Only allow integer type_caster to call int method when conversion is allowed; always call index #2698

YannickJadoul · 2020-11-26T22:16:47Z

Description

As the title says and the patch shows: currently, PyLong_AsLong/PyLong_AsLongLong gets called independently of convert in the type_caster for integral types. This seems inconsistent with other type casters.

Suggested changelog entry:

The ``type_caster`` for integers does not convert Python objects with ``__int__`` anymore with ``noconvert`` or during the first round of trying overloads.

…en conversion is allowed

…o int in 3.8+

henryiii · 2020-12-16T02:57:48Z

Milestone?

YannickJadoul · 2020-12-16T18:45:31Z

I'd consider this a fix; do you agree, @henryiii? If not, feel free to change to 2.7.0

rwgk · 2020-12-18T18:41:16Z

Observation only, no analysis yet:

This PR breaks https://github.com/google/mediapipe/blob/master/mediapipe/python/packet_test.py
The test output is below.
The test runs fine without this PR, after patching in this PR it fails.

[ RUN      ] PacketTest.test_boolean_packet
[       OK ] PacketTest.test_boolean_packet
[ RUN      ] PacketTest.test_bytes_packet
[       OK ] PacketTest.test_bytes_packet
[ RUN      ] PacketTest.test_detection_proto_packet
[       OK ] PacketTest.test_detection_proto_packet
[ RUN      ] PacketTest.test_double_packet
[       OK ] PacketTest.test_double_packet
[ RUN      ] PacketTest.test_empty_packet
[       OK ] PacketTest.test_empty_packet
[ RUN      ] PacketTest.test_float_array_packet
[       OK ] PacketTest.test_float_array_packet
[ RUN      ] PacketTest.test_float_image_frame_packet
[       OK ] PacketTest.test_float_image_frame_packet
[ RUN      ] PacketTest.test_float_packet
[       OK ] PacketTest.test_float_packet
[ RUN      ] PacketTest.test_float_vector_packet
[       OK ] PacketTest.test_float_vector_packet
[ RUN      ] PacketTest.test_image_frame_packet_copy_creation_with_cropping
[       OK ] PacketTest.test_image_frame_packet_copy_creation_with_cropping
[ RUN      ] PacketTest.test_image_frame_packet_creation_copy_mode
[       OK ] PacketTest.test_image_frame_packet_creation_copy_mode
[ RUN      ] PacketTest.test_image_frame_packet_creation_reference_mode
[       OK ] PacketTest.test_image_frame_packet_creation_reference_mode
[ RUN      ] PacketTest.test_int16_packet
[  FAILED  ] PacketTest.test_int16_packet
[ RUN      ] PacketTest.test_int32_packet
[  FAILED  ] PacketTest.test_int32_packet
[ RUN      ] PacketTest.test_int64_packet
[  FAILED  ] PacketTest.test_int64_packet
[ RUN      ] PacketTest.test_int8_packet
[  FAILED  ] PacketTest.test_int8_packet
[ RUN      ] PacketTest.test_int_array_packet
[       OK ] PacketTest.test_int_array_packet
[ RUN      ] PacketTest.test_int_packet
[  FAILED  ] PacketTest.test_int_packet
[ RUN      ] PacketTest.test_int_vector_packet
[       OK ] PacketTest.test_int_vector_packet
[ RUN      ] PacketTest.test_matrix_packet
[       OK ] PacketTest.test_matrix_packet
[ RUN      ] PacketTest.test_matrix_packet_with_non_c_contiguous_data
[       OK ] PacketTest.test_matrix_packet_with_non_c_contiguous_data
[ RUN      ] PacketTest.test_packet_vector_packet
[       OK ] PacketTest.test_packet_vector_packet
[ RUN      ] PacketTest.test_string_packet
[       OK ] PacketTest.test_string_packet
[ RUN      ] PacketTest.test_string_to_packet_map_packet
[       OK ] PacketTest.test_string_to_packet_map_packet
[ RUN      ] PacketTest.test_string_vector_packet
[       OK ] PacketTest.test_string_vector_packet
[ RUN      ] PacketTest.test_uint16_image_frame_packet
[       OK ] PacketTest.test_uint16_image_frame_packet
[ RUN      ] PacketTest.test_uint16_packet
[  FAILED  ] PacketTest.test_uint16_packet
[ RUN      ] PacketTest.test_uint32_packet
[  FAILED  ] PacketTest.test_uint32_packet
[ RUN      ] PacketTest.test_uint64_packet
[       OK ] PacketTest.test_uint64_packet
[ RUN      ] PacketTest.test_uint8_image_frame_packet
[       OK ] PacketTest.test_uint8_image_frame_packet
[ RUN      ] PacketTest.test_uint8_packet
[  FAILED  ] PacketTest.test_uint8_packet
======================================================================
ERROR: test_int16_packet (__main__.PacketTest)
PacketTest.test_int16_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 73, in test_int16_packet
    p2 = mp.packet_creator.create_int16(np.int16(1))
TypeError: create_int16(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_int32_packet (__main__.PacketTest)
PacketTest.test_int32_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 86, in test_int32_packet
    p2 = mp.packet_creator.create_int32(np.int32(1))
TypeError: create_int32(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_int64_packet (__main__.PacketTest)
PacketTest.test_int64_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 96, in test_int64_packet
    p2 = mp.packet_creator.create_int64(np.int64(1))
TypeError: create_int64(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_int8_packet (__main__.PacketTest)
PacketTest.test_int8_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 61, in test_int8_packet
    p2 = mp.packet_creator.create_int8(np.int8(1))
TypeError: create_int8(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_int_packet (__main__.PacketTest)
PacketTest.test_int_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 49, in test_int_packet
    p2 = mp.packet_creator.create_int(np.intc(1))
TypeError: create_int(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_uint16_packet (__main__.PacketTest)
PacketTest.test_uint16_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 120, in test_uint16_packet
    p2 = mp.packet_creator.create_uint16(np.uint16(1))
TypeError: create_uint16(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_uint32_packet (__main__.PacketTest)
PacketTest.test_uint32_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 132, in test_uint32_packet
    p2 = mp.packet_creator.create_uint32(np.uint32(1))
TypeError: create_uint32(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

