Using length comparison to narrow types for Tuples and Literals #10367

hatal175 · 2021-04-25T13:28:58Z

Description

This PR uses length based comparisons with Literals to narrow down types in branches. It is similar to the way mypy does it with isinstance.
There are three types of narrowing done:

Choosing Tuples from a union according to length
Reducing variant length tuples to specific sized tuples
Choosing Literals according to length

Test Plan

I've added unit tests based on both the use cases and bugs mypy primer found.

DevilXD · 2021-04-26T15:43:51Z

As someone who just ran into #1178 myself, this PR is a +1 from me, assuming it's correct and works.

JukkaL · 2021-05-05T15:40:12Z

Thanks for the PR! Since we are (hopefully) close to a release, I won't review and merge this PR just yet -- improvements to type inference can have tricky edge cases that may slow down the release.

MaddyGuthridge · 2022-02-08T01:46:47Z

Would love to see this get added soon - it'd improve my program's readability massively

hatal175 · 2022-04-08T23:09:13Z

I've added this pull request almost a year ago and there doesn't seem to be much progress with it. Has it been reviewed? Does it need fixes? What's up?

DevilXD · 2022-04-09T07:49:14Z

@hatal175 I think that may be because some tests on this one have failed. Here's the output from the Windows x64 test suite:

================================== FAILURES ===================================
___________________ testNarrowingLenBiggerThanVariantTuple ____________________
[gw1] win32 -- Python 3.7.9 D:\a\mypy\mypy\.tox\py37\Scripts\python.EXE
data: D:\a\mypy\mypy\test-data\unit\check-narrowing.test:1274:
D:\a\mypy\mypy\mypy\test\testcheck.py:140: in run_case
    self.run_case_once(testcase)
D:\a\mypy\mypy\mypy\test\testcheck.py:227: in run_case_once
    assert_string_arrays_equal(output, a, msg.format(testcase.file, testcase.line))
E   AssertionError: Unexpected type checker output (D:\a\mypy\mypy\test-data\unit\check-narrowing.test, line 1274)
---------------------------- Captured stderr call -----------------------------
Expected:
  main:10: note: Revealed type is "builtins.tuple[builtins.int]" (diff)
  main:12: note: Revealed type is "builtins.tuple[builtins.int]" (diff)
Actual:
  main:10: note: Revealed type is "builtins.tuple[builtins.int, ...]" (diff)
  main:12: note: Revealed type is "builtins.tuple[builtins.int, ...]" (diff)

Alignment of first line difference:
  E: ...s.tuple[builtins.int]"
  A: ...s.tuple[builtins.int, ...]"
                            ^
_____________________ testNarrowingLenVariantLengthTuple ______________________
[gw0] win32 -- Python 3.7.9 D:\a\mypy\mypy\.tox\py37\Scripts\python.EXE
data: D:\a\mypy\mypy\test-data\unit\check-narrowing.test:1156:
D:\a\mypy\mypy\mypy\test\testcheck.py:140: in run_case
    self.run_case_once(testcase)
D:\a\mypy\mypy\mypy\test\testcheck.py:227: in run_case_once
    assert_string_arrays_equal(output, a, msg.format(testcase.file, testcase.line))
E   AssertionError: Unexpected type checker output (D:\a\mypy\mypy\test-data\unit\check-narrowing.test, line 1156)
---------------------------- Captured stderr call -----------------------------
Expected:
  main:9: note: Revealed type is "Tuple[builtins.int, builtins.int, builtins....
  main:11: note: Revealed type is "builtins.tuple[builtins.int]" (diff)
  main:14: note: Revealed type is "builtins.tuple[builtins.int]" (diff)
  main:16: note: Revealed type is "Tuple[builtins.int, builtins.int, builtins...
Actual:
  main:9: note: Revealed type is "Tuple[builtins.int, builtins.int, builtins....
  main:11: note: Revealed type is "builtins.tuple[builtins.int, ...]" (diff)
  main:14: note: Revealed type is "builtins.tuple[builtins.int, ...]" (diff)
  main:16: note: Revealed type is "Tuple[builtins.int, builtins.int, builtins...

