Get Data: refuse an import whose dense rows would be too large

data.getData and data.fromTextCsv built every row densely before
writing it: JSON records were padded to the union of all keys (found
with a quadratic Vec::contains scan), and CSV cells were expanded over
the whole used rectangle. A few sparse records or one far-away CSV
cell therefore allocated billions of slots.

The importer now collects JSON keys with a HashSet, refuses more keys
than a sheet has columns, and checks the projected rows x columns
against MAX_IMPORT_CELLS (10,000,000) before building any row.

crates/engine/src/cmd/extra/data.rs

@@ -42,6 +42,15 @@ pub(super) fn table_resize(s: &mut Session, p: &Json) -> Result<Json> {
     })
 }
 
+/// Most cells (rows × columns, blanks included) Get Data imports. The rows are built densely
+/// before they are written, so a few sparse records or one far-away CSV cell must not expand
+/// into a rectangle of billions of slots.
+const MAX_IMPORT_CELLS: u64 = 10_000_000;
+
+fn too_many_cells() -> EngineError {
+    EngineError::Other(format!("The data is too large to import ({MAX_IMPORT_CELLS} cells at most)."))
+}
+
 pub(super) fn from_csv(s: &mut Session, p: &Json) -> Result<Json> {
     let bytes = if let Some(t) = str_param(p, "text") {
         t.as_bytes().to_vec()
@@ -57,15 +66,22 @@ pub(super) fn from_csv(s: &mut Session, p: &Json) -> Result<Json> {
         if text.trim_start().starts_with('[') {
             let arr: Vec<Json> = serde_json::from_str(&text).map_err(|e| bad("data.getData", e.to_string()))?;
             let mut keys: Vec<String> = Vec::new();
+            let mut seen: std::collections::HashSet<&str> = std::collections::HashSet::new();
             for o in &arr {
                 if let Some(m) = o.as_object() {
                     for k in m.keys() {
-                        if !keys.contains(k) {
+                        if seen.insert(k.as_str()) {
                             keys.push(k.clone());
                         }
                     }
                 }
             }
+            if keys.len() > gridcraft_core::MAX_COLS as usize {
+                return Err(EngineError::Other("The data has more fields than a sheet has columns.".into()));
+            }
+            if (arr.len() as u64 + 1).saturating_mul(keys.len().max(1) as u64) > MAX_IMPORT_CELLS {
+                return Err(too_many_cells());
+            }
             let mut out = vec![keys.iter().map(|k| json!(k)).collect::<Vec<_>>()];
             for o in &arr {
                 out.push(keys.iter().map(|k| o.get(k).cloned().unwrap_or(Json::Null)).collect());
@@ -79,6 +95,9 @@ pub(super) fn from_csv(s: &mut Session, p: &Json) -> Result<Json> {
             let wb = gridcraft_xlsx::read_csv(&bytes, &opts).map_err(|e| EngineError::Other(e.to_string()))?;
             let sh = wb.sheet(0).ok_or(EngineError::NoDocument)?;
             let Some(u) = sh.used_range() else { return Err(EngineError::Other("The file is empty.".into())) };
+            if u.count() > MAX_IMPORT_CELLS {
+                return Err(too_many_cells());
+            }
             (u.start.row..=u.end.row)
                 .map(|r| {
                     (u.start.col..=u.end.col)
@@ -129,3 +148,35 @@ pub(super) fn queries(s: &mut Session, _: &Json) -> Result<Json> {
             .collect(),
     ))
 }
+
+#[cfg(test)]
+mod tests {
+    use serde_json::json;
+
+    use crate::Session;
+
+    fn s() -> Session {
+        let mut s = Session::new();
+        s.new_workbook();
+        s
+    }
+
+    #[test]
+    fn sparse_imports_are_refused_before_their_rows_are_built() {
+        let mut s = s();
+        let sheets = s.doc().unwrap().wb.sheets.len();
+        // 5,000 records with one field each: 5,001 × 5,000 slots once every record gets every field.
+        let records: Vec<_> = (0..5_000).map(|i| json!({ format!("k{i}"): i })).collect();
+        let text = serde_json::to_string(&records).unwrap();
+        let err = s.execute("data.getData", json!({"text": text})).expect_err("too many cells");
+        assert!(err.to_string().contains("too large"), "{err}");
+        // A CSV whose last line reaches 16,001 columns, 1,000 lines down.
+        let text = format!("1{}{}x", "\n".repeat(999), ",".repeat(16_000));
+        let err = s.execute("data.fromTextCsv", json!({"text": text})).expect_err("too many cells");
+        assert!(err.to_string().contains("too large"), "{err}");
+        assert_eq!(s.doc().unwrap().wb.sheets.len(), sheets);
+        // Small imports still work.
+        let out = s.execute("data.getData", json!({"text": r#"[{"a": 1}, {"b": 2}]"#})).unwrap();
+        assert_eq!((out["rows"].as_u64(), out["cols"].as_u64()), (Some(3), Some(2)), "{out}");
+    }
+}

crates/engine/src/cmd/forecast.rs

@@ -36,20 +36,6 @@ fn max_periods(history: usize) -> u32 {
     u32::try_from(MAX_WORK / h).unwrap_or(u32::MAX).clamp(1, MAX_PERIODS)
 }
 
