Carry the typed IDs returned by node constructors through parsing and
recovery. Parse source-ordered argument and parameter lists directly.
Keep ownership of the tree with the parsed module.
Use the vendored runtime and generator instead of the registry dependency,
providing Unicode 17 names with Python-compatible matching. Update workspace
dependencies and tool configuration, add parser regressions, and document how
to update Unicode data when supporting a new Python version.
Use hashbrown's borrowed-entry insertion for long parser names so a
missing name reuses its lookup when it is inserted. Construct the
heap-backed `Name` only on a miss, while preserving the inline bypass
and shared allocations for repeated long names.
## Summary
This PR changes `ruff_python_ast::Name` from a 24-byte `CompactString`
to a 16-byte immutable `CharStr` from
[`astral-sh/char_str`](https://github.com/astral-sh/char_str).
`char_str` is an Astral-maintained fork of
[`lean_string`](https://github.com/ryota2357/lean_string) that provides
compact owned strings with mutable and immutable variants: growable
`CharString` for construction and immutable, exactly allocated `CharStr`
for storage. (It also provides Salsa and get-size features.)
Names up to 16 UTF-8 bytes remain inline, static names remain
allocation-free, and longer names use an exactly sized,
reference-counted allocation with no unused capacity.
Removing eight bytes from every `Name` also shrinks common AST nodes.
For example, `Identifier` and `ExprName` drop from 40 to 32 bytes, while
`ExprAttribute`, `StmtFunctionDef`, `StmtClassDef`, and several other
variants each drop by eight bytes.
The parser interns names that exceed the inline capacity, so repeated
longer names share one backing allocation within a parser invocation.
## Why 16 bytes?
Across the full `mypy_primer` corpus, there 20.8 million name
occurrences:
| UTF-8 length | Occurrences | Share | Unique name/file pairs | Share |
|---|---:|---:|---:|---:|
| 1–16 bytes | 18,955,757 | 91.02% | 3,192,424 | 79.42% |
| 17–24 bytes | 1,348,894 | 6.48% | 529,826 | 13.18% |
| 25+ bytes | 522,376 | 2.51% | 297,575 | 7.40% |
| Total | 20,827,027 | 100% | 4,019,825 | 100% |
The 16-byte cutoff keeps 91% of occurrences inline while saving eight
bytes for every occurrence. The 1–16-byte bucket alone removes 151.6 MB
of handle storage across the corpus.
The tradeoff is the 17–24-byte bucket: `CompactString` stored those
names inline, while `CharStr` allocates once per unique name within a
file. But the interning amortizes that cost when a name repeats. Names
longer than 24 bytes already allocated on `main`; for those names, this
change both reduces the handle size and deduplicates repeated backing
allocations.
## Memory impact
Accounting for the observed per-file multiplicities and structural
sharing, modeled `Name` storage across the corpus changes as follows:
| | `main` | This PR | Difference |
|---|---:|---:|---:|
| `Name` handles and backing storage | 526.18 MB | 366.27 MB | -159.91
MB (-30.4%) |
The current [instrumented memory
report](https://github.com/astral-sh/ruff/pull/26594#issuecomment-4906725175)
shows a consistent reduction in total retained memory:
| Project | `main` | This PR | Difference |
|---|---:|---:|---:|
| flake8 | 29.21 MB | 28.71 MB | -1.69% |
| trio | 71.01 MB | 70.01 MB | -1.41% |
| sphinx | 168.81 MB | 166.82 MB | -1.18% |
| prefect | 454.75 MB | 449.06 MB | -1.25% |
Within those totals, `parsed_module` decreases by roughly 2.1% and
`semantic_index` by 2.4–2.7%. Several name-heavy collections decrease by
considerably more.
## Summary
Use scratch buffers instead of allocating new `Vec`s and shrinking them
when constructing the AST node. While this doesn't remove the need for
copying the elements, it reduces the allocations within our parser
(`shrink_to_fit` often needs to allocate a new buffer and move all
elements). One shared scratch buffer per node type is sufficient,
because our parser is recursive (more nested parse rule always push to
the end, and parse function drain from the end as they complete).
