currently no
need for this method outside of `TypeInferenceBuilder`, so no reason to
expose it in `types.rs`.
## Test Plan
Pure refactor, no functional change; existing tests pass.
---------
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
This adds support for `type[a.X]`, where the `type` special form is
applied to a qualified name that resolves to a class literal. This works
for both nested classes and classes imported from another module.
Closes#14545
## Summary
Inferred and declared types for function parameters, in the function
body scope.
Fixes#13693.
## Test Plan
Added mdtests.
---------
Co-authored-by: Micha Reiser <micha@reiser.io>
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
## Summary
`typing.Never` and `typing.LiteralString` are only conditionally
exported from `typing` for Python versions 3.11 and later. We run the
Markdown tests with the default Python version of 3.9, so here we change
the import to `typing_extensions` instead, and add a new test to make
sure we'll continue to understand the `typing`-version of these symbols
for newer versions.
This didn't cause problems so far, as we don't understand
`sys.version_info` branches yet.
## Test Plan
New Markdown tests to make sure this will continue to work in the
future.
## Summary
This adds support for specifying the target Python version from a
Markdown test. It is a somewhat limited ad-hoc solution, but designed to
be future-compatible. TOML blocks can be added to arbitrary sections in
the Markdown block. They have the following format:
````markdown
```toml
[tool.knot.environment]
target-version = "3.13"
```
````
So far, there is nothing else that can be configured, but it should be
straightforward to extend this to things like a custom typeshed path.
This is in preparation for the statically-known branches feature where
we are going to have to specify the target version for lots of tests.
## Test Plan
- New Markdown test that fails without the explicitly specified
`target-version`.
- Manually tested various error paths when specifying a wrong
`target-version` field.
- Made sure that running tests is as fast as before.
## Summary
This is related to #13778, more specifically
https://github.com/astral-sh/ruff/issues/13778#issuecomment-2513556004.
This PR adds various test cases where a keyword is being where an
identifier is expected. The tests are to make sure that red knot doesn't
panic, raises the syntax error and the identifier is added to the symbol
table. The final part allows editor related features like renaming the
symbol.
## Summary
`typing_extensions` has a `>=3.13` re-export for the `typing.NoDefault`
singleton, but not for `typing._NoDefaultType`. This causes problems as
soon as we understand `sys.version_info` branches, so we explicity
switch to `typing._NoDefaultType` for Python 3.13 and later.
This is a part of #14759 that I thought might make sense to break out
and merge in isolation.
## Test Plan
New test that will become more meaningful with #12700
---------
Co-authored-by: Micha Reiser <micha@reiser.io>
## Summary
- Instead of seven (more or less similar) `setup_db` functions, use just
one in a single central place.
- For every test that needs customization beyond that, offer a
`TestDbBuilder` that can control the Python target version, custom
typeshed, and pre-existing files.
The main motivation for this is that we're soon going to need
customization of the Python version, and I didn't feel like adding this
to each of the existing `setup_db` functions.
## Summary
This changeset contains various improvements concerning non-fully-static
types and their relationships:
- Make sure that non-fully-static types do not participate in
equivalence or subtyping.
- Clarify what `Type::is_equivalent_to` actually implements.
- Introduce `Type::is_fully_static`
- New tests making sure that multiple `Any`/`Unknown`s inside unions and
intersections are collapsed.
closes#14524
## Test Plan
- Added new unit tests for union and intersection builder
- Added new unit tests for `Type::is_equivalent_to`
- Added new unit tests for `Type::is_subtype_of`
- Added new property test making sure that non-fully-static types do not
participate in subtyping
We already had a representation for the Any type, which we would use
e.g. for expressions without type annotations. We now recognize
`typing.Any` as a way to refer to this type explicitly. Like other
special forms, this is tracked correctly through aliasing, and isn't
confused with local definitions that happen to have the same name.
Closes#14544
## Summary
Minor change that uses two plain classes `A` and `B` instead of
`typing.Sized` and `typing.Hashable`.
The motivation is twofold: I remember that I was confused when I first
saw this test. Was there anything specific to `Sized` and `Hashable`
that was relevant here? (there is, these classes are not overlapping;
and you can build a proper intersection from them; but that's true for
almost all non-builtin classes).
I now ran into another problem while working on #14758: `Sized` and
`Hashable` are protocols that we don't fully understand yet. This
causing some trouble when trying to infer whether these are fully-static
types or not.
