FIX: Floating-point precision bug in FLOOR functions

Fixes #571

- Add EXCEL_PRECISION constant (15 significant digits)
- Fix FLOOR(7.1, 0.1) returning 7.0 instead of 7.1
- Apply to_excel_precision to ratio before floor/ceil operations
- Affects FLOOR, FLOOR.MATH, and FLOOR.PRECISE functions
- Add test_floor with 6 test cases

base/src/constants.rs

@@ -12,6 +12,9 @@ pub(crate) const DEFAULT_WINDOW_WIDTH: i64 = 800;
 pub(crate) const LAST_COLUMN: i32 = 16_384;
 pub(crate) const LAST_ROW: i32 = 1_048_576;
 
+// Excel uses 15 significant digits of precision for all numeric calculations.
+pub(crate) const EXCEL_PRECISION: usize = 15;
+
 // 693_594 is computed as:
 // NaiveDate::from_ymd(1900, 1, 1).num_days_from_ce() - 2
 // The 2 days offset is because of Excel 1900 bug

base/src/functions/mathematical.rs

@@ -1,9 +1,9 @@
 use crate::cast::NumberOrArray;
-use crate::constants::{LAST_COLUMN, LAST_ROW};
+use crate::constants::{EXCEL_PRECISION, LAST_COLUMN, LAST_ROW};
 use crate::expressions::parser::ArrayNode;
 use crate::expressions::types::CellReferenceIndex;
 use crate::functions::math_util::{from_roman, to_roman_with_form};
-use crate::number_format::to_precision;
+use crate::number_format::{to_excel_precision, to_precision};
 use crate::single_number_fn;
 use crate::{
     calc_result::CalcResult, expressions::parser::Node, expressions::token::Error, model::Model,
@@ -984,7 +984,9 @@ impl Model {
             };
         }
 
-        let result = f64::floor(value / significance) * significance;
+        // Apply Excel precision to the ratio to handle floating-point rounding errors
+        let ratio = to_excel_precision(value / significance, EXCEL_PRECISION);
+        let result = f64::floor(ratio) * significance;
         CalcResult::Number(result)
     }
 
@@ -1121,10 +1123,14 @@ impl Model {
         }
         let significance = significance.abs();
         if value < 0.0 && mode != 0.0 {
-            let result = f64::ceil(value / significance) * significance;
+            // Apply Excel precision to handle floating-point rounding errors
+            let ratio = to_excel_precision(value / significance, EXCEL_PRECISION);
+            let result = f64::ceil(ratio) * significance;
             CalcResult::Number(result)
         } else {
-            let result = f64::floor(value / significance) * significance;
+            // Apply Excel precision to handle floating-point rounding errors
+            let ratio = to_excel_precision(value / significance, EXCEL_PRECISION);
+            let result = f64::floor(ratio) * significance;
             CalcResult::Number(result)
         }
     }
@@ -1154,7 +1160,9 @@ impl Model {
             return CalcResult::Number(0.0);
         }
 
-        let result = f64::floor(value / significance) * significance;
+        // Apply Excel precision to handle floating-point rounding errors
+        let ratio = to_excel_precision(value / significance, EXCEL_PRECISION);
+        let result = f64::floor(ratio) * significance;
         CalcResult::Number(result)
     }
 

base/src/test/mod.rs

@@ -65,6 +65,7 @@ mod test_escape_quotes;
 mod test_even_odd;
 mod test_exp_sign;
 mod test_extend;
+mod test_floor;
 mod test_fn_datevalue_timevalue;
 mod test_fn_fv;
 mod test_fn_round;

