QuadPrecision(2) ** 100 crashes on Pyodide

[juntyr]

import numpy as np
import numpy_quaddtype

np.power(numpy_quaddtype.QuadPrecision(2), 112) # ok
np.array(numpy_quaddtype.QuadPrecision(2)) ** 112 # ok

numpy_quaddtype.QuadPrecision(2) ** 112 # crash

Pyodide has suffered a fatal error. Please report this to the Pyodide maintainers.
The cause of the fatal error was:
RuntimeError: null function or function signature mismatch
    at wasm://wasm/020fcee6:wasm-function[838]:0x152bd1
    at wasm://wasm/020fcee6:wasm-function[839]:0x152d5c
    at wasm://wasm/020fcee6:wasm-function[3533]:0x26dd6a
    at wasm://wasm/020fcee6:wasm-function[3531]:0x25ef9f
    at wasm://wasm/020fcee6:wasm-function[4464]:0x2d68c2
    at wasm://wasm/020fcee6:wasm-function[4465]:0x2d6ccf
    at wasm://wasm/020fcee6:wasm-function[4461]:0x2d64d8
    at wasm://wasm/020fcee6:wasm-function[4458]:0x2d5dc8
    at wasm://wasm/020fcee6:wasm-function[4456]:0x2d59d6
    at wasm://wasm/020fcee6:wasm-function[4466]:0x2d6d3a {
  pyodide_fatal_error: true
}
Stack (most recent call first):
  File "<stdin>-2", line 1 in <module>
  File "/lib/python313.zip/_pyrepl/main.py", line 30 in interactive_console
  File "/lib/python313.zip/_pyrepl/__main__.py", line 6 in <module>
  File "<frozen runpy>", line 88 in _run_code
  File "<frozen runpy>", line 198 in _run_module_as_main
RuntimeError: null function or function signature mismatch
    at wasm://wasm/020fcee6:wasm-function[838]:0x152bd1
    at wasm://wasm/020fcee6:wasm-function[839]:0x152d5c
    at wasm://wasm/020fcee6:wasm-function[3533]:0x26dd6a
    at wasm://wasm/020fcee6:wasm-function[3531]:0x25ef9f
    at wasm://wasm/020fcee6:wasm-function[4464]:0x2d68c2
    at wasm://wasm/020fcee6:wasm-function[4465]:0x2d6ccf
    at wasm://wasm/020fcee6:wasm-function[4461]:0x2d64d8
    at wasm://wasm/020fcee6:wasm-function[4458]:0x2d5dc8
    at wasm://wasm/020fcee6:wasm-function[4456]:0x2d59d6
    at wasm://wasm/020fcee6:wasm-function[4466]:0x2d6d3a {
  pyodide_fatal_error: true
}

[juntyr]

This was with numpy 2.3.5 and numpy-quaddtype 0.2.2

[SwayamInSync]

I think I caught the issue, this operator is registered as

.nb_power = (ternaryfunc)quad_binary_func<quad_pow, ld_pow>,

In Python, the ** calls to the pow(base, exp, mod) (a ternaryfunc) with mod = py_none

Where quad_pow is a binaryfunc but casted to a ternaryfunc. On native x86/ARM, this often "works" because extra stack arguments are simply ignored. But WebAssembly might having strict function signature enforcement in its call_indirect table.

I should fix this by making underline C method to also handle the mod parameter

[SwayamInSync]

@ngoldbaum just checked that numpy has a wasm build & test workflow emscripten.yml should we add one here to catch if more such cases are present?

[juntyr]

Here's my downstream CI which tests numpy-quaddtype in pyodide: https://github.com/juntyr/compression-safeguards/blob/906f7af74b8901802b253020e05894bffb6703ca/.github/workflows/ci.yml#L284-L381

with the recipes here:
https://github.com/juntyr/compression-safeguards/tree/906f7af74b8901802b253020e05894bffb6703ca/tools/ci/pyodide-recipes/packages

I think it will be important to have recipes for numpy and libsleef and numpy-quaddtype in this repo because (1) we can catch issues faster (2) we can iterate on them outside of Pyodide releases and (3) Pyodide numpy is a bit stuck at the moment (the upgrade to 2.4 will likely take a while) since it supports an old scipy and upgrading scipy always takes the Pyodide team a lot of work (whereas numpy nowadays just builds)

[seberg]

It's typical for Pyodide to notice these issues. Adding CI for it seems reasonable (but it should also be reasonable to drop it again if nobody maintains it).

FWIW, since the discussion is to create a new repo, I wouldn't add it here. Let's keep this to incubate new dtypes and have examples for weirder ones. For that such CI seems too much hassle.

[SwayamInSync]

Make sense, we can add that and move the current quadtype specific workflows to its repo only.