## Performance
Very consistent perf improvement
## Codspeed memory regression
This can cause the parser's peak memory to increase, because we retain
the scratch buffer allocation after parsing a clause completed. However,
that memory is only temporary and released as soon as the parser
completes. Most project-level memory reports show a memory reduction.
It's only specific parser benchmarks that show a memory increase
## Summary
Publish Ruff's production Rust crates to crates.io as part of the
existing release workflow. This follows uv's setup: workspace
dependencies carry registry-compatible version pins, internal crates
receive generated READMEs and an explicit versioning policy, CI dry-runs
the full workspace publish, and release jobs publish missing crate
versions idempotently through crates.io trusted publishing.
The publishing set contains 32 packages: Ruff's CLI, WASM package,
server, linter, formatter, parser, and the internal Ruff component
crates required by them. It also includes `ty_module_resolver`,
`ty_site_packages`, and `ty_static`, because those three crates are part
of Ruff's normal dependency closure.
The publishing set deliberately excludes:
- Test-only crates: `mdtest`, `ruff_mdtest`,
`ruff_python_ast_integration_tests`, and
`ruff_python_trivia_integration_tests`.
- Benchmark and development crates: `ruff_benchmark` and `ruff_dev`.
- All remaining `ty_*` crates.
These packages retain `publish = false`, so bootstrap, version bumping,
README generation, CI dry-runs, and release publishing skip them through
Cargo metadata instead of a separate hard-coded allowlist.
The crates.io job is intentionally independent of release announcement,
so a crates.io outage cannot block the existing binary, PyPI, npm, or
GitHub release outputs. A bootstrap script reserves new crate names,
configures the release workflow as each crate's trusted publisher,
enables trusted-publishing-only releases, and records configured crates
for idempotent reruns.
The full workspace publish dry-run packages and verifies all 32 selected
crates.
Closes https://github.com/astral-sh/ruff/issues/43.
## Summary
Setting `TY_MEMORY_REPORT=full` will generate and print a memory usage
report to the CLI after a `ty check` run:
```
=======SALSA STRUCTS=======
`Definition` metadata=7.24MB fields=17.38MB count=181062
`Expression` metadata=4.45MB fields=5.94MB count=92804
`member_lookup_with_policy_::interned_arguments` metadata=1.97MB fields=2.25MB count=35176
...
=======SALSA QUERIES=======
`File -> ty_python_semantic::semantic_index::SemanticIndex`
metadata=11.46MB fields=88.86MB count=1638
`Definition -> ty_python_semantic::types::infer::TypeInference`
metadata=24.52MB fields=86.68MB count=146018
`File -> ruff_db::parsed::ParsedModule`
metadata=0.12MB fields=69.06MB count=1642
...
=======SALSA SUMMARY=======
TOTAL MEMORY USAGE: 577.61MB
struct metadata = 29.00MB
struct fields = 35.68MB
memo metadata = 103.87MB
memo fields = 409.06MB
```
Eventually, we should integrate these numbers into CI in some form. The
one limitation currently is that heap allocations in salsa structs (e.g.
interned values) are not tracked, but memoized values should have full
coverage. We may also want a peak memory usage counter (that accounts
for non-salsa memory), but that is relatively simple to profile manually
(e.g. `time -v ty check`) and would require a compile-time option to
avoid runtime overhead.
## Summary
This PR adds support for a pragma-style header for inline parser tests
containing JSON-serialized `ParseOptions`. For example,
```python
# parse_options: { "target-version": "3.9" }
match 2:
case 1:
pass
```
The line must start with `# parse_options: ` and then the rest of the
(trimmed) line is deserialized into `ParseOptions` used for parsing the
the test.
## Test Plan
Existing inline tests, plus two new inline tests for
`match-before-py310`.
---------
Co-authored-by: Alex Waygood <alex.waygood@gmail.com>
This is pretty much just moving to the new API and taking care to use
byte offsets. This is *almost* enough. The next commit will fix a bug
involving the handling of unprintable characters as a result of
switching to byte offsets.
## Summary
This PR removes the following dependencies from the `ruff_python_parser`
crate:
* `anyhow` (moved to dev dependencies)
* `is-macro`
* `itertools`
The main motivation is that they aren't used much.
