The Mypy Blog: May 2026

We’ve just uploaded mypy 2.0.0 to the Python Package Index (PyPI). Mypy is a static type checker for Python. This release includes new features, performance improvements and bug fixes. There are also changes to options and defaults. You can install it as follows:

python3 -m pip install -U mypy

You can read the full documentation for this release on Read the Docs.

Enable `--local-partial-types` by Default

This flag affects the inference of types based on assignments in other scopes. For now, explicitly disabling this continues to be supported, but this support will be removed in the future as the legacy behaviour is hard to support with other current and future features in mypy, like the daemon or the new implementation of flexible redefinitions.

Contributed by Ivan Levkivskyi, Jukka Lehtosalo, Shantanu in PR 21163.

Enable `--strict-bytes` by Default

Per PEP 688, mypy no longer treats bytearray and memoryview values as assignable to the bytes type.

Contributed by Shantanu in PR 18371.

New Behavior for `--allow-redefinition`

The --allow-redefinition flag now behaves like --allow-redefinition-new in mypy 1.20 and earlier. The new behavior is generally more flexible. For example, you can have different types for a variable in different blocks:

# mypy: allow-redefinition

def foo(cond: bool) -> None:
    if cond:
        for x in ["a", "b"]:
            # Type of "x" is "str" here
            ...
    else:
        for x in [1, 2]:
            # Type of "x" is "int" here
            ...

The new behavior requires --local-partial-types, which is now enabled by default.

However, --allow-redefinition doesn't allow giving two type annotations for the same variable. The old behavior (sometimes) allows this. Code like this now generates an error when using --allow-redefinition:

def foo() -> None:
    x: list[int] = []
    ...
    x: list[str] = []  # Error: "x" redefined
    ...

You can still use --allow-redefinition-old to fall back to the old behavior. We have no plans to remove the legacy behavior, but the old functionality is maintained on a best effort basis.

Contributed by Jukka Lehtosalo in PR 21276.

Parallel Type Checking

Mypy now supports experimental parallel and incremental type checking. Use --num-workers N or -nN to use N worker processes to type check in parallel. The speedup depends on the import structure of your codebase and your environment, but for large projects we've seen performance gains of up to 5x when using 8 worker processes.

Parallel type checking implicitly enables the new native parser. There are still some minor semantic differences between parallel and non-parallel modes, which we will be fixing in future mypy releases.

Contributed by Ivan Levkivskyi, with additional contributions from Emma Smith and Jukka Lehtosalo.

Recent related changes since the last release:

Freeze garbage collection in parallel workers for 4-5% speedup (Ivan Levkivskyi, PR 21302)
Expose --num-workers and --native-parser (Ivan Levkivskyi, PR 21387)
Split type checking into interface and implementation in parallel workers (Ivan Levkivskyi, PR 21119)
Batch module groups for parallel processing (Ivan Levkivskyi, PR 21287)
Optimize parallel worker startup (Ivan Levkivskyi, PR 21203)
Parse files in parallel when possible (Ivan Levkivskyi, PR 21175)
Use parallel parsing at all stages (Ivan Levkivskyi, PR 21266)
Fix sequential bottleneck in parallel parsing (Jukka Lehtosalo, PR 21291)
Fail fast when a user tries to generate reports with parallel workers (Ivan Levkivskyi, PR 21341)
Partially support old NumPy plugin in parallel type checking (Ivan Levkivskyi, PR 21324)
Handle reachability consistently in parallel type checking (Ivan Levkivskyi, PR 21322)
Always respect @no_type_check in parallel type checking (Ivan Levkivskyi, PR 21320)
Minor fixes in parallel checking (Ivan Levkivskyi, PR 21319)
Fix plugin logic in parallel type checking (Ivan Levkivskyi, PR 21252)
Fix Windows IPC race condition when using parallel checking (Jukka Lehtosalo, PR 21228)
Report parallel worker exit status on receive failure (Jukka Lehtosalo, PR 21224)

Drop Support for Targeting Python 3.9

Mypy no longer supports type checking code with --python-version 3.9. Use --python-version 3.10 or newer.

Contributed by Shantanu, Marc Mueller in PR 21243.

