not-sure-if-bug: strange mypy behavior in the latest sqlglot

[pkit]

from sqlglot import parse_one, exp

a = parse_one("SELECT * FROM t").find(exp.Table)
print(a)

$ python -m mypy test.py
test.py:3: error: <nothing> has no attribute "find"  [attr-defined]
Found 1 error in 1 file (checked 1 source file)

Could be a mypy bug or the usual mypy idiocy, but just FYI.

[georgesittas]

Not sure why this is happening.. any execution path in parse_one should either raise or return an Expression.

IIRC, <nothing> means that mypy failed to solve the type constraints [1, 2, 3, 4] introduced by the use of TypeVars, so my guess is that it's either a mypy bug, or our @t.overload annotations are wrong for parse_one.

On another note: I think we should clean up IntoType at some point. It feels inelegant to have a str variant, even more so if the only reason to do have it is to enable the hack with "JoinType" in EXPRESSION_PARSERS.

[1] python/mypy#3283
[2] https://github.com/python/mypy/blob/6f2bfff521e708f015af1eec5118db4600b829be/mypy/typeops.py#L330
[3] https://github.com/python/mypy/blob/6f2bfff521e708f015af1eec5118db4600b829be/mypy/types.py#L3013
[4] https://github.com/python/mypy/wiki/Type-Checker#join-and-meet

[tobymao]

can you repro this @georgesittas ?

[georgesittas]

can you repro this @georgesittas ?

Yeah and it seems like an issue in general, not just our latest release.

[georgesittas]

I can also verify that by getting rid of the @t.overloads, the error disappears

# With just this mypy stopped complaining
def parse_one(
    sql: str,
    read: DialectType = None,
    into: t.Optional[exp.IntoType] = None,
    **opts,
) -> Expression:
    # Definition goes here

[pkit]

Yeah and it seems like an issue in general, not just our latest release.

I have upgraded both mypy and sqlglot recently. But I'm not sure anymore. Maybe it was indeed happening before, but just silenced.
In any case my investigation also didn't find any problems in sqlglot type hits.
That's why "not-sure-if-bug"

[pkit]

Looks like it's a mypy bug: python/mypy#10817

[georgesittas]

@pkit take a look at the PR if you like, it solves the issue.

[pkit]

Looks good, I still don't like mypy though.

[georgesittas]

Yeah.. It has some rough edges, but I like to think that in the long run it helps with the development process by catching errors early vs at runtime. Hopefully it also helps by "documenting" the code more than making it unreadable. :-)