======================================================================
ERROR: test_uint8_packet (__main__.PacketTest)
PacketTest.test_uint8_packet
----------------------------------------------------------------------
Traceback (most recent call last):
  File "mediapipe/python/packet_test.py", line 108, in test_uint8_packet
    p2 = mp.packet_creator.create_uint8(np.uint8(1))
TypeError: create_uint8(): incompatible function arguments. The following argument types are supported:
    1. (arg0: int) -> mediapipe.python._framework_bindings.packet.Packet

Invoked with: 1

----------------------------------------------------------------------
Ran 31 tests in 0.523s

FAILED (errors=8)

henryiii · 2020-12-18T20:11:02Z

__index__ is somewhat special. Would it work if __int__ was not called automatically, but __index__ was?

object.__index__(self)

Called to implement operator.index(), and whenever Python needs to losslessly convert the numeric object to an integer object (such as in slicing, or in the built-in bin(), hex() and oct() functions). Presence of this method indicates that the numeric object is an integer type. Must return an integer.

Therefore, having an __index__ means that this object is an integer type, with a lossless representation as an integer. __int__ means you can convert it to an integer, possibly/probably lossy, and the object is possibly not an integer.

rwgk · 2020-12-18T20:15:36Z

test_int_convert_numpy below is a minimal example of the mediapipe issue.

Main conclusion: this PR is working as intended, the mediapipe code needs to remove py::arg().noconvert().

My feeling about milestone: given that we now know this PR breaks someone, maybe 2.7 is better?

Might be nice to include the extra test below in this PR. It's similar to test_numpy_bool but uses int_passthrough, i.e. adds a little bit of coverage.

