gh-130317: fix PyFloat_Pack/Unpack[24] for NaN's with payload #130452

skirpichev · 2025-02-22T10:08:51Z

Previously (1) all NaN's payload in PyFloat_Pack/Unpack2() was ignored and (2) type conversions (float <-> double) in PyFloat_Pack/Unpack4() broke pack-unpack round-trip for sNaN's.

Issue: struct (un)packing of half-precision nan floats is non-invertible #130317

skirpichev · 2025-02-25T09:16:39Z

Ok, I've added new tests in test_capi.test_floats (in principle, test_struct's tests are redundant).

win32 behavior (you can't unset "quiet" bit for NaN) looks as a bug for me.

Objects/floatobject.c

vstinner · 2025-02-25T12:54:59Z

win32 behavior (you can't unset "quiet" bit for NaN) looks as a bug for me.

Any idea why there is a bug only on Windows?

skirpichev · 2025-02-25T13:34:32Z

Any idea why there is a bug only on Windows?

Only on win32. Though, I suspect the situation could be worse. Maybe I should revert win32-workaround and test this with some buildbots?

C17 says: "This specification does not define the behavior of signaling NaNs." But I don't think this means you can't flip appropriate bit in float/double to make a signaling NaN.

skirpichev · 2025-04-27T13:22:41Z

Ok, I think it's ready for review.

Test for Windows failed in 32-bit mode (signaling NaN type not preserved in roundtrip), that seems to be a known issue: https://developercommunity.visualstudio.com/t/155064 (sNaN returned from function becomes qNaN). See also https://developercommunity.visualstudio.com/t/903305 and https://en.delphipraxis.net/topic/12198-delphi-win32-quiets-signaling-nan-on-function-return/.

In principle, it's possible to workaround this for the struct module. But not for C-API (PyFloat_Unpack*).

Modules/_testcapi/float.c

Lib/test/test_capi/test_float.py

Objects/floatobject.c

Co-authored-by: Victor Stinner <vstinner@python.org>

vstinner

LGTM. I just have some minor comments.

Lib/test/test_capi/test_float.py

vstinner · 2025-04-28T11:59:13Z

Lib/test/test_capi/test_float.py

+    def test_pack_unpack_roundtrip_for_nans(self):
+        pack = _testcapi.float_pack
+        unpack = _testcapi.float_unpack
+        for _ in range(1000):


100 tests should be enough to validate the implementation, no?

Suggested change

for _ in range(1000):

for _ in range(100):

1000 tests might be a little bit too slow, I don't think that it's worth it.

Up to you, the test looks instantaneous on my system. 0.3sec vs 0.03. Where the threshold?

I don't think speed is really an issue here, but having the potential for 6000 failed test reports seems like major overkill. I think 10 would actually be plenty for this loop.

Lib/test/test_capi/test_float.py

Co-authored-by: Victor Stinner <vstinner@python.org>

vstinner · 2025-04-28T13:24:38Z

Merged, thank you.

vstinner · 2025-04-28T13:25:49Z

Should we backport this change? It's unclear to me if it's a new feature or a bugfix. The issue is reported as a bug report.

skirpichev · 2025-04-28T14:06:01Z

The issue is reported as a bug report.

Strictly speaking, it's a bug, though maybe a very minor one (that's why there was no label for 3.13 backport).

From IEEE 754 (2008), 6.2.3: "Conversion of a quiet NaN from a narrower format to a wider format in the same radix, and then back to the same narrower format, should not change the quiet NaN payload in any way except to make it canonical."

