Identifier case normalization causes failures (aliases + properties)

[prrao87]

Summary

The Cypher parser/planner lowercases unquoted identifiers, but the underlying schema and projection layer remain case‑sensitive. So anything camelCase in the query (aliases or property names) gets implicitly converted to lowercase during planning, and then lookup fails because the schema still has the original casing.

See the graph-benchmark repo for a real example.

• Query writes numFollowers → planner looks for numfollowers → schema has numFollowers → error.
• Query writes p.isMarried → planner looks for ismarried → schema has isMarried → error.
• So the bug is an internal mismatch: identifier normalization is happening in some parts of the pipeline but not consistently applied to schema lookup / binding.

Expected (Cypher‑like):

• Unquoted identifiers should be case‑insensitive OR
• Quoted identifiers should be supported to preserve case OR
• Ideally, there should be a config file or something similar to disable normalization.

Example

Alias case

import pyarrow as pa
from lance_graph import GraphConfig, CypherQuery

people = pa.table({"id": [1, 2]})
follows = pa.table({"src": [1], "dst": [2]})
cfg = (
    GraphConfig.builder()
    .with_node_label("Person", "id")
    .with_relationship("FOLLOWS", "src", "dst")
    .build()
)
datasets = {"Person": people, "FOLLOWS": follows}

query = """
MATCH (a:Person)-[:FOLLOWS]->(b:Person)
RETURN count(a.id) AS numFollowers
ORDER BY numFollowers DESC
"""
res = CypherQuery(query).with_config(cfg).execute(datasets)
print(res.to_pylist())

Actual

 Schema error: No field named numfollowers. ... Did you mean 'numFollowers'?

Property case

import pyarrow as pa
from lance_graph import GraphConfig, CypherQuery

people = pa.table({"id": [1], "isMarried": [True]})
cfg = GraphConfig.builder().with_node_label("Person", "id").build()
datasets = {"Person": people}

res = CypherQuery("MATCH (p:Person) RETURN p.isMarried").with_config(cfg).execute(datasets)
print(res.to_pylist())

Expected

Case-insensitive lookup for unquoted identifiers, or support for quoted identifiers / config to disable normalization.

Actual

Schema error: No field named ismarried. Did you mean 'person.isMarried'?

Proposal

Rather than the user manually lowercasing all column names and aliases to make queries work, does it make sense to treat column names as case-insensitive? (like Postgres).

Environment

The following environment was used to test this:

lance-graph 0.4.0
Python 3.13
macOS Tahoe 26.2

[ChunxuTang]

Thanks for raising the issue.

I think it's rooted in the syntax inconsistency between Cypher and SQL:

• Cypher: case-sensitive identifiers
• SQL(DataFusion): case-insensitive identifiers

So when processing the query, DataFusion normalizes the property to lower case, but later refers to the arrow dataset, which still uses the camel case column.

One possible way is to set config.sql_parser.enable_ident_normalization = false explicitly in the DataFusion processing, so DataFusion won't normalize it anymore. Cc @beinan in case you also have any suggestions.

[prrao87]

I think this is resolved to a better degree than before in 0.5.0. But am unsure if this is the complete solution. Here's how the alias casing looks:

import pyarrow as pa
from lance_graph import GraphConfig, CypherQuery

people = pa.table({"id": [1, 2]})
follows = pa.table({"src": [1], "dst": [2]})
cfg = (
    GraphConfig.builder()
    .with_node_label("Person", "id")
    .with_relationship("FOLLOWS", "src", "dst")
    .build()
)
datasets = {"Person": people, "FOLLOWS": follows}

query = """
MATCH (a:Person)-[:FOLLOWS]->(b:Person)
RETURN count(a.id) AS numFollowers
ORDER BY numFollowers DESC
"""
res = CypherQuery(query).with_config(cfg).execute(datasets)
print(res.to_pylist())

Gives:

Traceback (most recent call last):
  File "/Users/prrao/code/graph-benchmark-ldbc-snb/lance_graph/t.py", line 19, in <module>
    res = CypherQuery(query).with_config(cfg).execute(datasets)
ValueError: Query planning error: Failed to project after aggregate: Schema error: No field named numfollowers. Column names are case sensitive. You can use double quotes to refer to the "numfollowers" column or set the datafusion.sql_parser.enable_ident_normalization configuration. Did you mean 'numFollowers'?.

The error message is a little confusing because it makes it look like a schema error during aliasing, and the suggestion numFollowers is exactly what the query is using in the first place. So the user action isn't clear here.

However, setting the alias name to numfollowers works:

MATCH (a:Person)-[:FOLLOWS]->(b:Person)
RETURN count(a.id) AS numfollowers
ORDER BY numfollowers DESC

Gives:

[{'numfollowers': 1}]

I think it's worth keeping this issue open to handle the alias case a bit better.