Additionally, it updates the return type of `parse_type_annotation` to
use a more specific `ParseError` instead of the generic `anyhow::Error`.
## Test Plan
`cargo insta test`
## Summary
This PR updates the entire parser stack in multiple ways:
### Make the lexer lazy
* https://github.com/astral-sh/ruff/pull/11244
* https://github.com/astral-sh/ruff/pull/11473
Previously, Ruff's lexer would act as an iterator. The parser would
collect all the tokens in a vector first and then process the tokens to
create the syntax tree.
The first task in this project is to update the entire parsing flow to
make the lexer lazy. This includes the `Lexer`, `TokenSource`, and
`Parser`. For context, the `TokenSource` is a wrapper around the `Lexer`
to filter out the trivia tokens[^1]. Now, the parser will ask the token
source to get the next token and only then the lexer will continue and
emit the token. This means that the lexer needs to be aware of the
"current" token. When the `next_token` is called, the current token will
be updated with the newly lexed token.
The main motivation to make the lexer lazy is to allow re-lexing a token
in a different context. This is going to be really useful to make the
parser error resilience. For example, currently the emitted tokens
remains the same even if the parser can recover from an unclosed
parenthesis. This is important because the lexer emits a
`NonLogicalNewline` in parenthesized context while a normal `Newline` in
non-parenthesized context. This different kinds of newline is also used
to emit the indentation tokens which is important for the parser as it's
used to determine the start and end of a block.
Additionally, this allows us to implement the following functionalities:
1. Checkpoint - rewind infrastructure: The idea here is to create a
checkpoint and continue lexing. At a later point, this checkpoint can be
used to rewind the lexer back to the provided checkpoint.
2. Remove the `SoftKeywordTransformer` and instead use lookahead or
speculative parsing to determine whether a soft keyword is a keyword or
an identifier
3. Remove the `Tok` enum. The `Tok` enum represents the tokens emitted
by the lexer but it contains owned data which makes it expensive to
clone. The new `TokenKind` enum just represents the type of token which
is very cheap.
This brings up a question as to how will the parser get the owned value
which was stored on `Tok`. This will be solved by introducing a new
`TokenValue` enum which only contains a subset of token kinds which has
the owned value. This is stored on the lexer and is requested by the
parser when it wants to process the data. For example:
https://github.com/astral-sh/ruff/blob/8196720f809380d8f1fc7651679ff3fc2cb58cd7/crates/ruff_python_parser/src/parser/expression.rs#L1260-L1262
[^1]: Trivia tokens are `NonLogicalNewline` and `Comment`
### Remove `SoftKeywordTransformer`
* https://github.com/astral-sh/ruff/pull/11441
* https://github.com/astral-sh/ruff/pull/11459
* https://github.com/astral-sh/ruff/pull/11442
* https://github.com/astral-sh/ruff/pull/11443
* https://github.com/astral-sh/ruff/pull/11474
For context,
https://github.com/RustPython/RustPython/pull/4519/files#diff-5de40045e78e794aa5ab0b8aacf531aa477daf826d31ca129467703855408220
added support for soft keywords in the parser which uses infinite
lookahead to classify a soft keyword as a keyword or an identifier. This
is a brilliant idea as it basically wraps the existing Lexer and works
on top of it which means that the logic for lexing and re-lexing a soft
keyword remains separate. The change here is to remove
`SoftKeywordTransformer` and let the parser determine this based on
context, lookahead and speculative parsing.
* **Context:** The transformer needs to know the position of the lexer
between it being at a statement position or a simple statement position.
This is because a `match` token starts a compound statement while a
`type` token starts a simple statement. **The parser already knows
this.**
* **Lookahead:** Now that the parser knows the context it can perform
lookahead of up to two tokens to classify the soft keyword. The logic
for this is mentioned in the PR implementing it for `type` and `match
soft keyword.
* **Speculative parsing:** This is where the checkpoint - rewind
infrastructure helps. For `match` soft keyword, there are certain cases
for which we can't classify based on lookahead. The idea here is to
create a checkpoint and keep parsing. Based on whether the parsing was
successful and what tokens are ahead we can classify the remaining
cases. Refer to #11443 for more details.