E.g. in C nan's payload is preserved in conversions (e.g. float->double) or partially preserved (e.g. double -> float).

vstinner · 2025-04-29T10:22:53Z

test_capi.test_float fails on x86 Debian Installed with X 3.x: https://buildbot.python.org/#/builders/1244/builds/5176

test.pythoninfo:

CC.version: gcc (Debian 12.2.0-14) 12.2.0

builtins.float.double_format: IEEE, little-endian
builtins.float.float_format: IEEE, little-endian

platform.architecture: 32bit ELF
platform.freedesktop_os_release[NAME]: Debian GNU/Linux
platform.freedesktop_os_release[VERSION]: 12 (bookworm)
platform.libc_ver: glibc 2.36
platform.platform: Linux-6.1.0-33-686-pae-i686-with-glibc2.36

sys.float_info: sys.float_info(max=1.7976931348623157e+308, max_exp=1024, max_10_exp=308, min=2.2250738585072014e-308, min_exp=-1021, min_10_exp=-307, dig=15, mant_dig=53, epsilon=2.220446049250313e-16, radix=2, rounds=1)
sys.float_repr_style: short

sys.maxsize: 2147483647

There are failures on 16, 32 and 64 bit floats (size: 2, 4, 8). Examples:

FAIL: test_pack_unpack_roundtrip_for_nans (test.test_capi.test_float.CAPIFloatTest.test_pack_unpack_roundtrip_for_nans)
(data=b'\x7f\x98Pr', size=4, endian=0)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "test/test_capi/test_float.py", line 213, in test_pack_unpack_roundtrip_for_nans
    self.assertEqual(data1, data2)
    ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^
AssertionError: b'\x7f\x98Pr' != b'\x7f\xd8Pr'

======================================================================
FAIL: test_pack_unpack_roundtrip_for_nans (test.test_capi.test_float.CAPIFloatTest.test_pack_unpack_roundtrip_for_nans) 
(data=b'|\xd6', size=2, endian=0)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/buildbot/buildarea/3.x.ware-debian-x86.installed/build/target/lib/python3.14/test/test_capi/test_float.py", line 213, in test_pack_unpack_roundtrip_for_nans
    self.assertEqual(data1, data2)
    ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^
AssertionError: b'|\xd6' != b'~\xd6'

======================================================================
FAIL: test_pack_unpack_roundtrip_for_nans (test.test_capi.test_float.CAPIFloatTest.test_pack_unpack_roundtrip_for_nans) 
(data=b'\x7f\xf2h\xf8\x1bU*\xc1', size=8, endian=0)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/buildbot/buildarea/3.x.ware-debian-x86.installed/build/target/lib/python3.14/test/test_capi/test_float.py", line 213, in test_pack_unpack_roundtrip_for_nans
    self.assertEqual(data1, data2)
    ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^
AssertionError: b'\x7f\xf2h\xf8\x1bU*\xc1' != b'\x7f\xfah\xf8\x1bU*\xc1'

vstinner · 2025-04-29T10:58:04Z

I can reproduce the issue on Fedora 42 by building Python using gcc -m32:

./configure CFLAGS="-m32" LDFLAGS="-m32"
sed -i -e 's!#define PY_HAVE_PERF_TRAMPOLINE 1!#undef PY_HAVE_PERF_TRAMPOLINE!g' pyconfig.h
make
./python -m test -v test_capi.test_float

skirpichev · 2025-04-29T11:49:34Z

Hmm, again in all cases sNaN seems transformed to qNaN.

I think underlying reason is same and we should apply workaround for 32-bit modes.

vstinner · 2025-04-29T12:12:23Z

I think underlying reason is same and we should apply workaround for 32-bit modes.

I replaced if signaling and sys.platform == 'win32': with if signaling and True: in test_capi.test_float, but it didn't work around the issue :-( Maybe _testcapi.float_set_snan() doesn't work on x86.

vstinner · 2025-04-29T12:32:43Z

Aha, just copying a double to another double clears the sNaN flag:

(gdb) n
33	    double d = ((PyFloatObject *)obj)->ob_fval;
(gdb) n
35	    switch (size)
(gdb) p ((PyFloatObject *)obj)->ob_fval
$3 = nan(0x5e5454683683)
(gdb) p d
$4 = nan(0x85e5454683683)

d is not equal to ob_fval.

vstinner · 2025-04-29T13:12:54Z

The sNaN flag is easily lost. I added some debug of the uint64_t value in different functions. Between float_pack() and PyFloat_Pack4(), the sNaN flag is lost:

float_set_snan(): v=fffbce9e80000000 (before)
float_set_snan(): v=fff3ce9e80000000 (sNaN)
float_pack():     v=fff3ce9e80000000
PyFloat_Pack4():  v=fffbce9e80000000

PyFloat_Pack4() is called with the wrong value :-(

skirpichev · 2025-04-29T13:14:57Z

Maybe _testcapi.float_set_snan() doesn't work on x86.

