Errors and deprecation warnings from numba 0.58

[bdice]

The release of numba 0.58 has surfaced several errors and deprecation warnings. Below is an example:

numba.core.errors.TypingError: Internal error at resolving type of attribute "sum" of "group".
...
numba.core.errors.NumbaPendingDeprecationWarning: Failed in cuda mode pipeline (step: nopython frontend)
Code using Numba extension API maybe depending on 'old_style' error-capturing, which is deprecated and will be replaced by 'new_style' in a future release. See details at https://numba.readthedocs.io/en/latest/reference/deprecation.html#deprecation-of-old-style-numba-captured-errors

See details for a full log.

Details

____________________ test_groupby_apply_return_df[<lambda>] ____________________
[gw0] linux -- Python 3.10.13 /pyenv/versions/3.10.13/bin/python

typingctx = <numba.cuda.target.CUDATypingContext object at 0x7f72be9a3340>
targetctx = <numba.cuda.target.CUDATargetContext object at 0x7f7152cdb100>
interp = <numba.core.ir.FunctionIR object at 0x7f7152b50a60>
args = (Record([('a', {'type': Group(int64, int64), 'offset': 0, 'alignment': None, 'title': None, }), ('b', {'type': Group(int64, int64), 'offset': 24, 'alignment': None, 'title': None, })], 48, True),)
return_type = None, locals = {}, raise_errors = True

    def type_inference_stage(typingctx, targetctx, interp, args, return_type,
                             locals={}, raise_errors=True):
        if len(args) != interp.arg_count:
            raise TypeError("Mismatch number of argument types")
        warnings = errors.WarningsFixer(errors.NumbaWarning)
    
        infer = typeinfer.TypeInferer(typingctx, interp, warnings)
        callstack_ctx = typingctx.callstack.register(targetctx.target, infer,
                                                     interp.func_id, args)
        # Setup two contexts: 1) callstack setup/teardown 2) flush warnings
        with callstack_ctx, warnings:
            # Seed argument types
            for index, (name, ty) in enumerate(zip(interp.arg_names, args)):
                infer.seed_argument(name, index, ty)
    
            # Seed return type
            if return_type is not None:
                infer.seed_return(return_type)
    
            # Seed local types
            for k, v in locals.items():
                infer.seed_type(k, v)
    
            infer.build_constraint()
            # return errors in case of partial typing
>           errs = infer.propagate(raise_errors=raise_errors)

/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/typed_passes.py:91: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <numba.core.typeinfer.TypeInferer object at 0x7f7153827d60>
raise_errors = True

    def propagate(self, raise_errors=True):
        newtoken = self.get_state_token()
        oldtoken = None
        # Since the number of types are finite, the typesets will eventually
        # stop growing.
    
        while newtoken != oldtoken:
            self.debug.propagate_started()
            oldtoken = newtoken
            # Errors can appear when the type set is incomplete; only
            # raise them when there is no progress anymore.
            errors = self.constraints.propagate(self)
            newtoken = self.get_state_token()
            self.debug.propagate_finished()
        if errors:
            if raise_errors:
                force_lit_args = [e for e in errors
                                  if isinstance(e, ForceLiteralArg)]
                if not force_lit_args:
>                   raise errors[0]
E                   numba.core.errors.TypingError: Internal error at resolving type of attribute "sum" of "group".
E                   'sum'
E                   During: typing of get attribute at /__w/cudf/cudf/python/cudf/cudf/tests/test_groupby.py (678)
E                   Enable logging at debug level for details.
E                   
E                   File "python/cudf/cudf/tests/test_groupby.py", line 678:
E                       def func(df):
E                           <source elided>
E                   
E                   @pytest.mark.parametrize("func", [lambda group: group.sum()])
E                   ^

/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/typeinfer.py:1086: TypingError

During handling of the above exception, another exception occurred:

func = <function <lambda> at 0x7f718596d750>

    @pytest.mark.parametrize("func", [lambda group: group.sum()])
    def test_groupby_apply_return_df(func):
        # tests a UDF that reduces over a dataframe
        # and produces a series with the original column names
        # as its index, such as lambda group: group.sum() + group.min()
        df = cudf.DataFrame({"a": [1, 1, 2, 2], "b": [1, 2, 3, 4]})
        pdf = df.to_pandas()
    
        expect = pdf.groupby("a").apply(func)
>       got = df.groupby("a").apply(func)

