address review: build wildcard literal masks lazily with a bounded cache

After a mismatch the matcher restarted the pattern after the last
star at the next text position, so a long repetitive cell against a
near-matching pattern cost text x pattern comparisons. AutoFilter
custom criteria, Find and Replace all call it once per cell.

wildcard() now runs the pattern as a bit-parallel automaton: one pass
over the text, one step per 64 pattern positions for each character.
Results are unchanged; a test compares it with the old matcher on
every short text and pattern over a small alphabet.

crates/engine/src/cmd/edit/find.rs

@@ -14,8 +14,13 @@ fn matches(hay: &str, what: &str, case: bool, whole: bool) -> bool {
 }
 
 /// `*`/`?` wildcard match with `~` escapes.
+///
+/// The pattern runs as a bit-parallel automaton over the text: one pass, each character costing
+/// one step per 64 pattern positions. Backtracking to the last `*` after a mismatch rescans the
+/// text and costs text × pattern comparisons on a long, repetitive cell, for every cell a
+/// filter or Find visits.
 pub(crate) fn wildcard(text: &str, pat: &str) -> bool {
-    let t: Vec<char> = text.chars().collect();
+    // The pattern as (character, escaped) items, runs of `*` folded into one.
     let mut p: Vec<(char, bool)> = Vec::new();
     let mut it = pat.chars();
     while let Some(c) = it.next() {
@@ -23,32 +28,78 @@ pub(crate) fn wildcard(text: &str, pat: &str) -> bool {
             if let Some(n) = it.next() {
                 p.push((n, true));
             }
-        } else {
+        } else if c != '*' || p.last() != Some(&('*', false)) {
             p.push((c, false));
         }
     }
-    let (mut ti, mut pi) = (0usize, 0usize);
-    let (mut star, mut mark) = (None, 0usize);
-    while ti < t.len() {
-        if pi < p.len() && (p[pi] == ('?', false) || (p[pi].0 == t[ti] && p[pi] != ('*', false))) {
-            ti += 1;
-            pi += 1;
-        } else if pi < p.len() && p[pi] == ('*', false) {
-            star = Some(pi);
-            mark = ti;
-            pi += 1;
-        } else if let Some(sp) = star {
-            pi = sp + 1;
-            mark += 1;
-            ti = mark;
-        } else {
-            return false;
+    let words = (p.len() + 1).div_ceil(64);
+    let bit = |set: &mut [u64], i: usize| set[i / 64] |= 1 << (i % 64);
+    // Positions holding a `*` and a `?`, and where each literal character sits in the pattern.
+    let mut stars = vec![0u64; words];
+    let mut any = vec![0u64; words];
+    let mut at: std::collections::HashMap<char, Vec<usize>> = std::collections::HashMap::new();
+    for (i, item) in p.iter().enumerate() {
+        match item {
+            ('*', false) => bit(&mut stars, i),
+            ('?', false) => bit(&mut any, i),
+            (c, _) => at.entry(*c).or_default().push(i),
         }
     }
-    while pi < p.len() && p[pi] == ('*', false) {
-        pi += 1;
+    // A literal's bit mask is built when the text first reads it, and kept for the first
+    // `MAX_CACHED_MASKS` literals: a long pattern of mostly distinct characters would otherwise
+    // allocate a mask of the pattern's width for each one up front. Past the cap the mask is
+    // rebuilt in `scratch` from the character's positions, which costs about one step of the scan.
+    const MAX_CACHED_MASKS: usize = 256;
+    let mut masks: std::collections::HashMap<char, Vec<u64>> = std::collections::HashMap::new();
+    let mut scratch = vec![0u64; words];
+    // A set bit i: the text read so far matches the first i pattern items. A `*` matches nothing
+    // as well, so a state at a `*` is also at the item after it.
+    let close = |set: &mut Vec<u64>| {
+        let mut carry = 0u64;
+        for w in 0..words {
+            let s = set[w] & stars[w];
+            let next = (s << 1) | carry;
+            carry = s >> 63;
+            set[w] |= next;
+        }
+    };
+    let mut state = vec![0u64; words];
+    bit(&mut state, 0);
+    close(&mut state);
+    let mut next = vec![0u64; words];
+    for ch in text.chars() {
+        let lit: Option<&[u64]> = match at.get(&ch) {
+            None => None,
+            Some(positions) if masks.contains_key(&ch) || masks.len() < MAX_CACHED_MASKS => {
+                let mask = masks.entry(ch).or_insert_with(|| {
+                    let mut m = vec![0u64; words];
+                    positions.iter().for_each(|&i| bit(&mut m, i));
+                    m
+                });
+                Some(mask)
+            }
+            Some(positions) => {
+                scratch.fill(0);
+                positions.iter().for_each(|&i| bit(&mut scratch, i));
+                Some(&scratch)
+            }
+        };
+        let mut carry = 0u64;
+        let mut alive = 0u64;
+        for w in 0..words {
+            // Items that read `ch` advance; a `*` reads it and stays.
+            let m = state[w] & (any[w] | lit.map_or(0, |l| l[w]));
+            next[w] = (m << 1) | carry | (state[w] & stars[w]);
+            carry = m >> 63;
+            alive |= next[w];
+        }
+        if alive == 0 {
+            return false;
+        }
+        std::mem::swap(&mut state, &mut next);
+        close(&mut state);
     }
-    pi == p.len()
+    state[p.len() / 64] & (1 << (p.len() % 64)) != 0
 }
 