Alignment of first line difference:
  E: ...s.tuple[builtins.int]"
  A: ...s.tuple[builtins.int, ...]"
                            ^
=========================== short test summary info ===========================
FAILED mypy/test/testcheck.py::TypeCheckSuite::check-narrowing.test::testNarrowingLenBiggerThanVariantTuple
FAILED mypy/test/testcheck.py::TypeCheckSuite::check-narrowing.test::testNarrowingLenVariantLengthTuple
===== 2 failed, 9659 passed, 375 skipped, 7 xfailed in 1548.55s (0:25:48) =====
ERROR: InvocationError for command 'D:\a\mypy\mypy\.tox\py37\Scripts\python.EXE' -m pytest (exited with code 1)

In general, this section should be showing that all 14 checks were successful, and the branch needs to not have any conflicts, for a proper review to happen:

You can re-run the tests by making a push to this PR's branch - please note that a push can consist of several commits, so if you intend on fixing a bunch of things in separate commits and know that the tests will fail in-between them, it's best to hold on with any pushes until after you make all commits that you think will make the test suite pass this time. This just helps to save on unnecessary processing from the test suite side.

Side note: I've been following this PR closely as I've ran into several cases where this would be really helpful for my purposes. Thank you for it 🙂

DevilXD · 2022-04-09T07:57:13Z

Also, in addition to fixing the cause of the failing tests, I'd suggest taking a look at this specific primer output, since it may signify requiring a change too:

kornia (https://github.com/kornia/kornia)
+ kornia/augmentation/container/image.py:130: error: Value of type <nothing> is not indexable  [index]
+ kornia/augmentation/container/image.py:131: error: Value of type <nothing> is not indexable  [index]

Here's the relevant code from that point in time: https://github.com/kornia/kornia/blob/406f03aa3da709dddeee4a6992a1158d2054fc63/kornia/augmentation/container/image.py#L130-L131

hatal175 · 2022-04-09T09:43:27Z

Fixed tests and typing due to changes from last rebase done here (nothing ground breaking).
The primer error is an error in the codebase (random_apply should be a union of Tuple[int] as well).

github-actions · 2022-04-09T10:09:48Z

Diff from mypy_primer, showing the effect of this PR on open source code:

rich (https://github.com/willmcgugan/rich)
+ rich/padding.py:69: error: Redundant cast to "Tuple[int, int]"  [redundant-cast]
+ rich/padding.py:72: error: Redundant cast to "Tuple[int, int, int, int]"  [redundant-cast]

rotki (https://github.com/rotki/rotki)
+ rotkehlchen/api/server.py:284: error: Unused "type: ignore" comment
+ rotkehlchen/api/server.py:287: error: Unused "type: ignore" comment

anyio (https://github.com/agronholm/anyio)
+ src/anyio/_core/_sockets.py:499: error: Redundant cast to "Tuple[str, int, int, int]"  [redundant-cast]
+ src/anyio/_core/_sockets.py:506: error: Redundant cast to "Tuple[str, int]"  [redundant-cast]

kornia (https://github.com/kornia/kornia)
+ kornia/augmentation/container/image.py:132: error: Value of type <nothing> is not indexable  [index]
+ kornia/augmentation/container/image.py:133: error: Value of type <nothing> is not indexable  [index]

urllib3 (https://github.com/urllib3/urllib3)
+ src/urllib3/fields.py:236: error: Redundant cast to "Tuple[str, Union[str, bytes], str]"  [redundant-cast]
+ src/urllib3/fields.py:240: error: Redundant cast to "Tuple[str, Union[str, bytes]]"  [redundant-cast]

aiohttp (https://github.com/aio-libs/aiohttp)
+ aiohttp/resolver.py:47: error: Unused "type: ignore" comment

black (https://github.com/psf/black)
+ src/blib2to3/pgen2/tokenize.py:253:29: error: Redundant cast to "Tuple[int, str]"
+ src/blib2to3/pgen2/tokenize.py:255:49: error: Redundant cast to "Tuple[int, str, Tuple[int, int], Tuple[int, int], str]"

ilevkivskyi

Looks very good, thanks! I did a quick pass and left some comments. Also please resolve the merge conflicts and fix this mypy_primer output:

kornia (https://github.com/kornia/kornia)
+ kornia/augmentation/container/image.py:132: error: Value of type <nothing> is not indexable  [index]
+ kornia/augmentation/container/image.py:133: error: Value of type <nothing> is not indexable  [index]

ilevkivskyi · 2023-08-12T14:50:17Z

mypy/checker.py

+                    refers_to_fullname(expr.callee, 'builtins.len') and
+                    len(expr.args) == 1 and
+                    is_narrowable_literal(collapse_walrus(expr.args[0]))
+                )