Tangential: test_numpy_bool cant_convert(np.zeros(2, dtype="int")) seems to convolute two things: an array cannot be converted to a scalar, int vs bool. Note that cant_convert(np.zeros(2, dtype="bool")) also passes the test. How about this instead (to keep the two things separate)?

    pytest.raises(TypeError, convert, np.zeros(2, dtype="bool"))
    pytest.raises(TypeError, noconvert, np.int32(0))

diff --git a/tests/test_builtin_casters.py b/tests/test_builtin_casters.py
index 3942482..f130dd2 100644
--- a/tests/test_builtin_casters.py
+++ b/tests/test_builtin_casters.py
@@ -280,6 +280,16 @@ def test_int_convert():
     cant_convert(FuzzyThought())
 
 
+def test_int_convert_numpy():
+    np = pytest.importorskip("numpy")
+
+    convert, noconvert = m.int_passthrough, m.int_passthrough_noconvert
+
+    v = np.int32(84)
+    assert convert(v) == 84
+    pytest.raises(TypeError, noconvert, v)
+
+
 def test_tuple(doc):
     """std::pair <-> tuple & std::tuple <-> tuple"""
     assert m.pair_passthrough((True, "test")) == ("test", True)

YannickJadoul · 2020-12-18T20:38:21Z

__index__ is somewhat special. Would it work if __int__ was not called automatically, but __index__ was?

The problem is that both are treated equally by PYBIND11_LONG_CHECK, I believe?
But yes, that would make sense.

I feel stupid for not documenting this better, but I believe this was tackling a complaint from Gitter, where the conversion from float to int is considered noconvert (since float has __int__). I'll check tomorrow if I can figure out the context!

So this was also why I considered this as a bugfix.

My feeling about milestone: given that we now know this PR breaks someone, maybe 2.7 is better?

But @rwgk's comment also makes sense, from that respect, yes. Though ... well, it's very tricky behavior to depend on, I'd say?

Might be nice to include the extra test below in this PR. It's similar to test_numpy_bool but uses int_passthrough, i.e. adds a little bit of coverage.

I'll have a look at the tests as well, but in principle, these NumPy types are just as well just types with or without __int__/__index__ methods, no? So this should already be covered by my explicit, custom test classes DeepThought and/or ShallowThought, I believe? I'll check again, though.

rwgk · 2020-12-18T20:45:00Z

I'll have a look at the tests as well, but in principle, these NumPy types are just as well just types with or without __int__/__index__ methods, no? So this should already be covered by my explicit, custom test classes DeepThought and/or ShallowThought, I believe? I'll check again, though.

Yes, that's true, if you assume numpy behaves that way (seems like a very safe assumption).
Clearly, the test isn't essential, but I'd add those few lines there anyway, mainly to point people (like the mediapipe folks) to, which will imply the message: here, we thought about it, we know, that's the behavior we want, you need to adjust.

bstaletic · 2020-12-18T22:06:24Z

Those numpy types do simply implement __int__. I was going to argue in favour of sticking to milestone 2.6.2, but after thinking about it a bit more, I think @rwgk is right. We're essentially changing what pybind11 considers "an int".

henryiii · 2020-12-18T22:27:00Z

PYBIND11_LONG_CHECK

Does that need to be updated or a different check used? There should be a difference between a class with only __int__ (possibly lossy conversion, like a float), and __index__ ("I am an int").

henryiii · 2020-12-21T14:54:33Z

I think this is a 2.7 fix then, and I also still think __index__ should not be "a conversion", but __int__ should be - that's what those methods mean.

…h np.intc has an __index__ method

YannickJadoul · 2020-12-29T14:46:41Z

I fixed the __index__ thing, demonstrated by a new test (which failed, before). This change should now be way less intrusive (not saying I need it into 2.6.2, but there's a chance it would fit, now).

PyLong_AsLongLong (and others) only call __index__ from 3.8 onwards, but always call __int__, so I think we're fine in this respect? Only if __int__ and __index__ do different things, we're in trouble, but if that's the case, I'm not blaming pybind11. (Also, __int__ being deprecated is now deprecated and gives a warning in Python >= 3.8, so I assume this is part of the same change, where PyLong_AsLongLong switches to ónly take __index__ into account?)

Let's see why Python 2.7 is unhappy. sigh

YannickJadoul · 2020-12-29T14:48:27Z

tests/test_builtin_casters.py

+    assert convert(np.intc(42)) == 42
+    assert noconvert(np.intc(42)) == 42
+
+    assert convert(np.float32(3.14159)) == 3  # This might be wrong/unwanted?


This is also kind of weird. It's because we only explicitly test for float and not for "float-like" types. I'm not sure if there's a general way to recognize "float-like" types, though. We could just add a special case for NumPy float types?