Hmm, indeed. I should finally do memcpy() to created PyFloat's ob_fval, then return it.

Does it work:

diff --git a/Modules/_testcapi/float.c b/Modules/_testcapi/float.c
index 2feeb205d8..b382883fa8 100644
--- a/Modules/_testcapi/float.c
+++ b/Modules/_testcapi/float.c
@@ -181,9 +181,12 @@ _testcapi_float_set_snan(PyObject *module, PyObject *obj)
     }
     uint64_t v;
     memcpy(&v, &d, 8);
-    v &= ~(1ULL << 51); /* make sNaN */
-    memcpy(&d, &v, 8);
-    return PyFloat_FromDouble(d);
+    if (v & (1ULL << 51)) {
+        v -= (1ULL << 51); /* make sNaN */
+    }
+    PyObject *ret = PyFloat_FromDouble(0.0);
+    memcpy(&(((PyFloatObject *)ret)->ob_fval), &v, 8);
+    return ret;
 }
 
 static PyMethodDef test_methods[] = {

?

vstinner · 2025-04-29T13:23:06Z

I wrote #133150 to fix 3 bugs, but it's not enough to fix the test on x86.

bedevere-app bot added the awaiting review label Feb 22, 2025

bedevere-app bot mentioned this pull request Feb 22, 2025

struct (un)packing of half-precision nan floats is non-invertible #130317

Closed

skirpichev marked this pull request as draft February 22, 2025 10:40

bedevere-app bot removed the awaiting review label Feb 22, 2025

skirpichev force-pushed the fix-nan-packing/130317 branch from 7da9a1f to a6e99f9 Compare February 22, 2025 11:19

pythongh-130317: fix PyFloat_Pack2/Unpack2 for NaN's with payload

971fd98

skirpichev force-pushed the fix-nan-packing/130317 branch from a6e99f9 to 971fd98 Compare February 23, 2025 03:15

skirpichev marked this pull request as ready for review February 23, 2025 03:53

bedevere-app bot added the awaiting review label Feb 23, 2025

This comment was marked as outdated.

Sign in to view

skirpichev added 2 commits February 25, 2025 03:20

Merge branch 'master' into fix-nan-packing/130317

b05720a

more tests in test_capi, fix also Pack/Unpack4

f513b14

skirpichev marked this pull request as draft February 25, 2025 06:04

bedevere-app bot removed the awaiting review label Feb 25, 2025

skirpichev added 3 commits February 25, 2025 09:26

+ different fix for Pack/Unpack2 (works as for floats)

218f6b3

blacklist on win32

a7f5ed9

cleanup and comments

c7c08ff

skirpichev changed the title ~~gh-130317: fix PyFloat_Pack2/Unpack2 for NaN's with payload~~ gh-130317: fix PyFloat_Pack/Unpack[24] for NaN's with payload Feb 25, 2025

skirpichev added 2 commits February 25, 2025 10:29

+ enable test_pack_unpack_roundtrip_nans on win32 for qNaN's

de7b6aa

+1

e6c1c12

skirpichev force-pushed the fix-nan-packing/130317 branch from 139501d to e6c1c12 Compare February 25, 2025 08:24