If the soft keyword is being parsed in an identifier context, it'll be
converted to an identifier and the emitted token will be updated as
well. Refer
https://github.com/astral-sh/ruff/blob/8196720f809380d8f1fc7651679ff3fc2cb58cd7/crates/ruff_python_parser/src/parser/expression.rs#L487-L491.
The `case` soft keyword doesn't require any special handling because
it'll be a keyword only in the context of a match statement.
### Update the parser API
* https://github.com/astral-sh/ruff/pull/11494
* https://github.com/astral-sh/ruff/pull/11505
Now that the lexer is in sync with the parser, and the parser helps to
determine whether a soft keyword is a keyword or an identifier, the
lexer cannot be used on its own. The reason being that it's not
sensitive to the context (which is correct). This means that the parser
API needs to be updated to not allow any access to the lexer.
Previously, there were multiple ways to parse the source code:
1. Passing the source code itself
2. Or, passing the tokens
Now that the lexer and parser are working together, the API
corresponding to (2) cannot exists. The final API is mentioned in this
PR description: https://github.com/astral-sh/ruff/pull/11494.
### Refactor the downstream tools (linter and formatter)
* https://github.com/astral-sh/ruff/pull/11511
* https://github.com/astral-sh/ruff/pull/11515
* https://github.com/astral-sh/ruff/pull/11529
* https://github.com/astral-sh/ruff/pull/11562
* https://github.com/astral-sh/ruff/pull/11592
And, the final set of changes involves updating all references of the
lexer and `Tok` enum. This was done in two-parts:
1. Update all the references in a way that doesn't require any changes
from this PR i.e., it can be done independently
* https://github.com/astral-sh/ruff/pull/11402
* https://github.com/astral-sh/ruff/pull/11406
* https://github.com/astral-sh/ruff/pull/11418
* https://github.com/astral-sh/ruff/pull/11419
* https://github.com/astral-sh/ruff/pull/11420
* https://github.com/astral-sh/ruff/pull/11424
2. Update all the remaining references to use the changes made in this
PR
For (2), there were various strategies used:
1. Introduce a new `Tokens` struct which wraps the token vector and add
methods to query a certain subset of tokens. These includes:
1. `up_to_first_unknown` which replaces the `tokenize` function
2. `in_range` and `after` which replaces the `lex_starts_at` function
where the former returns the tokens within the given range while the
latter returns all the tokens after the given offset
2. Introduce a new `TokenFlags` which is a set of flags to query certain
information from a token. Currently, this information is only limited to
any string type token but can be expanded to include other information
in the future as needed. https://github.com/astral-sh/ruff/pull/11578
3. Move the `CommentRanges` to the parsed output because this
information is common to both the linter and the formatter. This removes
the need for `tokens_and_ranges` function.
## Test Plan
- [x] Update and verify the test snapshots
- [x] Make sure the entire test suite is passing
- [x] Make sure there are no changes in the ecosystem checks
- [x] Run the fuzzer on the parser
- [x] Run this change on dozens of open-source projects
### Running this change on dozens of open-source projects
Refer to the PR description to get the list of open source projects used
for testing.
Now, the following tests were done between `main` and this branch:
1. Compare the output of `--select=E999` (syntax errors)
2. Compare the output of default rule selection
3. Compare the output of `--select=ALL`
**Conclusion: all output were same**
## What's next?
The next step is to introduce re-lexing logic and update the parser to
feed the recovery information to the lexer so that it can emit the
correct token. This moves us one step closer to having error resilience
in the parser and provides Ruff the possibility to lint even if the
source code contains syntax errors.
## Summary
This PR adds a new `ExpressionContext` struct which is used in
expression parsing.
This solves the following problem:
1. Allowing starred expression with different precedence
2. Allowing yield expression in certain context
3. Remove ambiguity with `in` keyword when parsing a `for ... in`
statement
For context, (1) was solved by adding `parse_star_expression_list` and
`parse_star_expression_or_higher` in #10623, (2) was solved by by adding
`parse_yield_expression_or_else` in #10809, and (3) was fixed in #11009.