Not really related to this PR, though, so definitely another PR.

I agree, although it looks tricky to implement, therefore I wonder about cost/benefit of working on this.
Pragmatic idea: report the situation in a comment and wait for an opportune moment in the future (e.g. when there is an actual problem that needs to be resolved). My first stab:

# The implicit conversion from np.float is undesirable but difficult to detect, without making the pybind11 # int converter dependent on numpy. Currently Python does not have a generalized concept of float-like types.

We do already have something like this (a special case for NumPy) in the type_caster for bool, though.

But yes, this might be something we'll have to live with? At the very least, this should now be considered a "conversion".

The Google-global testing of this PR had no issues.
I see two behavior changes here: 1. the one currently implemented, and 2. potentially also raising a TypeError for np.float types (with ifdefs I guess to use the numpy api if available).
Given that this PR is exhaustively tested and doesn't disturb anything anymore, I feel it's best to merge now (for 2.6.2), and try the 2. behavior change in a separate PR. The main rationale for this two-stage approach is to use our time efficiently. Putting off this well-tested change will cost extra time.
For the comment, this is my 2nd suggested version:

# The implicit conversion from np.float is undesirable but currently accepted.

Just a hint that we thought about it, leaving all options open for the future.

Agreed. Never meant to lump this in this PR. I really thought it would just be a straightforward one, but once again, Python had some surprises.

Btw, to make things more complicated, np.float to int isn't allowed, because np.float is just a Python float (I'm not joking; this is really the case). Anyway, I'll add 32 and commit this suggestion.

YannickJadoul · 2020-12-29T15:08:35Z

Let's see why Python 2.7 is unhappy. sigh

even deeper sigh Old-style classes...

YannickJadoul · 2020-12-29T15:35:28Z

tests/test_builtin_casters.py

+    cant_convert(ShallowThought())
+    cant_convert(FuzzyThought())
+    if not env.PY2:
+        # I have no clue why __index__ is not picked up by Python 2's PyIndex_check


Too bad for Python 2? I don't know what went wrong, here...

This does work in Python 2.7, though:

Python 2.7.17 (default, Sep 30 2020, 13:38:04) [GCC 7.5.0] on linux2 Type "help", "copyright", "credits" or "license" for more information. >>> x = list(range(50)) >>> class IndexedThought(object): ... def __index__(self): ... return 42 ... >>> x[IndexedThought()] 42

YannickJadoul · 2020-12-29T15:35:52Z

Ready for another review

include/pybind11/cast.h

rwgk · 2020-12-29T18:45:19Z

tests/test_builtin_casters.py

+    assert convert(np.intc(42)) == 42
+    assert noconvert(np.intc(42)) == 42
+
+    assert convert(np.float32(3.14159)) == 3  # This might be wrong/unwanted?


I agree, although it looks tricky to implement, therefore I wonder about cost/benefit of working on this.
Pragmatic idea: report the situation in a comment and wait for an opportune moment in the future (e.g. when there is an actual problem that needs to be resolved). My first stab:

# The implicit conversion from np.float is undesirable but difficult to detect, without making the pybind11 # int converter dependent on numpy. Currently Python does not have a generalized concept of float-like types.

YannickJadoul · 2020-12-29T18:59:00Z

I need to look into these new failures, but, @rwgk, this might break a lot fewer tests, on your end?

YannickJadoul · 2020-12-29T19:43:54Z

I think this should fix the tests (it does so on my machine anyway, where I had failed to run a 3 <= Python version < 3.8, before).

Remaining issue: PyLong_AsLongLong does not call __index__ before 3.8. This is consistent with calling int(...) but not with some_list[...].

Python 3.8:

>>> int(IndexedThought())
42
>>> list(range(100))[IndexedThought()]
42

Python 3.6:

>>> int(IndexedThought())
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: int() argument must be a string, a bytes-like object or a number, not 'IndexedThought'
>>> list(range(100))[IndexedThought()]
42

Python 2.7:

>>> int(IndexedThought())
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: int() argument must be a string or a number, not 'IndexedThought'
>>> list(range(100))[IndexedThought()]
42

So should we make this consistent, and always call PyNumber_Index in versions < 3.8? I'm not sure, but we could. We could also say: "this is just how Python versions < 3.8 handle int conversions, so pybind11 is not going to change that (and you should update anyway)".