## Summary
This PR adds a new `property_tests` module with quickcheck-based tests
that verify certain properties of types. The following properties are
currently checked:
* `is_equivalent_to`:
* is reflexive: `T` is equivalent to itself
* `is_subtype_of`:
* is reflexive: `T` is a subtype of `T`
* is antisymmetric: if `S <: T` and `T <: S`, then `S` is equivalent to
`T`
* is transitive: `S <: T` & `T <: U` => `S <: U`
* `is_disjoint_from`:
* is irreflexive: `T` is not disjoint from `T`
* is symmetric: `S` disjoint from `T` => `T` disjoint from `S`
* `is_assignable_to`:
* is reflexive
* `negate`:
* is an involution: `T.negate().negate()` is equivalent to `T`
There are also some tests that validate higher-level properties like:
* `S <: T` implies that `S` is not disjoint from `T`
* `S <: T` implies that `S` is assignable to `T`
* A singleton type must also be single-valued
These tests found a few bugs so far:
- #14177
- #14195
- #14196
- #14210
- #14731
Some additional notes:
- Quickcheck-based property tests are non-deterministic and finding
counter-examples might take an arbitrary long time. This makes them bad
candidates for running in CI (for every PR). We can think of running
them in a cron-job way from time to time, similar to fuzzing. But for
now, it's only possible to run them locally (see instructions in source
code).
- Some tests currently find false positive "counterexamples" because our
understanding of equivalence of types is not yet complete. We do not
understand that `int | str` is the same as `str | int`, for example.
These tests are in a separate `property_tests::flaky` module.
- Properties can not be formulated in every way possible, due to the
fact that `is_disjoint_from` and `is_subtype_of` can produce false
negative answers.
- The current shrinking implementation is very naive, which leads to
counterexamples that are very long (`str & Any & ~tuple[Any] &
~tuple[Unknown] & ~Literal[""] & ~Literal["a"] | str & int & ~tuple[Any]
& ~tuple[Unknown]`), requiring the developer to simplify manually. It
has not been a major issue so far, but there is a comment in the code
how this can be improved.
- The tests are currently implemented using a macro. This is a single
commit on top which can easily be reverted, if we prefer the plain code
instead. With the macro:
```rs
// `S <: T` implies that `S` can be assigned to `T`.
type_property_test!(
subtype_of_implies_assignable_to, db,
forall types s, t. s.is_subtype_of(db, t) => s.is_assignable_to(db, t)
);
```
without the macro:
```rs
/// `S <: T` implies that `S` can be assigned to `T`.
#[quickcheck]
fn subtype_of_implies_assignable_to(s: Ty, t: Ty) -> bool {
let db = get_cached_db();
let s = s.into_type(&db);
let t = t.into_type(&db);
!s.is_subtype_of(&*db, t) || s.is_assignable_to(&*db, t)
}
```
## Test Plan
```bash
while cargo test --release -p red_knot_python_semantic --features property_tests types::property_tests; do :; done
```
## Summary
`KnownInstance::instance_fallback` may return instances of supertypes.
For example, it returns an instance of `_SpecialForm` for `Literal`.
This means it can't be used on the right-hand side of `is_subtype_of`
relationships, because it might lead to false positives.
I can lead to false negatives on the left hand side of `is_subtype_of`,
but this is at least a known limitation. False negatives are fine for
most applications, but false positives can lead to wrong results in
intersection-simplification, for example.
closes#14731
## Test Plan
Added regression test
## Summary
Simplify tuples containing `Never` to `Never`:
```py
from typing import Never
def never() -> Never: ...
reveal_type((1, never(), "foo")) # revealed: Never
```
I should note that mypy and pyright do *not* perform this
simplification. I don't know why.
There is [only one
place](5137fcc9c8/crates/red_knot_python_semantic/src/types/infer.rs (L1477-L1484))
where we use `TupleType::new` directly (instead of `Type::tuple`, which
changes behavior here). This appears when creating `TypeVar`
constraints, and it looks to me like it should stay this way, because
we're using `TupleType` to store a list of constraints there, instead of
an actual type. We also store `tuple[constraint1, constraint2, …]` as
the type for the `constraint1, constraint2, …` tuple expression. This
would mean that we infer a type of `tuple[str, Never]` for the following
type variable constraints, without simplifying it to `Never`. This seems
like a weird edge case that's maybe not worth looking further into?!
```py
from typing import Never
# vvvvvvvvvv
def f[T: (str, Never)](x: T):
pass
```
## Test Plan
- Added a new unit test. Did not add additional Markdown tests as that
seems superfluous.
- Tested the example above using red knot, mypy, pyright.
- Verified that this allows us to remove `contains_never` from the
property tests
(https://github.com/astral-sh/ruff/pull/14178#discussion_r1866473192)
Resolves https://github.com/astral-sh/ruff/issues/14547 by delegating
narrowing to `E` for `bool(E)` where `E` is some expression.
This change does not include other builtin class constructors which
should also work in this position, like `int(..)` or `float(..)`, as the
original issue does not mention these. It should be easy enough to add
checks for these as well if we want to.
I don't see a lot of markdown tests for malformed input, maybe there's a
better place for the no args and too many args cases to go?
I did see after the fact that it looks like this task was intended for a
new hire.. my apologies. I got here from
https://github.com/astral-sh/ruff/issues/13694, which is marked
help-wanted.
---------
Co-authored-by: David Peter <mail@david-peter.de>
## Summary
Closes: https://github.com/astral-sh/ruff/issues/14593
The final type of a variable after if-statement without explicit else
branch should be similar to having an explicit else branch.
## Test Plan
Originally failed test cases from the bug are added.
---------
Co-authored-by: Carl Meyer <carl@astral.sh>
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