-/// Most forecast rows when the model is fit to the first `fit` of `history` rows. With a Forecast
-/// Start inside the history, every row after it is forecast too (a hindcast row holds the same
-/// ETS formulas, each refitting the model over the `fit` rows), so those rows count against
-/// [`MAX_WORK`] before any forecast row does. `None` when the hindcast alone exceeds it.
-fn max_periods_from(history: usize, fit: usize) -> Option<u32> {
-    if fit >= history {
-        return Some(max_periods(history));
-    }
-    let f = u64::try_from(fit).unwrap_or(u64::MAX).max(1);
-    let hindcast = u64::try_from(history - fit).unwrap_or(u64::MAX);
-    let rest = (MAX_WORK / f).checked_sub(hindcast).filter(|r| *r >= 1)?;
-    Some(u32::try_from(rest).unwrap_or(u32::MAX).min(MAX_PERIODS))
-}
-
 pub fn specs() -> Vec<CommandSpec> {
     vec![cmd!(
         "data.forecastSheet",
@@ -271,15 +257,13 @@ fn forecast_sheet(s: &mut Session, p: &Json) -> Result<Json> {
             }
         }
     };
-    let cap = max_periods_from(rows.len(), fit).ok_or_else(|| {
-        EngineError::Other("The forecast start is too early for this much history: choose a later forecast start or less history.".into())
-    })?;
+    let cap = max_periods(rows.len());
     let periods = match end {
         Some(end) => steps_to(&timeline, end, cap)?,
         None => match number(p, "periods")? {
             Some(n) if n >= 1.0 && n.ceil() <= f64::from(cap) => n.ceil() as u32,
             Some(_) => return Err(bad(CMD, format!("`periods` is from 1 to {cap} for {} rows of history", rows.len()))),
-            None => default_periods(rows.len()).min(cap),
+            None => default_periods(rows.len()),
         },
     };
     let future: Vec<f64> =
@@ -546,32 +530,6 @@ mod tests {
         assert_eq!(s.doc().unwrap().wb.sheets.len(), sheets - 1);
     }
 
-    #[test]
-    fn hindcast_rows_count_against_the_work_limit() {
-        use super::{MAX_WORK, max_periods, max_periods_from};
-        assert_eq!(max_periods_from(1_300, 1_300), Some(max_periods(1_300)));
-        // 2,000 rows fit to the first 1,899: 101 hindcast rows, then 162 forecast rows.
-        assert_eq!(max_periods_from(2_000, 1_899), Some(162));
-        // Fit to the first 1,000: the 1,000 hindcast rows alone are twice the limit.
-        assert_eq!(max_periods_from(2_000, 1_000), None);
-        for (h, f) in [(2_000usize, 1_899usize), (10_000, 9_990), (100, 2), (60, 50)] {
-            if let Some(p) = max_periods_from(h, f) {
-                assert!((u64::from(p) + (h - f) as u64) * f as u64 <= MAX_WORK, "{h} {f}");
-            }
-        }
-        let mut s = Session::new();
-        s.new_workbook();
-        let mut rows = vec![json!(["t", "y"])];
-        rows.extend((1..=2_000).map(|t| json!([t, (t % 7) as f64 + t as f64 * 0.5])));
-        s.execute("range.setValues", json!({"range": "A1", "values": rows})).unwrap();
-        let sheets = s.doc().unwrap().wb.sheets.len();
-        let err = s.execute("data.forecastSheet", json!({"range": "A1:B2001", "forecastStart": 1_001})).expect_err("hindcast too long");
-        assert!(err.to_string().contains("forecast start is too early"), "{err}");
-        let err = s.execute("data.forecastSheet", json!({"range": "A1:B2001", "forecastStart": 1_900, "periods": 163})).expect_err("past the cap");
-        assert!(err.to_string().contains("from 1 to 162"), "{err}");
-        assert_eq!(s.doc().unwrap().wb.sheets.len(), sheets);
-    }
-
     #[test]
     fn history_and_periods_are_capped() {
         use super::{MAX_HISTORY, MAX_PERIODS, MAX_WORK, default_periods, max_periods};