Minor extra inconsistency: in versions < 3.6, we now check PyIndex_Check (whether it has an __index__ and is thus an integer numeric type!) to determine if the conversion is noconvert, but later, PyLong_AsLongLong still calls __int__. That's ... weird, at least. And maybe an argument to say we should manually call PyNumber_Index anyway?

rwgk · 2020-12-30T01:24:33Z

I quick observation: https://github.com/google/mediapipe/blob/master/mediapipe/python/packet_test.py passes with the current version of this PR (915423d).
I verified that the mediapipe bindings still have the .noconvert() (that code was last changed in on Nov 17).

rwgk · 2020-12-30T09:11:45Z

tests/test_builtin_casters.py

+    assert convert(np.intc(42)) == 42
+    assert noconvert(np.intc(42)) == 42
+
+    assert convert(np.float32(3.14159)) == 3  # This might be wrong/unwanted?


The Google-global testing of this PR had no issues.
I see two behavior changes here: 1. the one currently implemented, and 2. potentially also raising a TypeError for np.float types (with ifdefs I guess to use the numpy api if available).
Given that this PR is exhaustively tested and doesn't disturb anything anymore, I feel it's best to merge now (for 2.6.2), and try the 2. behavior change in a separate PR. The main rationale for this two-stage approach is to use our time efficiently. Putting off this well-tested change will cost extra time.
For the comment, this is my 2nd suggested version:

# The implicit conversion from np.float is undesirable but currently accepted.

Just a hint that we thought about it, leaving all options open for the future.

YannickJadoul · 2020-12-30T12:48:21Z

Added back to 2.6.2, as @rwgk suggested; we can always switch back to 2.7 (or 2.6.3), if there's no time to get this fully reviewed and in.

There's still the issue from #2698 (comment) that doesn't really feel nice to me. I'm wondering if we shouldn't strive for more consistency in pybind11, and adopt 3.8+ behavior.

YannickJadoul · 2021-01-16T23:20:09Z

I've moved the last question to be decided to a more concrete PR, #2801, demonstrating the implementation is straightforward. All that's left is arguing why we do or don't want this.

…eck (unnoticed because we currently don't have PyPy >= 3.8)

YannickJadoul · 2021-01-17T00:32:26Z

#2801 still discovered an inconsistency with PyPy. @rwgk, do have a look at the latest 0e89b14, but I'm assuming you're fine with this since it doesn't change CPython behavior?

rwgk

Yes, the PyPy fix looks great to me. Thank @YannickJadoul!

YannickJadoul · 2021-01-17T00:43:45Z

Except that ICC isn't happy with this. It's hardly been a day... :-|

henryiii · 2021-01-17T00:45:02Z

Aren't you glad we added that check? 😉

YannickJadoul · 2021-01-17T00:49:02Z

Aren't you glad we added that check? wink

I'm thrilled. Anyway, yeah, otherwise that other PR had to do it.

I'm trying to get rid of that lambda, by means of a macro, I guess. Wouldn't know what else it could be the issue.

Sorry for the mess. Things worked smoothly, locally.

tests/test_builtin_casters.cpp

YannickJadoul · 2021-01-17T01:04:15Z

Submitted as https://foss.heptapod.net/pypy/pypy/-/issues/3383

YannickJadoul · 2021-01-17T01:19:13Z

Oof, all green again 😅
That stress really wasn't necessary, ICC.

YannickJadoul · 2021-01-17T01:27:50Z

@henryiii, #2801 is also green, now, so as far as I'm concerned, this can go in, as the one loose end is handled in #2801.

