PY_SSIZE_T_CLEAN error when loading Triton-generated extension on Windows (Python 3.12.9, VS2022, PyTorch 2.8+cu129, triton-windows 3.4)

[happye]

Environment
OS: Windows 11, x64.
GPU: 4080s, Driver Version: 581.29, CUDA Version: 13.0
ComfyUI embedded Python: 3.12.9 at G:\Tools\ComfyUI_wan2.1\python.
PyTorch: 2.8+cu129.
triton-windows: 3.4.
CUDA: Toolkit 12.4.
MSVC: Visual Studio 2022 (x64 Native Tools Command Prompt used).
vcredist already installed.

Reproduction steps
Start VS2022 x64 Native Tools Command Prompt.

Ensure embedded Python is used: G:\Tools\ComfyUI_wan2.1>.\python\python.exe -m pip show torch triton-windows.

Run the README test script test_triton.py (exact code from triton-windows README) using the embedded python: .\python\python.exe test_triton.py

Observed result (critical log)
Compilation phases show creation of cuda_utils and __triton_launcher libraries, then failure on load with traceback ending in: SystemError: PY_SSIZE_T_CLEAN macro must be defined for '#' formats

sysconfig check printed: Python 3.12.9 and returned Py_ENABLE_SHARED=None, LIBDIR=None for the embedded Python.

What I tried

Deleted caches: %USERPROFILE%.triton\cache and %LOCALAPPDATA%\Temp\torchinductor_, and removed temp files with __triton_launcher / cuda_utils prefixes.

Uninstalled VS2019 and installed VS2022; ran tests from VS2022 dev prompt to ensure cl/link come from VS2022.

Cleared leftover env vars (VCINSTALLDIR, VCToolsVersion, WindowsSdkDir, WindowsSDKVersion) in-session and invoked vcvars64.bat..

Verified include/libs placed into embedded Python per README guidance (include and libs for Python 3.12 series).

Reinstalled triton-windows in the same session and repeated test.

Confirmed compilation run occurs but load fails with the PY_SSIZE_T_CLEAN error despite full cleanup and VS2022 environment.

Facts from the project README that I followed or considered: the project requires copying include and libs into the Python folder for embedded Python; the README documents the need to delete triton/torchinductor cache when changing compilers, Python, or Triton; and the README explicitly lists the PY_SSIZE_T_CLEAN error as a symptom recommending cache deletion and further diagnostics

[woct0rdho]

Can you show the error log?

[happye]

Can you show the error log?

G:\Tools\ComfyUI_wan2.1>.\python\python.exe test_triton.py
Traceback (most recent call last):
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 25, in <module>
    b_compiled = add(a, a)
                 ^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 20, in add
    add_kernel[grid](x, y, output, n_elements, BLOCK_SIZE=1024)
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 390, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 617, in run
    kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
    ^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 498, in __getattribute__
    self._init_handles()
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 490, in _init_handles
    self.module, self.function, self.n_regs, self.n_spills, self.n_max_threads = driver.active.utils.load_binary(
                                                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
SystemError: PY_SSIZE_T_CLEAN macro must be defined for '#' formats

[woct0rdho]

You've done a pretty detailed troubleshooting and I think it's unlikely that the issue still happens. Anyway, let's check if it's because of some uncommon configuration of your cache.

Please modify the test script test_triton.py, and add the following lines at the beginning:

import os
os.environ["TRITON_CACHE_DIR"] = "tmp_triton_cache"

Then run the test script again. If it works, it will create a new folder tmp_triton_cache in the current folder, and it will not be affected by anything in the default cache folder.

[happye]

You've done a pretty detailed troubleshooting and I think it's unlikely that the issue still happens. Anyway, let's check if it's because of some uncommon configuration of your cache.

Please modify the test script , and add the following lines at the beginning:test_triton.py

import os
os.environ["TRITON_CACHE_DIR"] = "tmp_triton_cache"
Then run the test script again. If it works, it will create a new folder in the current folder, and it will not be affected by anything in the default cache folder.tmp_triton_cache

Sadly, it still happens. With the same Traceback except the cache folder was created in the current folder...

[woct0rdho]

Please send the folder tmp_triton_cache here. I'll try to analyze it.

Then let's check some environments. Please show the full output of:

G:\Tools\ComfyUI_wan2.1>.\python\python.exe -m pip list

And please run the following in Python and show the full output:

import os
print(os.environ)
import sysconfig
print(sysconfig.get_config_vars())

Also, where did you download 'ComfyUI_wan2.1' from? In the official ComfyUI portable package, Python should be in the folder python_embeded.

[happye]

Please send the folder tmp_triton_cache here. I'll try to analyze it.

Then let's check some environments. Please show the full output of:

G:\Tools\ComfyUI_wan2.1>.\python\python.exe -m pip list

And please run the following in Python and show the full output:

import os
print(os.environ)
import sysconfig
print(sysconfig.get_config_vars())
Also, where did you download 'ComfyUI_wan2.1' from? In the official ComfyUI portable package, Python should be in the folder python_embeded.

tmp_triton_cache.zip

requirements.txt

python_commands_full_output.txt

I appreciate that you can keep tracking this issue! I downloaded ComfyUI_wan2.1 from a third-party community. But sorry, I don't know where it is from...

[woct0rdho]

I can reproduce that your cuda_utils.cp312-win_amd64.pyd cannot be loaded, and I need to know how it was compiled. Please modify the file G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\build.py:

triton-windows/python/triton/runtime/build.py

Line 116 in 4d53687

cc_cmd = _cc_cmd(cc, src, so, include_dirs, library_dirs, libraries)

Add a line print('cc_cmd', cc_cmd) after line 116, then clear the caches, run the test script again, and show the error log.