skirpichev marked this pull request as ready for review February 25, 2025 09:12

bedevere-app bot added the awaiting review label Feb 25, 2025

skirpichev requested a review from vstinner February 25, 2025 09:16

vstinner reviewed Feb 25, 2025

View reviewed changes

Objects/floatobject.c Outdated Show resolved Hide resolved

address review: warning on Windows

ac7bdcb

skirpichev marked this pull request as ready for review April 27, 2025 13:10

bedevere-app bot added the awaiting review label Apr 27, 2025

vstinner reviewed Apr 27, 2025

View reviewed changes

Modules/_testcapi/float.c Outdated Show resolved Hide resolved

Modules/_testcapi/float.c Outdated Show resolved Hide resolved

address review: redo helper

00dfd66

skirpichev requested a review from vstinner April 27, 2025 15:18

vstinner reviewed Apr 28, 2025

View reviewed changes

skirpichev and others added 3 commits April 28, 2025 13:42

Apply suggestions from code review

b7f727c

Co-authored-by: Victor Stinner <vstinner@python.org>

restore comment

13e8310

address review: ensure code works with strict aliasing

bcf54f6

skirpichev requested a review from vstinner April 28, 2025 11:57

vstinner approved these changes Apr 28, 2025

View reviewed changes

bedevere-app bot added awaiting merge and removed awaiting review labels Apr 28, 2025

skirpichev commented Apr 28, 2025

View reviewed changes

Lib/test/test_capi/test_float.py Show resolved Hide resolved

Apply suggestions from code review

9d20633

Co-authored-by: Victor Stinner <vstinner@python.org>

vstinner merged commit 6157135 into python:main Apr 28, 2025
42 checks passed

bedevere-app bot removed the awaiting merge label Apr 28, 2025

skirpichev deleted the fix-nan-packing/130317 branch April 28, 2025 13:29

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

gh-130317: fix PyFloat_Pack/Unpack[24] for NaN's with payload #130452

gh-130317: fix PyFloat_Pack/Unpack[24] for NaN's with payload #130452

skirpichev commented Feb 22, 2025 •

edited

Loading

This comment was marked as outdated.

This comment was marked as outdated.

skirpichev commented Feb 25, 2025

vstinner commented Feb 25, 2025

skirpichev commented Feb 25, 2025

skirpichev commented Apr 27, 2025

vstinner left a comment

vstinner Apr 28, 2025

skirpichev Apr 28, 2025

zware Apr 29, 2025

vstinner commented Apr 28, 2025

vstinner commented Apr 28, 2025

skirpichev commented Apr 28, 2025

vstinner commented Apr 29, 2025

vstinner commented Apr 29, 2025

skirpichev commented Apr 29, 2025

vstinner commented Apr 29, 2025

vstinner commented Apr 29, 2025

vstinner commented Apr 29, 2025

skirpichev commented Apr 29, 2025

vstinner commented Apr 29, 2025

gh-130317: fix PyFloat_Pack/Unpack[24] for NaN's with payload #130452

gh-130317: fix PyFloat_Pack/Unpack[24] for NaN's with payload #130452

Conversation

skirpichev commented Feb 22, 2025 • edited Loading

This comment was marked as outdated.

This comment was marked as outdated.

skirpichev commented Feb 25, 2025

vstinner commented Feb 25, 2025

skirpichev commented Feb 25, 2025

skirpichev commented Apr 27, 2025

vstinner left a comment

Choose a reason for hiding this comment

vstinner Apr 28, 2025

Choose a reason for hiding this comment

skirpichev Apr 28, 2025

Choose a reason for hiding this comment

zware Apr 29, 2025

Choose a reason for hiding this comment

vstinner commented Apr 28, 2025

vstinner commented Apr 28, 2025

skirpichev commented Apr 28, 2025

vstinner commented Apr 29, 2025

vstinner commented Apr 29, 2025

skirpichev commented Apr 29, 2025

vstinner commented Apr 29, 2025

vstinner commented Apr 29, 2025

vstinner commented Apr 29, 2025

skirpichev commented Apr 29, 2025

vstinner commented Apr 29, 2025

skirpichev commented Feb 22, 2025 •

edited

Loading