[ty] Increase worker-thread stack size by sharkdp · Pull Request #17869 · astral-sh/ruff

sharkdp · 2025-05-05T19:16:20Z

Summary

This is obviously just a band-aid solution to this problem (in that you can always make your pathological inputs bigger and it will still crash), but I think this is not an unreasonable change — even if we add more sophisticated solutions later. I tried using stacker as suggested by @MichaReiser, and it works. But it's unclear where exactly would be the right place to put it, and even for the sympy problem, we would need to add it both in the semantic index builder AST traversal and in type inference. Increasing the default stack size for worker threads, as proposed here, doesn't solve the underlying problem (that there is a hard limit), but it is more universal in the sense that it is not specific to large binary-operator expression chains.

To determine a reasonable stack size, I created files that look like

right associative:

from typing import reveal_type
total = (1 + (1 + (1 + (1 + (… + 1)))))
reveal_type(total)

left associative

from typing import reveal_type
total = 1 + 1 + 1 + 1 + … + 1
reveal_type(total)

with a variable amount of operands (N). I then chose the stack size large enough to still be able to handle cases that existing type checkers can not:

right

  N = 20: mypy takes ~ 1min
  N = 350: pyright crashes with a stack overflow (mypy fails with "too many nested parentheses")
  N = 800: ty(main) infers Literal[800] instantly
  N = 1000: ty(main) crashes with "thread '<unknown>' has overflowed its stack"

  N = 7000: ty(this branch) infers Literal[7000] instantly
  N = 8000+: ty(this branch) crashes


left

  N = 300: pyright emits "Maximum parse depth exceeded; break expression into smaller sub-expressions"
           total is inferred as Unknown
  N = 5500: mypy crashes with "INTERNAL ERROR"
  N = 2500: ty(main) infers Literal[2500] instantly
  N = 3000: ty(main) crashes with "thread '<unknown>' has overflowed its stack"

  N = 22000: ty(this branch) infers Literal[22000] instantly
  N = 23000+: ty(this branch) crashes

Test Plan

New regression test.

github-actions · 2025-05-05T19:19:11Z

`mypy_primer` results

No ecosystem changes detected ✅

sharkdp · 2025-05-05T19:23:34Z

crates/ty/src/main.rs

+        // binary operators (x + x + … + x) both during the AST walk in semantic index building as
+        // well as during type checking. Using this stack size, we can handle handle expressions
+        // that are several times larger than the corresponding limits in existing type checkers.
+        .stack_size(16 * 1024 * 1024)


It's probably worth noting that this does not generally increase memory usage by 16 MiB × num_threads. It's just the max. available stack space that a single thread can use.

And 16MiB/thread is not even detectable, relative to the memory we need for source texts and AST, for any project large enough to care about memory usage.

carljm

This looks great, thank you

carljm · 2025-05-05T19:25:00Z

crates/ty/src/main.rs

+        // binary operators (x + x + … + x) both during the AST walk in semantic index building as
+        // well as during type checking. Using this stack size, we can handle handle expressions
+        // that are several times larger than the corresponding limits in existing type checkers.
+        .stack_size(16 * 1024 * 1024)


And 16MiB/thread is not even detectable, relative to the memory we need for source texts and AST, for any project large enough to care about memory usage.

sharkdp · 2025-05-05T19:28:00Z

crates/ty/tests/cli.rs

+        total = 1{plus_one_repeated}
+        reveal_type(total)
+        ",
+        plus_one_repeated = " + 1".repeat(2000 - 1)


I chose a much smaller limit than what we can handle (but still bigger than what we can handle on main) for two reasons:

Debug builds take up more stack space and crash at smaller input sizes

We don't want minor increases of stack frame sizes to lead to a test failure here.

## Summary Following #17869, we can now run `ty` on `sympy`.

MichaReiser · 2025-05-05T21:15:16Z

I think the long term solution here would be to implement binary expression traversal in a loop rather than with recursion. But this is fine for now