Remove Special Casing of Legacy Bundled Stubs

Mypy used to bundle stubs for a few packages in versions 0.812 and earlier. To navigate the transition, mypy used to report missing types for these packages even if --ignore-missing-imports was set. Mypy now consistently respects --ignore-missing-imports for all packages.

Contributed by Shantanu in PR 18372.

Prevent Assignment to None for Non-Optional Class Variables with Type Comments

Mypy used to allow assignment to None for class variables when using type comments. This was a common idiom in Python 3.5 and earlier, prior to the introduction of variable annotations. However, this was a soundness hole and has now been removed.

Contributed by Shantanu in PR 20054.

librt.strings: String and Bytes Primitives for Mypyc

In mypy 1.20, we introduced librt as a standard library for mypyc that fills in some gaps in the Python standard library and the C API. This release adds the new module librt.strings, which contains utilities for building string and bytes objects, and for accessing and generating binary data:

StringWriter and BytesWriter classes allow quickly building str and bytes objects from parts.
read_* and write_* functions provide fast reading and writing of binary-encoded data.

Refer to the documentation for the details.

Contributed by Jukka Lehtosalo.

Mypyc Improvements

Document librt.time (Jukka Lehtosalo, PR 21372)
Mark librt.time.time() non-experimental (Ivan Levkivskyi, PR 21310)
Fix librt.time primitive now that it is no longer experimental (Ivan Levkivskyi, PR 21318)
Fix librt API/ABI version checks (Jukka Lehtosalo, PR 21311)
Generate more type methods for classes with attribute dictionaries (Piotr Sawicki, PR 21290)
Fix reference counting for tuple items during deallocation (Shantanu, PR 21245)
Release new instances when __init__ raises (Shantanu, PR 21248)
Fix @property getter memory leak (Vaggelis Danias, PR 21230)
Fix semantics for walrus expression in tuple (Shantanu, PR 21249)
Fix crash on import errors during cleanup (Shantanu, PR 21247)
Fix reference leak in str index (Shantanu, PR 21251)
Fix memory leak in integer true division (Shantanu, PR 21246)
Fix reference leaks in list.clear()/dict.clear() (Shantanu, PR 21244)
Resolve type aliases in function specialization (esarp, PR 21233)
Report an error if an acyclic class inherits from non-acyclic (Piotr Sawicki, PR 21227)
Fix b64decode to match new CPython behavior (Piotr Sawicki, PR 21200)

Fixes to Crashes

Fix crash when a file does not exist during semantic analysis (Ivan Levkivskyi, PR 21379)
Fix parallel worker crash on syntax error (Ivan Levkivskyi, PR 21202)

Changes to Messages

Improve error messages for unexpected keyword arguments in overloaded functions (Kevin Kannammalil, PR 20592)
Don't suggest Foo[...] when Foo(arg=...) is used in annotation (Yosof Badr, PR 21238)
Mention what codes are actually ignored in "not covered by type: ignore comment" note (wyattscarpenter, PR 19904)
Improve error messages when positional argument is missing (Kevin Kannammalil, PR 20591)
Improve "name is not defined" errors with fuzzy matching (Kevin Kannammalil, PR 20693)
Add suggestions for misspelled module imports (Kevin Kannammalil, PR 20695)

Performance Improvements

Replace NamedTuple with faster regular classes in hot paths (Shantanu, PR 21326)
Avoid calling best-match suggestions unless the message is shown (Ivan Levkivskyi, PR 21307)
Order cases in native parser based on AST node frequency (Jukka Lehtosalo, PR 21219)

Stubtest Improvements

Basic support for unpack kwargs (Shantanu, PR 21024)
Fix false positive for properties with a deleter (Pranav Manglik, PR 21259)

Documentation Updates

Rename "value restriction" to "value-constrained type variable" (Leo Ji, PR 21112)
Clarify that invariant-by-default applies to legacy TypeVar syntax (Leo Ji, PR 21108)

Improvements to the Native Parser

The new native parser is still experimental.