python/cudf/cudf/tests/test_groupby.py:687: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/pyenv/versions/3.10.13/lib/python3.10/site-packages/nvtx/nvtx.py:115: in inner
    result = func(*args, **kwargs)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/cudf/core/groupby/groupby.py:1420: in apply
    if (not grouped_values._has_nulls) and _can_be_jitted(
/pyenv/versions/3.10.13/lib/python3.10/site-packages/cudf/core/udf/groupby_utils.py:226: in _can_be_jitted
    _get_udf_return_type(dataframe_group_type, func, args)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/nvtx/nvtx.py:115: in inner
    result = func(*args, **kwargs)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/cudf/core/udf/utils.py:88: in _get_udf_return_type
    ptx, output_type = cudautils.compile_udf(func, compile_sig)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/cudf/utils/cudautils.py:126: in compile_udf
    ptx_code, return_type = cuda.compile_ptx_for_current_device(
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/cuda/compiler.py:283: in compile_ptx_for_current_device
    return compile_ptx(pyfunc, sig, debug=debug, lineinfo=lineinfo,
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_lock.py:35: in _acquire_compile_lock
    return func(*args, **kwargs)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/cuda/compiler.py:253: in compile_ptx
    cres = compile_cuda(pyfunc, return_type, args, debug=debug,
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_lock.py:35: in _acquire_compile_lock
    return func(*args, **kwargs)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/cuda/compiler.py:194: in compile_cuda
    cres = compiler.compile_extra(typingctx=typingctx,
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler.py:770: in compile_extra
    return pipeline.compile_extra(func)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler.py:461: in compile_extra
    return self._compile_bytecode()
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler.py:529: in _compile_bytecode
    return self._compile_core()
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler.py:508: in _compile_core
    raise e
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler.py:495: in _compile_core
    pm.run(self.state)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_machinery.py:368: in run
    raise patched_exception
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_machinery.py:356: in run
    self._runPass(idx, pass_inst, state)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_lock.py:35: in _acquire_compile_lock
    return func(*args, **kwargs)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_machinery.py:311: in _runPass
    mutated |= check(pss.run_pass, internal_state)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/compiler_machinery.py:273: in check
    mangled = func(compiler_state)
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/typed_passes.py:110: in run_pass
    typemap, return_type, calltypes, errs = type_inference_stage(
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/typed_passes.py:76: in type_inference_stage
    with callstack_ctx, warnings:
/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/errors.py:512: in __exit__
    self.flush()
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <numba.core.errors.WarningsFixer object at 0x7f71720e0af0>

    def flush(self):
        """
        Emit all stored warnings.
        """
        def key(arg):
            # It is possible through codegen to create entirely identical
            # warnings, this leads to comparing types when sorting which breaks
            # on Python 3. Key as str() and if the worse happens then `id`
            # creates some uniqueness
            return str(arg) + str(id(arg))
    
        for (filename, lineno, category), messages in sorted(
                self._warnings.items(), key=key):
            for msg in sorted(messages):
>               warnings.warn_explicit(msg, category, filename, lineno)
E               numba.core.errors.NumbaPendingDeprecationWarning: Failed in cuda mode pipeline (step: nopython frontend)
E               Code using Numba extension API maybe depending on 'old_style' error-capturing, which is deprecated and will be replaced by 'new_style' in a future release. See details at https://numba.readthedocs.io/en/latest/reference/deprecation.html#deprecation-of-old-style-numba-captured-errors
E               Exception origin:
E                 File "/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/types/npytypes.py", line 189, in typeof
E                   return self.fields[key].type

/pyenv/versions/3.10.13/lib/python3.10/site-packages/numba/core/errors.py:505: NumbaPendingDeprecationWarning

https://github.com/rapidsai/cudf/actions/runs/6264918730/job/17013087530?pr=14156#step:9:1773

[bdice]

cc: @gmarkall @brandon-b-miller

[brandon-b-miller]

Taking a look at this 👍

[brandon-b-miller]

Having poked around a bit I believe some of the changes needed to fix this overlap/are related to changes in #13854. I think it might fix some of these things outright, or at the very least it centralizes the kinds of numba errors we raise in such a way that we could probably fix things in fewer places.

[bdice]

@brandon-b-miller Is #13854 ready to merge? It would need to retarget 24.02. I'd like to unpin numba<0.58 for the 24.02 release so ironing out any known issues would be good.

[brandon-b-miller]

Thanks for bumping this. Will investigate and see what if anything still needs to be done here.

[brandon-b-miller]

The addition of this configuration option in #13854 seems to have resolved this.

[bdice]

@brandon-b-miller Can this issue be closed, then? Or should it be closed by #14616?

[brandon-b-miller]

I think it should be closed by #14616. Will update