@MichaReiser

## Summary closes astral-sh#17472 This is obviously just a band-aid solution to this problem (in that you can always make your [pathological inputs](https://github.com/sympy/sympy/blob/28994edd82a18c22a98a52245c173f3c836dcad3/sympy/polys/numberfields/resolvent_lookup.py) bigger and it will still crash), but I think this is not an unreasonable change — even if we add more sophisticated solutions later. I tried using `stacker` as suggested by @MichaReiser, and it works. But it's unclear where exactly would be the right place to put it, and even for the `sympy` problem, we would need to add it both in the semantic index builder AST traversal and in type inference. Increasing the default stack size for worker threads, as proposed here, doesn't solve the underlying problem (that there is a hard limit), but it is more universal in the sense that it is not specific to large binary-operator expression chains. To determine a reasonable stack size, I created files that look like *right associative*: ```py from typing import reveal_type total = (1 + (1 + (1 + (1 + (… + 1))))) reveal_type(total) ``` *left associative* ```py from typing import reveal_type total = 1 + 1 + 1 + 1 + … + 1 reveal_type(total) ``` with a variable amount of operands (`N`). I then chose the stack size large enough to still be able to handle cases that existing type checkers can not: ``` right N = 20: mypy takes ~ 1min N = 350: pyright crashes with a stack overflow (mypy fails with "too many nested parentheses") N = 800: ty(main) infers Literal[800] instantly N = 1000: ty(main) crashes with "thread '<unknown>' has overflowed its stack" N = 7000: ty(this branch) infers Literal[7000] instantly N = 8000+: ty(this branch) crashes left N = 300: pyright emits "Maximum parse depth exceeded; break expression into smaller sub-expressions" total is inferred as Unknown N = 5500: mypy crashes with "INTERNAL ERROR" N = 2500: ty(main) infers Literal[2500] instantly N = 3000: ty(main) crashes with "thread '<unknown>' has overflowed its stack" N = 22000: ty(this branch) infers Literal[22000] instantly N = 23000+: ty(this branch) crashes ``` ## Test Plan New regression test.

## Summary Following astral-sh#17869, we can now run `ty` on `sympy`.

[ty] Increase worker-thread stack size

d7fce30

sharkdp added the ty Multi-file analysis & type inference label May 5, 2025

sharkdp requested review from AlexWaygood, MichaReiser, carljm and dcreager as code owners May 5, 2025 19:16

AlexWaygood removed their request for review May 5, 2025 19:20

sharkdp commented May 5, 2025

View reviewed changes

carljm approved these changes May 5, 2025

View reviewed changes

sharkdp commented May 5, 2025

View reviewed changes

sharkdp merged commit de78da5 into main May 5, 2025
35 checks passed

sharkdp deleted the david/increase-default-stack-size branch May 5, 2025 19:31

sharkdp mentioned this pull request May 5, 2025

[ty] ecosystem: activate running on 'sympy' #17870

Merged

sharkdp added a commit that referenced this pull request May 5, 2025

[ty] ecosystem: activate running on 'sympy' (#17870)

9a6633d

## Summary Following #17869, we can now run `ty` on `sympy`.

Glyphack pushed a commit to Glyphack/ruff that referenced this pull request May 6, 2025

[ty] ecosystem: activate running on 'sympy' (astral-sh#17870)

dc1dd9d

## Summary Following astral-sh#17869, we can now run `ty` on `sympy`.

sharkdp mentioned this pull request May 7, 2025

Stack overflow for large chain of inter-dependent definitions astral-sh/ty#96

Open

MichaReiser mentioned this pull request Jan 7, 2026

[ty] Fix stack overflow due to too small stack size #22433

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[ty] Increase worker-thread stack size#17869

[ty] Increase worker-thread stack size#17869
sharkdp merged 1 commit intomainfrom
david/increase-default-stack-size

sharkdp commented May 5, 2025 •

edited

Loading

Uh oh!

github-actions bot commented May 5, 2025

Uh oh!

sharkdp May 5, 2025

Uh oh!

carljm May 5, 2025

Uh oh!

carljm left a comment

Uh oh!

carljm May 5, 2025

Uh oh!

sharkdp May 5, 2025

Uh oh!

Uh oh!

MichaReiser commented May 5, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

sharkdp commented May 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test Plan

Uh oh!

github-actions bot commented May 5, 2025

mypy_primer results

Uh oh!

sharkdp May 5, 2025

Choose a reason for hiding this comment

Uh oh!

carljm May 5, 2025

Choose a reason for hiding this comment

Uh oh!

carljm left a comment

Choose a reason for hiding this comment

Uh oh!

carljm May 5, 2025

Choose a reason for hiding this comment

Uh oh!

sharkdp May 5, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

MichaReiser commented May 5, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

sharkdp commented May 5, 2025 •

edited

Loading

`mypy_primer` results