Mypyc generates a slow code for nested functions. Can you move these two out of the class, e.g. put them near len_of_type()?

ilevkivskyi · 2023-08-12T14:51:41Z

mypy/checker.py

+                    ind in narrowable_len_operand_index_to_hash
+                    for ind in expr_indices)
+
+                if not is_len_check_expression:


FWIW I don't like huge if blocks. Can you factor out the corresponding logic in this block into a helper method?

ilevkivskyi · 2023-08-12T14:52:43Z

mypy/checker.py

+                continue
+
+            # we already checked that operand[i] is CallExpr since it is narrowable
+            expr = operands[i].args[0]  # type: ignore[attr-defined]


Please don't use # type: ignore, use assert isinstance(..., CalExpr) instead.

ilevkivskyi · 2023-08-12T14:56:59Z

mypy/checker.py

@@ -5774,6 +5969,17 @@ def conditional_types_to_typemaps(expr: Expression,
    return cast(Tuple[TypeMap, TypeMap], tuple(maps))


+def len_of_type(typ: Type) -> Optional[int]:
+    """Takes a type and returns an int that represents the length
+    of instances of that type or None if not applicable or variant length"""


Please use a single line docstring here, for example Return the length of objects of given type if known statically.

Also please replace "variant length" -> "variable length" everywhere in the PR (including test cases).

ilevkivskyi · 2023-08-12T14:58:27Z

mypy/checker.py

+    if isinstance(proper_type, TupleType):
+        return len(proper_type.items)
+    if isinstance(proper_type, LiteralType) and isinstance(proper_type.value, str):
+        return len(proper_type.value)


I would propose to leave out literal strings from the initial version of this feature. It is quite niche and should be easy to add if people will ask. You can leave a TODO item here to add support for literal strings in the future.

TeamSpen210 · 2023-08-12T21:04:12Z

mypy/checker.py

+            return typ
+        proper_type = get_proper_type(typ)
+        if (isinstance(proper_type, Instance) and proper_type.type.fullname == "builtins.tuple"
+                and length >= 0):


I notice this excludes zero-length tuples, and it doesn't seem like they're tested. We might want a special case for them, though it's not too likely in code that you'd do anything with the result.

ilevkivskyi · 2023-10-08T18:27:24Z

Superseded by #16237 (note I am not including LiteralType support in the initial implementation)

@hatal175

Fixes #1178 Supersedes #10367 This is includes implementation for fixed length tuples, homogeneous tuples, and variadic tuples (and combinations of those). Generally implementation is straightforward. Some notes: * Unfortunately, it is necessary to add a new attribute `min_len` to `TypeVarTupleType`, which is probably fine, as it doesn't have that many attributes so far. * Supporting more general use cases (like `>` comparisons for variadic tuples) can cause quick proliferation of unions. I added two mechanisms to counteract this: not applying the narrowing if the integer literal in comparison is itself large, and collapsing unions of tuples into a single tuple (if possible) after we are done with the narrowing. This looks a bit arbitrary, but I think it is important to have. * Main missing feature here is probably not inferring type information from indirect comparisons like `len(x) > foo() > 1`. Supporting this kind of things in full generality is cumbersome, and we may add cases that turn out to be important later. * Note I am quite careful with indexing "inside" a `TypeVarTuple`, it is not really needed now, but I wanted to make everything future proof, so that it will be easy to add support for upper bounds for `TypeVarTuple`s, like `Nums = TypeVarTuple("Nums", bound=tuple[float, ...])`. * I also fix couple existing inconsistencies with `Any` handling in type narrowing. It looks like they stem from the old incorrect logic that meet of `Any` and `X` should be `X`, while in fact it should be `Any`. These fixes are not strictly necessary, but otherwise there may be new false positives, because I introduce a bunch of additional type narrowing scenarios here. cc @hatal175, thanks for the test cases, and for your nice first attempt to implement this! Co-authored-by: Tal Hayon <talhayon1@gmail.com>

This comment has been minimized.

