ux(sql): unhelpful assertion error message: `AssertionError: num_froms == 2`

Here is one way to trigger it:

>>> import numpy as np
>>> import pandas as pd
>>> import ibis
>>> df = pd.DataFrame({"a": np.random.randn(10000)})
>>> t = ibis.memtable(df)
>>> t.mutate(chunk=ibis.row_number() // 100).group_by(ibis._.chunk).aggregate(b=t.a.mean()).execute()
Traceback (most recent call last):
  Input In [5] in <cell line: 5>
    t.mutate(chunk=ibis.row_number() // 100).group_by(ibis._.chunk).aggregate(b=t.a.mean()).execute()
  File ~/code/ibis/ibis/expr/types/core.py:274 in execute
    return self._find_backend(use_default=True).execute(
  File ~/code/ibis/ibis/backends/base/sql/__init__.py:247 in execute
    sql = query_ast.compile()
  File ~/code/ibis/ibis/backends/base/sql/compiler/base.py:39 in compile
    compiled_queries = [q.compile() for q in self.queries]
  File ~/code/ibis/ibis/backends/base/sql/compiler/base.py:39 in <listcomp>
    compiled_queries = [q.compile() for q in self.queries]
  File ~/code/ibis/ibis/backends/base/sql/alchemy/query_builder.py:186 in compile
    frag = step(frag)
  File ~/code/ibis/ibis/backends/base/sql/alchemy/query_builder.py:279 in _add_select
    assert num_froms == 1, f"num_froms == {num_froms:d}"
AssertionError: num_froms == 2

while using a deferred field reference would have worked:

>>> t.mutate(chunk=ibis.row_number() // 100).group_by(ibis._.chunk).aggregate(b=ibis._.a.mean()).execute()
    chunk         b
0       0 -0.002539
1       1  0.211258
2       2 -0.034153
3       3  0.054345
4       4 -0.063897
..    ...       ...
95     83  0.077232
96     87 -0.018670
97     91 -0.153058
98     95  0.062328
99     99 -0.131931

[100 rows x 2 columns]

I am not sure if this is a bug or just a bad error message.

[cpcloud]

Generally speaking you’ll probably have a bad time if you reference tables further down the tree than the current table. In some cases ibis tries to be smart and infer what you mean (in this case when you reference t), but fails to actually be smart and this is a case of that.

[cpcloud]

The assertion is correct but you shouldn’t be hitting it. We need to catch the error earlier and give a more helpful and more usefully-typed message.