 pub(super) fn find(s: &mut Session, p: &Json) -> Result<Json> {

crates/engine/src/cmd/edit/tests.rs

@@ -302,6 +302,91 @@ fn wildcards() {
     assert!(wildcard("", "*"));
 }
 
+/// The backtracking matcher `wildcard` replaced, as the reference for its results.
+fn wildcard_backtracking(text: &str, pat: &str) -> bool {
+    let t: Vec<char> = text.chars().collect();
+    let mut p: Vec<(char, bool)> = Vec::new();
+    let mut it = pat.chars();
+    while let Some(c) = it.next() {
+        if c == '~' {
+            if let Some(n) = it.next() {
+                p.push((n, true));
+            }
+        } else {
+            p.push((c, false));
+        }
+    }
+    let (mut ti, mut pi) = (0usize, 0usize);
+    let (mut star, mut mark) = (None, 0usize);
+    while ti < t.len() {
+        if pi < p.len() && (p[pi] == ('?', false) || (p[pi].0 == t[ti] && p[pi] != ('*', false))) {
+            ti += 1;
+            pi += 1;
+        } else if pi < p.len() && p[pi] == ('*', false) {
+            star = Some(pi);
+            mark = ti;
+            pi += 1;
+        } else if let Some(sp) = star {
+            pi = sp + 1;
+            mark += 1;
+            ti = mark;
+        } else {
+            return false;
+        }
+    }
+    while pi < p.len() && p[pi] == ('*', false) {
+        pi += 1;
+    }
+    pi == p.len()
+}
+
+#[test]
+fn wildcard_agrees_with_backtracking() {
+    // Every text up to 5 characters over a small alphabet against every pattern up to 4 items.
+    let alphabet = ['a', 'b', '*', '?', '~'];
+    let all = |n: usize, chars: &[char]| -> Vec<String> {
+        let mut out = vec![String::new()];
+        let mut level = vec![String::new()];
+        for _ in 0..n {
+            level = level.iter().flat_map(|s| chars.iter().map(move |c| format!("{s}{c}"))).collect();
+            out.extend(level.iter().cloned());
+        }
+        out
+    };
+    let texts = all(5, &['a', 'b', '*']);
+    for pat in all(4, &alphabet) {
+        for text in &texts {
+            assert_eq!(wildcard(text, &pat), wildcard_backtracking(text, &pat), "{text:?} ~ {pat:?}");
+        }
+    }
+    // Patterns longer than one 64-bit word.
+    let long = format!("*{}*b?", "a".repeat(70));
+    for text in [format!("x{}yb1", "a".repeat(70)), format!("{}b", "a".repeat(71)), "a".repeat(200)] {
+        assert_eq!(wildcard(&text, &long), wildcard_backtracking(&text, &long), "{text}");
+    }
+}
+
+#[test]
+fn wildcard_with_many_distinct_literals_agrees_with_backtracking() {
+    // 600 distinct characters in the pattern: more than the masks kept, so the rest are rebuilt.
+    let chars: Vec<char> = (0x4E00u32..0x4E00 + 600).filter_map(char::from_u32).collect();
+    let word: String = chars.iter().collect();
+    let pat = format!("*{word}?{}", chars[3]);
+    for text in [format!("x{word}yq{}", chars[3]), word.clone(), format!("{word}{word}"), chars.iter().rev().collect::<String>()] {
+        assert_eq!(wildcard(&text, &pat), wildcard_backtracking(&text, &pat), "{} chars", text.chars().count());
+    }
+    assert!(wildcard(&format!("{word}zz{}", chars[3]).replace("zz", "z"), &pat));
+}
+
+#[test]
+fn wildcard_on_long_repetitive_text_is_linear() {
+    // The backtracking matcher compared about text × pattern characters here (10^8).
+    let text = "a".repeat(20_000);
+    let pat = format!("*{}b", "a".repeat(5_000));
+    assert!(!wildcard(&text, &pat));
+    assert!(wildcard(&format!("{text}b"), &pat));
+}
+
 #[test]
 fn tsv() {
     assert_eq!(parse_tsv("a\tb\n1\t2\n"), vec![vec!["a", "b"], vec!["1", "2"]]);

