Could you add a test case for that? I think testing that the initial reproducer doesn't crash would be enough

Could you add a test case for that? I think testing that the initial reproducer doesn't crash would be enough

Yep, but maybe I'll try to make the reproducer a bit smaller first, it's quite big 😅 :

>>> s = open("reproducer.txt").read()
>>> len(s)
782813

Could you add a test case for that? I think testing that the initial reproducer doesn't crash would be enough

Yep, but maybe I'll try to make the reproducer a bit smaller first, it's quite big 😅 :

>>> s = open("reproducer.txt").read()
>>> len(s)
782813

Here's how it can be reduced to avoid using the file:

import ast
repro = "{0x0" + "e" * 250_000 + "%" + "e" * 250_000 + "1j}"
print(ast.literal_eval(repro))

Test added :)

Ah, in fact, here's a repro without even ast.parse():

~/dev/cpython (main)⚡ % ./python.exe -c 'import ast; repr(ast.Constant(value=eval("0x0" + "e" * 4_000)))'
./Include/refcount.h:474: _Py_NegativeRefcount: Assertion failed: object has negative ref count
<object at 0x15a858220 is freed>
Fatal Python error: _PyObject_AssertFailed: _PyObject_AssertFailed
Python runtime state: initialized
ValueError: Exceeds the limit (4300 digits) for integer string conversion; use sys.set_int_max_str_digits() to increase the limit

Current thread 0x00000001f39d4f40 (most recent call first):
  File "<string>", line 1 in <module>
zsh: abort      ./python.exe -c 

I simplified the test to use Alex's smaller reproducer and removed the note about sys.set_int_max_str_digits since the source is now a valid syntax and does not raise the other ValueError anymore.

If @JelleZijlstra has no further comments, I'll merge this tomorrow.