All of the mentioned functions have been removed in favor of the context
flags.
As mentioned in #11009, an ideal solution would be to implement an
expression context which is what this PR implements. This is passed
around as function parameter and the call stack is used to automatically
reset the context.
### Recovery
How should the parser recover if the target expression is invalid when
an expression can consume the `in` keyword?
1. Should the `in` keyword be part of the target expression?
2. Or, should the expression parsing stop as soon as `in` keyword is
encountered, no matter the expression?
For example:
```python
for yield x in y: ...
# Here, should this be parsed as
for (yield x) in (y): ...
# Or
for (yield x in y): ...
# where the `in iter` part is missing
```
Or, for binary expression parsing:
```python
for x or y in z: ...
# Should this be parsed as
for (x or y) in z: ...
# Or
for (x or y in z): ...
# where the `in iter` part is missing
```
This need not be solved now, but is very easy to change. For context
this PR does the following:
* For binary, comparison, and unary expressions, stop at `in`
* For lambda, yield expressions, consume the `in`
## Test Plan
1. Add test cases for the `for ... in` statement and verify the
snapshots
2. Make sure the existing test suite pass
3. Run the fuzzer for around 3000 generated source code
4. Run the updated logic on a dozen or so open source repositories
(codename "parser-checkouts")
(Supersedes #9152, authored by @LaBatata101)
## Summary
This PR replaces the current parser generated from LALRPOP to a
hand-written recursive descent parser.
It also updates the grammar for [PEP
646](https://peps.python.org/pep-0646/) so that the parser outputs the
correct AST. For example, in `data[*x]`, the index expression is now a
tuple with a single starred expression instead of just a starred
expression.
Beyond the performance improvements, the parser is also error resilient
and can provide better error messages. The behavior as seen by any
downstream tools isn't changed. That is, the linter and formatter can
still assume that the parser will _stop_ at the first syntax error. This
will be updated in the following months.
For more details about the change here, refer to the PR corresponding to
the individual commits and the release blog post.
## Test Plan
Write _lots_ and _lots_ of tests for both valid and invalid syntax and
verify the output.
## Acknowledgements
- @MichaReiser for reviewing 100+ parser PRs and continuously providing
guidance throughout the project
- @LaBatata101 for initiating the transition to a hand-written parser in
#9152
- @addisoncrump for implementing the fuzzer which helped
[catch](https://github.com/astral-sh/ruff/pull/10903)
[a](https://github.com/astral-sh/ruff/pull/10910)
[lot](https://github.com/astral-sh/ruff/pull/10966)
[of](https://github.com/astral-sh/ruff/pull/10896)
[bugs](https://github.com/astral-sh/ruff/pull/10877)
---------
Co-authored-by: Victor Hugo Gomes <labatata101@linuxmail.org>
Co-authored-by: Micha Reiser <micha@reiser.io>
## Summary
This PR modifies our `Cargo.toml` files to use workspace dependencies
for _all_ dependencies, rather than the status quo of sporadically
trying to use workspace dependencies for those dependencies that are
used across multiple crates. I find the current situation more confusing
and harder to manage, since we have a mix of workspace and crate-local
dependencies, whereas this setup consistently uses the same approach for
all dependencies.
Update to [Rust
1.74](https://blog.rust-lang.org/2023/11/16/Rust-1.74.0.html) and use
the new clippy lints table.
The update itself introduced a new clippy lint about superfluous hashes
in raw strings, which got removed.
I moved our lint config from `rustflags` to the newly stabilized
[workspace.lints](https://doc.rust-lang.org/stable/cargo/reference/workspaces.html#the-lints-table).
One consequence is that we have to `unsafe_code = "warn"` instead of
"forbid" because the latter now actually bans unsafe code:
```
error[E0453]: allow(unsafe_code) incompatible with previous forbid
--> crates/ruff_source_file/src/newlines.rs:62:17
|
62 | #[allow(unsafe_code)]
| ^^^^^^^^^^^ overruled by previous forbid
|
= note: `forbid` lint level was set on command line
```
---------
Co-authored-by: Charlie Marsh <charlie.r.marsh@gmail.com>