Cannot iterate over fields of record array

When iterating over the fields of a record array, numba complains about a not implemented getitem. Accessing the fields by name outside a loop works. I am using numba 0.52, the problem occurs on both Windows and linux

This is the ipython cell that produces the error:

%pylab
import numpy.random
from numba import njit

@njit
def recarriterator(rec):
    for o in 'a','b','c':
        print(rec[o][0])
    return
        
@njit
def recarriterator2(rec):
    print(rec['a'][0])
    print(rec['b'][0])
    print(rec['c'][0])
    return
        
ldd=[]
ldk=[]
keys='a','b','c'
for o in keys:
    #if not (o == 'observation_id' or o == 'report_id' or o == 'sensor_id' or o == 'source_id'):
        ldd.append(random.rand(5))
        ldk.append((o,ldd[-1].dtype))
rec = numpy.rec.fromarrays(ldd, dtype=ldk)
    
recarriterator2(rec)
recarriterator(rec)

The error code:

Using matplotlib backend: Qt5Agg
Populating the interactive namespace from numpy and matplotlib
0.26156099390229226
0.4919898206995005
0.363624558244539
---------------------------------------------------------------------------
TypingError                               Traceback (most recent call last)
<ipython-input-1-56c6c55bddcf> in <module>
     26 
     27 recarriterator2(rec)
---> 28 recarriterator(rec)

~\anaconda3\lib\site-packages\numba\core\dispatcher.py in _compile_for_args(self, *args, **kws)
    413                 e.patch_message(msg)
    414 
--> 415             error_rewrite(e, 'typing')
    416         except errors.UnsupportedError as e:
    417             # Something unsupported is present in the user code, add help info

~\anaconda3\lib\site-packages\numba\core\dispatcher.py in error_rewrite(e, issue_type)
    356                 raise e
    357             else:
--> 358                 reraise(type(e), e, None)
    359 
    360         argtypes = []

~\anaconda3\lib\site-packages\numba\core\utils.py in reraise(tp, value, tb)
     78         value = tp()
     79     if value.__traceback__ is not tb:
---> 80         raise value.with_traceback(tb)
     81     raise value
     82 

TypingError: Failed in nopython mode pipeline (step: nopython frontend)
No implementation of function Function(<built-in function getitem>) found for signature:
 
 >>> getitem(unaligned array(Record(a[type=float64;offset=0],b[type=float64;offset=8],c[type=float64;offset=16];24;False), 1d, C), unicode_type)
 
There are 22 candidate implementations:
  - Of which 20 did not match due to:
  Overload of function 'getitem': File: <numerous>: Line N/A.
    With argument(s): '(unaligned array(Record(a[type=float64;offset=0],b[type=float64;offset=8],c[type=float64;offset=16];24;False), 1d, C), unicode_type)':
   No match.
  - Of which 2 did not match due to:
  Overload in function 'GetItemBuffer.generic': File: numba\core\typing\arraydecl.py: Line 162.
    With argument(s): '(unaligned array(Record(a[type=float64;offset=0],b[type=float64;offset=8],c[type=float64;offset=16];24;False), 1d, C), unicode_type)':
   Rejected as the implementation raised a specific error:
     TypeError: unsupported array index type unicode_type in [unicode_type]
  raised from C:\Users\leopo\anaconda3\lib\site-packages\numba\core\typing\arraydecl.py:68

During: typing of intrinsic-call at <ipython-input-1-56c6c55bddcf> (8)

File "<ipython-input-1-56c6c55bddcf>", line 8:
def recarriterator(rec):
    <source elided>
    for o in 'a','b','c':
        print(rec[o][0])
        ^

[stuartarchibald]

This should work:

import numpy as np
import numba as nb
from numba import literal_unroll

dt = np.dtype([("a", np.uint8), ("b", np.uint8)])
dt_names = dt.names

@nb.njit
def test():
    rec = np.zeros(1, dtype=dt)
    for f in literal_unroll(dt_names):
        rec[0][f] = 1
    return rec

print(test())

note that literal_unroll is imported explicitly. This is a known bug, think it’s a duplicate of https://github.com/numba/numba/issues/5344.

[esc]

@pwuertz thank you for submitting this. I can confirm that I am able to reproduce on current main (3781d6836d78f3467b02500b19a889039bdaf3bb) with the following error:

Traceback (most recent call last):
  File "/Users/vhaenel/git/numba/issue_6634.py", line 14, in <module>
    test()
  File "/Users/vhaenel/git/numba/numba/core/dispatcher.py", line 468, in _compile_for_args
    error_rewrite(e, 'typing')
  File "/Users/vhaenel/git/numba/numba/core/dispatcher.py", line 409, in error_rewrite
    raise e.with_traceback(None)
numba.core.errors.TypingError: Failed in nopython mode pipeline (step: nopython frontend)
No implementation of function Function(<built-in function setitem>) found for signature:

 >>> setitem(Record(a[type=uint8;offset=0],b[type=uint8;offset=1];2;False), unicode_type, Literal[int](1))

There are 16 candidate implementations:
  - Of which 16 did not match due to:
  Overload of function 'setitem': File: <numerous>: Line N/A.
    With argument(s): '(Record(a[type=uint8;offset=0],b[type=uint8;offset=1];2;False), unicode_type, int64)':
   No match.

During: typing of setitem at /Users/vhaenel/git/numba/issue_6634.py (11)

File "issue_6634.py", line 11:
def test():
    <source elided>
    for f in nb.literal_unroll(dt_names):
        rec[0][f] = 1