Sign in to view

vnmabus mentioned this pull request Jul 22, 2021

Typing rect. Tuxemon/Tuxemon#921

Merged

This comment has been minimized.

Sign in to view

GBeauregard mentioned this pull request Oct 22, 2021

tuple length type narrowing microsoft/pyright#2474

Closed

97littleleaf11 closed this Jan 12, 2022

97littleleaf11 reopened this Jan 12, 2022

This comment has been minimized.

Sign in to view

hatal175 added 3 commits April 9, 2022 11:14

Narrow down types based on len

49430f8

Fix mypy primer bugs

e763fbc

Make other comparison operators work

41e00fc

hatal175 force-pushed the NarrowingTypeLen branch from a040393 to 42b3aa6 Compare April 9, 2022 09:41

Fix tests and typing

884d47c

hatal175 force-pushed the NarrowingTypeLen branch from 42b3aa6 to 884d47c Compare April 9, 2022 09:49

This comment has been minimized.

Sign in to view

AlexWaygood requested a review from JukkaL April 9, 2022 10:09

hauntsaninja added the upnext label Apr 21, 2023

ilevkivskyi reviewed Aug 12, 2023

View reviewed changes

TeamSpen210 reviewed Aug 12, 2023

View reviewed changes

ilevkivskyi mentioned this pull request Sep 9, 2023

Fix tuple subtyping #16073

Closed

ilevkivskyi mentioned this pull request Oct 8, 2023

Narrow tuple types using len() #16237

Merged

ilevkivskyi closed this Oct 8, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Using length comparison to narrow types for Tuples and Literals #10367

Using length comparison to narrow types for Tuples and Literals #10367

hatal175 commented Apr 25, 2021

This comment has been minimized.

DevilXD commented Apr 26, 2021

JukkaL commented May 5, 2021

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

MaddyGuthridge commented Feb 8, 2022

hatal175 commented Apr 8, 2022

DevilXD commented Apr 9, 2022 •

edited

Loading

DevilXD commented Apr 9, 2022

hatal175 commented Apr 9, 2022

This comment has been minimized.

github-actions bot commented Apr 9, 2022

ilevkivskyi left a comment

ilevkivskyi Aug 12, 2023

ilevkivskyi Aug 12, 2023

ilevkivskyi Aug 12, 2023

ilevkivskyi Aug 12, 2023

ilevkivskyi Aug 12, 2023

TeamSpen210 Aug 12, 2023

ilevkivskyi commented Oct 8, 2023

Using length comparison to narrow types for Tuples and Literals #10367

Using length comparison to narrow types for Tuples and Literals #10367

Conversation

hatal175 commented Apr 25, 2021

Description

Test Plan

This comment has been minimized.

DevilXD commented Apr 26, 2021

JukkaL commented May 5, 2021

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

MaddyGuthridge commented Feb 8, 2022

hatal175 commented Apr 8, 2022

DevilXD commented Apr 9, 2022 • edited Loading

DevilXD commented Apr 9, 2022

hatal175 commented Apr 9, 2022

This comment has been minimized.

github-actions bot commented Apr 9, 2022

ilevkivskyi left a comment

Choose a reason for hiding this comment

ilevkivskyi Aug 12, 2023

Choose a reason for hiding this comment

ilevkivskyi Aug 12, 2023

Choose a reason for hiding this comment

ilevkivskyi Aug 12, 2023

Choose a reason for hiding this comment

ilevkivskyi Aug 12, 2023

Choose a reason for hiding this comment

ilevkivskyi Aug 12, 2023

Choose a reason for hiding this comment

TeamSpen210 Aug 12, 2023

Choose a reason for hiding this comment

ilevkivskyi commented Oct 8, 2023

DevilXD commented Apr 9, 2022 •

edited

Loading