[BUG] deeplake v4.5.3 produces a segmentation fault with WHERE (NOT ((f9['e1'] IS NOT NULL)))

[zhuang-keju]

Severity

P0 - Critical breaking issue or missing functionality

Current Behavior

Deeplake produces a segmentation fault when queried with the clause WHERE (NOT ((f9['e1'] IS NOT NULL))).

Steps to Reproduce

The following testcase produces this error: deeplake_bug_trigger.py.txt. The sample logic of the code is shown below.

import deeplake as deeplake_lib
from deeplake import types as dl_types
import numpy as np
import faulthandler

faulthandler.enable()

# create collection, reset old collections if any
path = "mem://test_collection"
try:
    deeplake_lib.delete(path)
except:
    pass
ds = deeplake_lib.create(path)


# define schema
ds.add_column("f0", dl_types.Bool())
ds.add_column("f1", dl_types.Int64())
ds.add_column("f2", dl_types.Embedding(35, dtype='float32'))
ds.add_column("f3", dl_types.Text())
ds.add_column("f4", dl_types.Float32())
ds.add_column("f5", dl_types.Dict())
ds.add_column("f6", dl_types.Array("float32", 1))
ds.add_column("f7", dl_types.Text())
ds.add_column("f8", dl_types.Dict())
ds.add_column("f9", dl_types.Dict())
ds.add_column("f10", dl_types.Embedding(95, dtype='float32'))

# the data to be inserted
data_list = [ ... ]

# insert data
pkey_name = 'f7'
pk_to_idx = {}
if pkey_name is not None and len(ds) > 0:
    col = ds[pkey_name].numpy().flatten()
    for i, v in enumerate(col):
        pk_to_idx[v] = i
for data in data_list:
    pk = data.get(pkey_name) if pkey_name is not None else None
    idx = pk_to_idx.get(pk) if pk is not None else None
    if idx is None:
        row = {k: [v] for k, v in data.items()}
        ds.append(row)
        if pk is not None:
            pk_to_idx[pk] = len(ds) - 1
    else:
        for field, v in data.items():
            if field in ['f0', 'f1', 'f2', 'f3', 'f4', 'f5', 'f6', 'f7', 'f8', 'f9', 'f10']:
                ds[field][idx] = v
ds.commit()

# create index
ds["f1"].create_index(dl_types.Inverted)
ds["f3"].create_index(dl_types.Inverted)
ds["f4"].create_index(dl_types.Inverted)
ds["f8"].create_index(dl_types.Inverted)

# build query
target_vector = [0, -0.73974, -0.56952, 0.32329, 0, 0.18876, 0, -0.48297, -0.76431, -0.561, 0, 0.38881, 0, -0.51161, 0, 0.3196, -0.50481, 0.36672, 0.7916, 0.72589, -0.15253, -0.81857, 0.54581, 0.54685, 0.53233, 0.87906, 0.46351, 0.80606, 0.02326, -0.99057, 0.01479, 0.74169, 0.72541, 0.48778, 0.19673]
anns_field = 'f2'
limit = 8875
q = "SELECT * FROM (SELECT *, COSINE_SIMILARITY(f2, ARRAY[0.0,-0.73974,-0.56952,0.32329,0.0,0.18876,0.0,-0.48297,-0.76431,-0.561,0.0,0.38881,0.0,-0.51161,0.0,0.3196,-0.50481,0.36672,0.7916,0.72589,-0.15253,-0.81857,0.54581,0.54685,0.53233,0.87906,0.46351,0.80606,0.02326,-0.99057,0.01479,0.74169,0.72541,0.48778,0.19673]) AS distance_score ) WHERE (NOT ((f9['e1'] IS NOT NULL))) ORDER BY distance_score DESC LIMIT 8875"

# this query should trigger the segmentation fault
view = ds.query(q)

Expected/Desired Behavior

ds.query(q) should return the matching data.

Python Version

3.10

OS

Ubuntu 22.04

IDE

No response

Packages

No response

Additional Context

No response

Possible Solution

No response

Are you willing to submit a PR?

• I'm willing to submit a PR (Thank you!)