YannickJadoul added 3 commits November 26, 2020 23:13

Only allow integer type_caster to call __int__ or __index__ method wh…

34893a4

…en conversion is allowed

Remove tests for __index__ as this seems to only be used to convert t…

dc28416

…o int in 3.8+

Take both int and long types into account for Python 2

8f00086

YannickJadoul added this to the v2.6.2 milestone Dec 16, 2020

henryiii modified the milestones: v2.6.2, v2.7 Dec 21, 2020

YannickJadoul changed the title ~~Only allow integer type_caster to call __int__ or __index__ method when conversion is allowed~~ Only allow integer type_caster to call __int__ method when conversion is allowed; always call __index__ Dec 29, 2020

YannickJadoul added 2 commits December 29, 2020 15:04

Add test_numpy_int_convert to assert tests currently fail, even thoug…

05d1a03

…h np.intc has an __index__ method

Also consider __index__ as noconvert to a C++ integer

cecbb98

YannickJadoul commented Dec 29, 2020

View reviewed changes

YannickJadoul added 2 commits December 29, 2020 16:15

New-style classes for Python 2.7; sigh

c4f25f8

Add some tests on types with custom __index__ method

1380611

YannickJadoul commented Dec 29, 2020

View reviewed changes

rwgk reviewed Dec 29, 2020

View reviewed changes

Ignore some tests in Python <3.8

915423d

rwgk approved these changes Dec 30, 2020

View reviewed changes

Update comment about conversion from np.float32 to C++ int

3609fe5

YannickJadoul modified the milestones: v2.7, v2.6.2 Dec 30, 2020

YannickJadoul mentioned this pull request Jan 16, 2021

Always call PyNumber_Index when casting from Python to a C++ integral type, also pre-3.8 #2801

Merged

Workaround difference between CPython and PyPy's different PyIndex_Ch…

0e89b14

…eck (unnoticed because we currently don't have PyPy >= 3.8)

rwgk approved these changes Jan 17, 2021

View reviewed changes

henryiii reviewed Jan 17, 2021

View reviewed changes

tests/test_builtin_casters.cpp Outdated Show resolved Hide resolved

Avoid ICC segfault with py::arg()

b942b62

YannickJadoul force-pushed the noconvert-int branch from c47e21b to b942b62 Compare January 17, 2021 00:55

henryiii merged commit 8449a80 into pybind:master Jan 17, 2021

github-actions bot added the needs changelog Possibly needs a changelog entry label Jan 17, 2021

YannickJadoul deleted the noconvert-int branch January 17, 2021 01:55

henryiii removed the needs changelog Possibly needs a changelog entry label Jan 25, 2021

henryiii mentioned this pull request Jan 26, 2021

chore: prepare for the 2.6.2 release #2821

Merged

rwgk mentioned this pull request Feb 10, 2023

FWD pybind11 google/pybind11clif#2698

Closed

Only allow integer type_caster to call __int__ method when conversion is allowed; always call __index__ #2698

Only allow integer type_caster to call __int__ method when conversion is allowed; always call __index__ #2698

Conversation

YannickJadoul commented Nov 26, 2020 • edited Loading

Description

Suggested changelog entry:

henryiii commented Dec 16, 2020

YannickJadoul commented Dec 16, 2020

rwgk commented Dec 18, 2020

henryiii commented Dec 18, 2020

rwgk commented Dec 18, 2020 • edited Loading

YannickJadoul commented Dec 18, 2020

rwgk commented Dec 18, 2020

bstaletic commented Dec 18, 2020

henryiii commented Dec 18, 2020

henryiii commented Dec 21, 2020

YannickJadoul commented Dec 29, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

YannickJadoul commented Dec 29, 2020

Choose a reason for hiding this comment

YannickJadoul commented Dec 29, 2020

Choose a reason for hiding this comment

YannickJadoul commented Dec 29, 2020

YannickJadoul commented Dec 29, 2020

rwgk commented Dec 30, 2020

Choose a reason for hiding this comment

YannickJadoul commented Dec 30, 2020

YannickJadoul commented Jan 16, 2021

YannickJadoul commented Jan 17, 2021

rwgk left a comment

Choose a reason for hiding this comment

YannickJadoul commented Jan 17, 2021

henryiii commented Jan 17, 2021

YannickJadoul commented Jan 17, 2021

YannickJadoul commented Jan 17, 2021

YannickJadoul commented Jan 17, 2021

YannickJadoul commented Jan 17, 2021

Only allow integer type_caster to call int method when conversion is allowed; always call index #2698

Only allow integer type_caster to call int method when conversion is allowed; always call index #2698

YannickJadoul commented Nov 26, 2020 •

edited

Loading

rwgk commented Dec 18, 2020 •

edited

Loading