base/src/test/test_floor.rs

@@ -0,0 +1,123 @@
+#![allow(clippy::unwrap_used)]
+
+use crate::test::util::new_empty_model;
+
+#[test]
+fn test_floor_floating_point_precision() {
+    // This test specifically checks the floating-point precision bug fix
+    // Bug: FLOOR(7.1, 0.1) was returning 7.0 instead of 7.1
+    let mut model = new_empty_model();
+
+    // FLOOR tests
+    model._set("C5", "=FLOOR(7.1, 0.1)");
+    model._set("H7", "=FLOOR(-7.1, -0.1)");
+
+    // FLOOR.PRECISE tests
+    model._set("C53", "=FLOOR.PRECISE(7.1, 0.1)");
+    model._set("H53", "=FLOOR.PRECISE(7.1, -0.1)");
+
+    // FLOOR.MATH tests
+    model._set("C101", "=FLOOR.MATH(7.1, 0.1)");
+    model._set("H101", "=FLOOR.MATH(7.1, -0.1)");
+
+    model.evaluate();
+
+    // All should return 7.1
+    assert_eq!(model._get_text("C5"), *"7.1");
+    assert_eq!(model._get_text("H7"), *"-7.1");
+    assert_eq!(model._get_text("C53"), *"7.1");
+    assert_eq!(model._get_text("H53"), *"7.1");
+    assert_eq!(model._get_text("C101"), *"7.1");
+    assert_eq!(model._get_text("H101"), *"7.1");
+}
+
+#[test]
+fn test_floor_additional_precision_cases() {
+    let mut model = new_empty_model();
+    model._set("A1", "=FLOOR(7.9, 0.1)");
+    model._set("A2", "=FLOOR(2.6, 0.5)");
+    model._set("A3", "=FLOOR(0.3, 0.1)"); // 0.1 + 0.2 type scenario
+
+    model.evaluate();
+
+    assert_eq!(model._get_text("A1"), *"7.9");
+    assert_eq!(model._get_text("A2"), *"2.5");
+    assert_eq!(model._get_text("A3"), *"0.3");
+}
+
+#[test]
+fn test_floor_basic_cases() {
+    let mut model = new_empty_model();
+    model._set("A1", "=FLOOR(3.7, 2)");
+    model._set("A2", "=FLOOR(3.2, 1)");
+    model._set("A3", "=FLOOR(10, 3)");
+    model._set("A4", "=FLOOR(7, 2)");
+
+    model.evaluate();
+
+    assert_eq!(model._get_text("A1"), *"2");
+    assert_eq!(model._get_text("A2"), *"3");
+    assert_eq!(model._get_text("A3"), *"9");
+    assert_eq!(model._get_text("A4"), *"6");
+}
+
+#[test]
+fn test_floor_negative_numbers() {
+    let mut model = new_empty_model();
+    // Both negative: rounds toward zero
+    model._set("A1", "=FLOOR(-2.5, -2)");
+    model._set("A2", "=FLOOR(-11, -3)");
+
+    // Negative number, positive significance: rounds away from zero
+    model._set("A3", "=FLOOR(-11, 3)");
+    model._set("A4", "=FLOOR(-2.5, 2)");
+
+    model.evaluate();
+
+    assert_eq!(model._get_text("A1"), *"-2");
+    assert_eq!(model._get_text("A2"), *"-9");
+    assert_eq!(model._get_text("A3"), *"-12");
+    assert_eq!(model._get_text("A4"), *"-4");
+}
+
+#[test]
+fn test_floor_error_cases() {
+    let mut model = new_empty_model();
+    // Positive number with negative significance should error
+    model._set("A1", "=FLOOR(2.5, -2)");
+    model._set("A2", "=FLOOR(10, -3)");
+
+    // Division by zero
+    model._set("A3", "=FLOOR(5, 0)");
+
+    // Wrong number of arguments
+    model._set("A4", "=FLOOR(5)");
+    model._set("A5", "=FLOOR(5, 1, 1)");
+
+    model.evaluate();
+
+    assert_eq!(model._get_text("A1"), *"#NUM!");
+    assert_eq!(model._get_text("A2"), *"#NUM!");
+    assert_eq!(model._get_text("A3"), *"#DIV/0!");
+    assert_eq!(model._get_text("A4"), *"#ERROR!");
+    assert_eq!(model._get_text("A5"), *"#ERROR!");
+}
+
+#[test]
+fn test_floor_edge_cases() {
+    let mut model = new_empty_model();
+    // Zero value
+    model._set("A1", "=FLOOR(0, 5)");
+    model._set("A2", "=FLOOR(0, 0)");
+
+    // Exact multiples
+    model._set("A3", "=FLOOR(10, 5)");
+    model._set("A4", "=FLOOR(9, 3)");
+
+    model.evaluate();
+
+    assert_eq!(model._get_text("A1"), *"0");
+    assert_eq!(model._get_text("A2"), *"0");
+    assert_eq!(model._get_text("A3"), *"10");
+    assert_eq!(model._get_text("A4"), *"9");
+}