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

@@ -42,15 +42,6 @@ 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()
@@ -66,27 +57,15 @@ 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 !seen.contains(k.as_str()) {
-                            // Stop collecting at the column limit.
-                            if keys.len() >= gridcraft_core::MAX_COLS as usize {
-                                return Err(EngineError::Other("The data has more fields than a sheet has columns.".into()));
-                            }
-                            seen.insert(k.as_str());
+                        if !keys.contains(k) {
                             keys.push(k.clone());
                         }
                     }
                 }
             }
-            if arr.len() as u64 + 1 > u64::from(gridcraft_core::MAX_ROWS) {
-                return Err(EngineError::Other("The data has more rows than a sheet has.".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());
@@ -100,9 +79,6 @@ 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)
@@ -153,35 +129,3 @@ 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/tests.rs

@@ -1312,18 +1312,6 @@ fn scripts_leave_file_access_working_after_they_run() {
     let _ = std::fs::remove_dir_all(&dir);
 }
 
-#[test]
-fn get_data_refuses_json_with_too_many_fields_or_rows_before_building_rows() {
-    let mut s = s();
-    let wide = format!("[{}]", (0..=gridcraft_core::MAX_COLS).map(|i| format!("{{\"k{i}\":1}}")).collect::<Vec<_>>().join(","));
-    let err = s.execute("data.fromTextCsv", json!({"text": wide})).unwrap_err().to_string();
-    assert!(err.contains("more fields"), "{err}");
-    // MAX_ROWS objects plus the header row are one too many.
-    let tall = format!("[{}]", vec!["{}"; gridcraft_core::MAX_ROWS as usize].join(","));
-    let err = s.execute("data.fromTextCsv", json!({"text": tall})).unwrap_err().to_string();
-    assert!(err.contains("more rows"), "{err}");
-}
-
 #[test]
 fn scripts_cannot_reach_outside_the_workbook_even_when_trusted() {
     let dir = std::env::temp_dir().join(format!("gridcraft-script-deny-{}", std::process::id()));