---

Also, your environment variable PATH shows that you have many different Python installations on your computer, and that will make things complicated. What if you clear the caches, then temporarily clear PATH, then run the test script again?

To temporarily clear PATH in the current command window, run the following in cmd:

set PATH=

or the following in PowerShell:

$Env:PATH = ""

(For now don't permanently modify PATH in that System Properties window.)

[happye]

New log (with default PATH):

G:\Tools\ComfyUI_wan2.1>.\python\python.exe test_triton.py
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp87rbf7i8\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp87rbf7i8\\cuda_utils.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmp87rbf7i8', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp0j9_hwl0\\__triton_launcher.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp0j9_hwl0\\__triton_launcher.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmp0j9_hwl0', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
Traceback (most recent call last):
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 27, in <module>
    b_compiled = add(a, a)
                 ^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 22, in add
    add_kernel[grid](x, y, output, n_elements, BLOCK_SIZE=1024)
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 390, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 617, in run
    kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
    ^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 498, in __getattribute__
    self._init_handles()
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 490, in _init_handles
    self.module, self.function, self.n_regs, self.n_spills, self.n_max_threads = driver.active.utils.load_binary(
                                                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
SystemError: PY_SSIZE_T_CLEAN macro must be defined for '#' formats

New log (with set PATH=):

G:\Tools\ComfyUI_wan2.1>.\python\python.exe test_triton.py
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmpd92cpz61\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmpd92cpz61\\cuda_utils.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmpd92cpz61', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmps_9inn66\\__triton_launcher.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmps_9inn66\\__triton_launcher.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmps_9inn66', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
Traceback (most recent call last):
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 27, in <module>
    b_compiled = add(a, a)
                 ^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 22, in add
    add_kernel[grid](x, y, output, n_elements, BLOCK_SIZE=1024)
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 390, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 617, in run
    kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
    ^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 498, in __getattribute__
    self._init_handles()
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 490, in _init_handles
    self.module, self.function, self.n_regs, self.n_spills, self.n_max_threads = driver.active.utils.load_binary(
                                                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
SystemError: PY_SSIZE_T_CLEAN macro must be defined for '#' formats

seems there's no differences between them...

[happye]

New log (with default PATH):

G:\Tools\ComfyUI_wan2.1>.\python\python.exe test_triton.py
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp87rbf7i8\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp87rbf7i8\\cuda_utils.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmp87rbf7i8', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp0j9_hwl0\\__triton_launcher.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmp0j9_hwl0\\__triton_launcher.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmp0j9_hwl0', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
Traceback (most recent call last):
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 27, in <module>
    b_compiled = add(a, a)
                 ^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 22, in add
    add_kernel[grid](x, y, output, n_elements, BLOCK_SIZE=1024)
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 390, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 617, in run
    kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
    ^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 498, in __getattribute__
    self._init_handles()
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 490, in _init_handles
    self.module, self.function, self.n_regs, self.n_spills, self.n_max_threads = driver.active.utils.load_binary(
                                                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
SystemError: PY_SSIZE_T_CLEAN macro must be defined for '#' formats

New log (with set PATH=):

G:\Tools\ComfyUI_wan2.1>.\python\python.exe test_triton.py
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmpd92cpz61\\cuda_utils.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmpd92cpz61\\cuda_utils.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmpd92cpz61', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
cc_cmd ['G:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\runtime\\tcc\\tcc.exe', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmps_9inn66\\__triton_launcher.c', '-O3', '-shared', '-Wno-psabi', '-o', 'C:\\Users\\Crux\\AppData\\Local\\Temp\\tmps_9inn66\\__triton_launcher.cp312-win_amd64.pyd', '-fPIC', '-D_Py_USE_GCC_BUILTIN_ATOMICS', '-lcuda', '-lpython312', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\lib', '-LC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\lib\\x64', '-LG:\\Tools\\ComfyUI_wan2.1\\python\\libs', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Lib\\site-packages\\triton\\backends\\nvidia\\include', '-IC:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\include', '-IC:\\Users\\Crux\\AppData\\Local\\Temp\\tmps_9inn66', '-IG:\\Tools\\ComfyUI_wan2.1\\python\\Include']
Traceback (most recent call last):
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 27, in <module>
    b_compiled = add(a, a)
                 ^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\test_triton.py", line 22, in add
    add_kernel[grid](x, y, output, n_elements, BLOCK_SIZE=1024)
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 390, in <lambda>
    return lambda *args, **kwargs: self.run(grid=grid, warmup=False, *args, **kwargs)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\runtime\jit.py", line 617, in run
    kernel.run(grid_0, grid_1, grid_2, stream, kernel.function, kernel.packed_metadata, launch_metadata,
    ^^^^^^^^^^
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 498, in __getattribute__
    self._init_handles()
  File "G:\Tools\ComfyUI_wan2.1\python\Lib\site-packages\triton\compiler\compiler.py", line 490, in _init_handles
    self.module, self.function, self.n_regs, self.n_spills, self.n_max_threads = driver.active.utils.load_binary(
                                                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
SystemError: PY_SSIZE_T_CLEAN macro must be defined for '#' formats

seems there's no differences between them...

[woct0rdho]

Thanks, I'll try to analyze what it means.

In the meanwhile, what if you install and configure MSVC, see Step 5 in my README? Without MSVC, triton-windows uses the bundled TinyCC, which may be problematic in some environments.

[hyyy1211]

或许回退版本可以解决问题？我从3.5.0回退到了3.4.0，测试样例就正常了。
Perhaps rolling back a version would solve the problem? I rolled back from 3.5.0 to 3.4.0, and the test cases worked fine.