Cross-dataset arithmetic fails — alias syntax not parsed, multi-source queries broken for datasets

[jayendra13]

Summary

Two related issues prevent computing expressions across two datasets (e.g., forecast.temp - target.temp):

1. Alias syntax doesn't parse for datasets: FROM dataset1 a, dataset2 b fails with a parse error.
2. Multi-variable SELECT from datasets not supported: Even within a single dataset, SELECT u_wind - v_wind FROM climate fails.

Reproduction

Issue 1: Alias syntax parse failure

REGISTER DATASET climate FROM 'zarr:///path/to/tutorial_climate.zarr';

-- Expected: compute self-difference (should be all zeros)
-- Actual: parse error
SELECT a.temperature - b.temperature FROM climate a, climate b;
-- Error: query error: unexpected trailing tokens after statement: Ident("a")

The parser doesn't recognize FROM <dataset_name> <alias> — it treats the alias as a new statement.

Issue 2: Multi-variable expressions from datasets

-- Expected: element-wise difference between u_wind and v_wind arrays
-- Actual: error
SELECT u_wind - v_wind FROM climate;
-- Error: not implemented: multi-variable SELECT from datasets is not yet supported (found: u_wind, v_wind)

When a dataset has multiple data variables, the query planner can't resolve expressions that reference more than one variable.

Expected Behavior

Cross-dataset queries

REGISTER DATASET era5 FROM 'zarr:///path/to/era5.zarr';
REGISTER DATASET hres FROM 'zarr:///path/to/hres.zarr';

-- Should align on shared dimensions (lat, lon, time) and compute element-wise difference
SELECT a.temperature - b.temperature FROM era5 a, hres b;

Multi-variable expressions

-- Should compute element-wise difference between two variables sharing the same dimensions
SELECT u_wind - v_wind FROM climate;

-- Wind speed from components
SELECT sqrt(u_wind * u_wind + v_wind * v_wind) FROM climate;

Context

Cross-dataset arithmetic is essential for forecast evaluation metrics:

• MAE = avg_cells(abs(forecast.temp - target.temp))
• RMSE = sqrt(avg_cells((forecast.temp - target.temp)^2))

The array-level query system (REGISTER ARRAY) does support multi-source queries with aliases (FROM arr1 a, arr2 b) via the join logic in crates/arrdb-query/src/planner/logical.rs:209-240. The gap is in the dataset-to-array resolution layer — dataset queries need to expand variable references into the underlying array sources.

Files

• crates/arrdb-query/src/parser/stmt.rs — FROM clause parsing for datasets may not handle aliases
• crates/arrdb-query/src/planner/logical.rs:90-106 — multi-source plan building (works for arrays)
• crates/arrdb-exec/src/session.rs — dataset query path needs multi-variable expression support