Speed up make_simplified_union, fix recursive tuple crash by hauntsaninja · Pull Request #15128 · python/mypy

hauntsaninja · 2023-04-25T16:21:43Z

The following code optimises make_simplified_union in the common case that there are exact duplicates in the union. In this regard, this is similar to #15104

There's a behaviour change in one unit test. I think it's good? We'll see what mypy_primer has to say.

To get this to work, I needed to use partial tuple fallbacks in a couple places (these maybe had the potential to be latent crashes anyway?) There were some interesting things going on with recursive type aliases and type state assumptions

This is about a 25% speedup on the pydantic codebase and about a 2% speedup on self check (measured with uncompiled mypy)

The following code optimises make_simplified_union in the common case that there are exact duplicates in the union. In this regard, this is similar to python#15104 To get this to work, I needed to use partial tuple fallbacks in a couple places (these maybe had the potential to be latent crashes anyway?) There were some interesting things going on with recursive type aliases and type state assumptions This is about a 25% speedup on the pydantic codebase and about a 2% speedup on self check (measured with uncompiled mypy)

for more information, see https://pre-commit.ci

hauntsaninja · 2023-04-25T17:34:10Z

Looks like there's a major performance regression in materialize

for more information, see https://pre-commit.ci

github-actions · 2023-04-25T23:59:20Z

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

pandas-stubs (https://github.com/pandas-dev/pandas-stubs) got 1.14 faster (68.0s -> 59.5s)

speedrun.com_global_scoreboard_webapp (https://github.com/Avasam/speedrun.com_global_scoreboard_webapp) got 1.24 faster (43.8s -> 35.2s)

hauntsaninja · 2023-04-26T05:22:10Z

(note to self: once this is merged, rename is_simple_literal to is_simple_str_literal or see if we can get rid of it)

JukkaL · 2023-05-03T13:39:12Z

I did a measurement using misc/perf-compare.py and this seems roughly performance-neutral for self check using a compiled mypy:

=== Results ===

hauntsaninja-msu          4.393s (0.0%)
master                    4.405s (+0.3%)

JukkaL

Looks good! I'm happy to see additional performance wins. I finally got around to reviewing this now that my jet lag is better.

Left a few additional ideas (optional, since this already seems like a nice improvement).

mypy/typeops.py

JukkaL · 2023-05-03T13:44:13Z

mypy/typeops.py

+                # If deleted subtypes had more general truthiness, use that
+                orig_item = new_items[duplicate_index]
+                if not orig_item.can_be_true and ti.can_be_true:
+                    new_items[duplicate_index] = true_or_false(orig_item)


Should we only change can_be_true, or is it ok to only adjust can_be_false?

This is a good question. I'm going to preserve the existing behaviour for now and explore this in another PR. These code paths don't have many unit tests (see #15094 and #15098), so I want to be careful here

JukkaL · 2023-05-03T13:44:21Z

mypy/typeops.py

+                if not orig_item.can_be_true and ti.can_be_true:
+                    new_items[duplicate_index] = true_or_false(orig_item)
+                elif not orig_item.can_be_false and ti.can_be_false:
+                    new_items[duplicate_index] = true_or_false(orig_item)


Similar to above.

hauntsaninja · 2023-05-05T20:38:02Z

This fixes a crash that just got reported #15192, so I added a regression test for it.

I added the two micro optimisations suggested.

github-actions · 2023-05-05T20:54:12Z

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

pandas-stubs (https://github.com/pandas-dev/pandas-stubs) got 1.15x faster (66.6s -> 58.0s)

speedrun.com_global_scoreboard_webapp (https://github.com/Avasam/speedrun.com_global_scoreboard_webapp) got 1.27x faster (33.3s -> 26.2s)

hauntsaninja · 2023-05-06T20:23:36Z

Nice, looks like the primer timings are pretty consistent between:
#15128 (comment)
#15128 (comment)
and some of the earlier commits on this PR (although those earlier commits had the regression on Materialize)

ilevkivskyi · 2023-05-10T14:57:13Z

This PR actually introduced a recursive tuple crash

[case testAliasRecursiveUnpack]
from typing import Tuple, TypeVar, Optional

T = TypeVar("T")
S = TypeVar("S")

A = Tuple[T, S, Optional[A[T, S]]]  # Must have two non-recursive items to crash
x: A[int, str]

*_, last = x
if last is not None:
    reveal_type(last)
[builtins fixtures/tuple.pyi]

I will check maybe there is a simple fix for it.

ilevkivskyi · 2023-05-10T15:10:00Z

It looks like there is (a bug is quite obvious). I will submit a PR shortly.

ilevkivskyi · 2023-05-10T15:44:45Z

See #15216

hauntsaninja and others added 2 commits April 25, 2023 10:18

[pre-commit.ci] auto fixes from pre-commit.com hooks

cd5f40b

for more information, see https://pre-commit.ci

This comment has been minimized.

Sign in to view

Merge branch 'master' into msu

13c4007

This comment has been minimized.

Sign in to view

hauntsaninja and others added 2 commits April 25, 2023 15:38

fix performance on materialize

b009483

[pre-commit.ci] auto fixes from pre-commit.com hooks

bea67d2

for more information, see https://pre-commit.ci

This comment has been minimized.

Sign in to view

more perf

fd8463b

hauntsaninja marked this pull request as ready for review April 26, 2023 00:11

hauntsaninja mentioned this pull request Apr 26, 2023

mypy 0.990 is 1000x slow on pydantic codebase vs 0.982 #14034

Closed

JukkaL reviewed May 3, 2023

View reviewed changes

hauntsaninja mentioned this pull request May 5, 2023

Segfault when manipulating a recursive union #15192

Closed

hauntsaninja added 2 commits May 5, 2023 13:30

add regression test for crash

b132663

micro optimisations

bc79133

hauntsaninja changed the title ~~Speed up make_simplified_union, remove a potential crash~~ Speed up make_simplified_union, fix a crash May 5, 2023

hauntsaninja changed the title ~~Speed up make_simplified_union, fix a crash~~ Speed up make_simplified_union, fix recursive tuple crash May 6, 2023

hauntsaninja merged commit 7832e1f into python:master May 6, 2023

hauntsaninja deleted the msu branch May 6, 2023 20:22

hauntsaninja mentioned this pull request Jun 14, 2023

Improve performance reports hauntsaninja/mypy_primer#81

Merged

Uh oh!

Comments

Conversation

hauntsaninja commented Apr 25, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

hauntsaninja commented Apr 25, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

github-actions bot commented Apr 25, 2023

Uh oh!

hauntsaninja commented Apr 26, 2023

Uh oh!

JukkaL commented May 3, 2023

Uh oh!

JukkaL left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

JukkaL May 3, 2023

Choose a reason for hiding this comment

Uh oh!

hauntsaninja May 5, 2023

Choose a reason for hiding this comment

Uh oh!

JukkaL May 3, 2023

Choose a reason for hiding this comment

Uh oh!

hauntsaninja commented May 5, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented May 5, 2023

Uh oh!

hauntsaninja commented May 6, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ilevkivskyi commented May 10, 2023

Uh oh!

ilevkivskyi commented May 10, 2023

Uh oh!

ilevkivskyi commented May 10, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

hauntsaninja commented Apr 25, 2023 •

edited

Loading

hauntsaninja commented Apr 25, 2023 •

edited

Loading

hauntsaninja commented May 5, 2023 •

edited

Loading

hauntsaninja commented May 6, 2023 •

edited

Loading