Make new parser consistent with the old one (Ivan Levkivskyi, PR 21377)
Support --package-root with the native parser (Ivan Levkivskyi, PR 21321)
Improve call expressions in type annotations with the native parser (Jukka Lehtosalo, PR 21300)
Depend on ast-serialize by default (Jukka Lehtosalo, PR 21297)

Other Notable Fixes and Improvements

Fix narrowing for AbstractSet and Mapping (Shantanu, PR 21352)
Preserve gradual guarantee when narrowing Any union via equality (Shantanu, PR 21368)
Make type variable upper bound narrowing symmetric (Ivan Levkivskyi, PR 21350)
Behave consistently when type-checking a stub package directly (Ivan Levkivskyi, PR 21330)
Add support for Final[...] in dataclasses (Ivan Levkivskyi, PR 21334)
Narrow more sequence parents (Shantanu, PR 21327)
Better narrowing for enums and other types with known equality (Shantanu, PR 21281)
Fix pathspec error (Ivan Levkivskyi, PR 21296)
Use sharding for the SQLite cache (Jukka Lehtosalo, PR 21292)
Limit type inference context fallback to the walrus operator only (Ivan Levkivskyi, PR 21294)
Support .git/info/exclude for --exclude-gitignore (RogerJinIS, PR 21286)
Let --allow-redefinition widen a global in a function with None initialization (Jukka Lehtosalo, PR 21285)
Delete Python 2 extra (Shantanu, PR 18374)
No longer narrow final globals in functions (Ivan Levkivskyi, PR 21241)
Narrow unions containing Any in conditional branches (Shantanu, PR 21231)
Propagate narrowing within chained comparisons (Shantanu, PR 21160)
Add proper lazy deserialization (Ivan Levkivskyi, PR 21198)
Add install_types to options affecting cache (Brian Schubert, PR 21070)
Narrow Any in conditional type checks (Shantanu, PR 21167)
Fix exception handler target location in new parser (Ivan Levkivskyi, PR 21185)
Improve traceback display (Shantanu, PR 21155)
Include two more files in the sdist: CREDITS and the typeshed README (Michael R. Crusoe, PR 21131)

Typeshed Updates

Please see git log for full list of standard library typeshed stub changes.

Acknowledgements

Thanks to all mypy contributors who contributed to this release:

Brian Schubert
Ethan Sarp
Ivan Levkivskyi
Jukka Lehtosalo
Kevin Kannammalil
Leo Ji
Marc Mueller
Michael R. Crusoe
Piotr Sawicki
Pranav Manglik
RogerJinIS
Shantanu
Vaggelis Danias
wyattscarpenter
Yosof Badr

I’d also like to thank my employer, Dropbox, for supporting mypy development.

The Mypy Blog

Wednesday, 6 May 2026

Mypy 2.0 Relased

Enable `--local-partial-types` by Default

Enable `--strict-bytes` by Default

New Behavior for `--allow-redefinition`

Parallel Type Checking

Drop Support for Targeting Python 3.9

Remove Special Casing of Legacy Bundled Stubs

Prevent Assignment to None for Non-Optional Class Variables with Type Comments

librt.strings: String and Bytes Primitives for Mypyc

Mypyc Improvements

Fixes to Crashes

Changes to Messages

Performance Improvements

Stubtest Improvements

Documentation Updates

Improvements to the Native Parser

Other Notable Fixes and Improvements

Typeshed Updates

Acknowledgements

Contributors

Wednesday, 6 May 2026

Mypy 2.0 Relased

Enable --local-partial-types by Default

Enable --strict-bytes by Default

New Behavior for --allow-redefinition

Parallel Type Checking

Drop Support for Targeting Python 3.9

Remove Special Casing of Legacy Bundled Stubs

Prevent Assignment to None for Non-Optional Class Variables with Type Comments

librt.strings: String and Bytes Primitives for Mypyc

Mypyc Improvements

Fixes to Crashes

Changes to Messages

Performance Improvements

Stubtest Improvements

Documentation Updates

Improvements to the Native Parser

Other Notable Fixes and Improvements

Typeshed Updates

Acknowledgements

Enable `--local-partial-types` by Default

Enable `--strict-bytes` by Default

New Behavior for `--allow-redefinition`