Object mode code fails to load from cache

Reporting a bug

• I have tried using the latest released version of Numba (most recent is visible in the change log (https://github.com/numba/numba/blob/master/CHANGE_LOG).
• I have included a self contained code sample to reproduce the problem. i.e. it’s possible to run as ‘python bug.py’.

Object mode code below fails to load from cache. The problem is that the objmode dispatcher is serialized and compiled during runtime (see here). However, the function IR saved is invalid. The issue is that func_ir.copy() is used to avoid compiler’s mutations, but this doesn’t copy the IR nodes, which are modified in places like here. Avoiding these copy() calls before dispatcher compilation but adding remove_dels after compilation solves the problem. Are these copy() calls necessary?

import numba
import pandas as pd

@numba.njit(cache=True)
def f():

  with numba.objmode(T="int64"):
    typemap = {}
    typemap.update({i:"Int64" for i in range(3)})
    df = pd.DataFrame()
    arrs = []
    for i in range(df.shape[1]):
      arrs.append(df.iloc[:, i].values)
    T = len(arrs)
  return (T,)


f()

 ✗ python ../tmp/objmode_test.py
 ✗ python ../tmp/objmode_test.py
Traceback (most recent call last):
  File "/Users/ehsan/dev/bodo/../tmp/objmode_test.py", line 18, in <module>
    f()
  File "/Users/ehsan/dev/numba/numba/core/pythonapi.py", line 1682, in _call_objmode_dispatcher
    entrypt = dispatcher.compile(argtypes)
  File "/Users/ehsan/dev/numba/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
  File "/Users/ehsan/dev/numba/numba/core/dispatcher.py", line 1329, in compile
    return super().compile(sig)
  File "/Users/ehsan/dev/numba/numba/core/dispatcher.py", line 1210, in compile
    cres = compiler.compile_ir(typingctx=self.typingctx,
  File "/Users/ehsan/dev/numba/numba/core/compiler.py", line 724, in compile_ir
    norw_cres = compile_local(func_ir.copy(), norw_flags)
  File "/Users/ehsan/dev/numba/numba/core/compiler.py", line 720, in compile_local
    return pipeline.compile_ir(func_ir=the_ir, lifted=lifted,
  File "/Users/ehsan/dev/numba/numba/core/compiler.py", line 438, in compile_ir
    return self._compile_ir()
  File "/Users/ehsan/dev/numba/numba/core/compiler.py", line 499, in _compile_ir
    return self._compile_core()
  File "/Users/ehsan/dev/numba/numba/core/compiler.py", line 471, in _compile_core
    raise e
  File "/Users/ehsan/dev/numba/numba/core/compiler.py", line 462, in _compile_core
    pm.run(self.state)
  File "/Users/ehsan/dev/numba/numba/core/compiler_machinery.py", line 343, in run
    raise patched_exception
  File "/Users/ehsan/dev/numba/numba/core/compiler_machinery.py", line 334, in run
    self._runPass(idx, pass_inst, state)
  File "/Users/ehsan/dev/numba/numba/core/compiler_lock.py", line 35, in _acquire_compile_lock
    return func(*args, **kwargs)
  File "/Users/ehsan/dev/numba/numba/core/compiler_machinery.py", line 289, in _runPass
    mutated |= check(pss.run_pass, internal_state)
  File "/Users/ehsan/dev/numba/numba/core/compiler_machinery.py", line 262, in check
    mangled = func(compiler_state)
  File "/Users/ehsan/dev/numba/numba/core/typed_passes.py", line 794, in run_pass
    post_proc.run(emit_dels=False)
  File "/Users/ehsan/dev/numba/numba/core/postproc.py", line 78, in run
    self.func_ir.blocks = transforms.canonicalize_cfg(self.func_ir.blocks)
  File "/Users/ehsan/dev/numba/numba/core/transforms.py", line 315, in canonicalize_cfg
    return canonicalize_cfg_single_backedge(blocks)
  File "/Users/ehsan/dev/numba/numba/core/transforms.py", line 244, in canonicalize_cfg_single_backedge
    cfg = compute_cfg_from_blocks(blocks)
  File "/Users/ehsan/dev/numba/numba/core/analysis.py", line 243, in compute_cfg_from_blocks
    cfg.add_edge(k, target)
  File "/Users/ehsan/dev/numba/numba/core/controlflow.py", line 112, in add_edge
    raise ValueError("Cannot add edge as dest node %s not in nodes %s" %
ValueError: Failed in object mode pipeline (step: remove phis nodes)
Cannot add edge as dest node 119 not in nodes {102, 70, 72, 12, 14, 112}

[njriasan]

IIRC this got discussed and we’re waiting on a PR or example of an approach to resolve this issue? ISTR from the discussion that there was a risk that removing all copies would result in unintended mutation of the original in some places.

Hi @gmarkall. Sorry about that. I did have a PR and reran all of the tests. None of the ones that we suspected would induce issues did, but there was an issue in another test. I will open the PR after I debug it, I just haven’t had time yet (and will this week).

[gmarkall]

IIRC this got discussed and we’re waiting on a PR or example of an approach to resolve this issue? ISTR from the discussion that there was a risk that removing all copies would result in unintended mutation of the original in some places.