[FIX] charts: non-invertible matrix returns NaN as predicted data

Task Description

When trying to make a trending line, almost all model use matrix
computation to get the predicted data, with some matrix
multiplications and inversions. When the matrixes aren't invertible
the user get a traceback, which is not fine.
This PR aims to fix these issues by returning an array of NaN when
there is an error in the computation of the predicted dataset.

Related Task

Task: 4328743

Author: anhe-odoo

src/helpers/figures/charts/chart_common.ts

@@ -535,41 +535,45 @@ export function interpolateData(
   const labelRange = labelMax - labelMin;
   const normalizedLabels = labels.map((v) => (v - labelMin) / labelRange);
   const normalizedNewLabels = newLabels.map((v) => (v - labelMin) / labelRange);
-  switch (config.type) {
-    case "polynomial": {
-      const order = config.order ?? 2;
-      if (order === 1) {
-        return predictLinearValues([values], [normalizedLabels], [normalizedNewLabels], true)[0];
+  try {
+    switch (config.type) {
+      case "polynomial": {
+        const order = config.order ?? 2;
+        if (order === 1) {
+          return predictLinearValues([values], [normalizedLabels], [normalizedNewLabels], true)[0];
+        }
+        const coeffs = polynomialRegression(values, normalizedLabels, order, true).flat();
+        return normalizedNewLabels.map((v) => evaluatePolynomial(coeffs, v, order));
       }
-      const coeffs = polynomialRegression(values, normalizedLabels, order, true).flat();
-      return normalizedNewLabels.map((v) => evaluatePolynomial(coeffs, v, order));
-    }
-    case "exponential": {
-      const positiveLogValues: number[] = [];
-      const filteredLabels: number[] = [];
-      for (let i = 0; i < values.length; i++) {
-        if (values[i] > 0) {
-          positiveLogValues.push(Math.log(values[i]));
-          filteredLabels.push(normalizedLabels[i]);
+      case "exponential": {
+        const positiveLogValues: number[] = [];
+        const filteredLabels: number[] = [];
+        for (let i = 0; i < values.length; i++) {
+          if (values[i] > 0) {
+            positiveLogValues.push(Math.log(values[i]));
+            filteredLabels.push(normalizedLabels[i]);
+          }
         }
+        if (!filteredLabels.length) {
+          return Array.from({ length: newLabels.length }, () => NaN);
+        }
+        return expM(
+          predictLinearValues([positiveLogValues], [filteredLabels], [normalizedNewLabels], true)
+        )[0];
       }
-      if (!filteredLabels.length) {
-        return [];
+      case "logarithmic": {
+        return predictLinearValues(
+          [values],
+          logM([normalizedLabels]),
+          logM([normalizedNewLabels]),
+          true
+        )[0];
       }
-      return expM(
-        predictLinearValues([positiveLogValues], [filteredLabels], [normalizedNewLabels], true)
-      )[0];
-    }
-    case "logarithmic": {
-      return predictLinearValues(
-        [values],
-        logM([normalizedLabels]),
-        logM([normalizedNewLabels]),
-        true
-      )[0];
+      default:
+        return [];
     }
-    default:
-      return [];
+  } catch (e) {
+    return Array.from({ length: newLabels.length }, () => NaN);
   }
 }
 

tests/figures/chart/chart_plugin.test.ts

@@ -3188,6 +3188,20 @@ describe("trending line", () => {
     }
   });
 
+  test("non-invertible matrix doesn't throw error", () => {
+    // prettier-ignore
+    setGrid(model, {
+      A1: "label",
+      A2: "0",
+      A3: "1",
+    });
+    updateChart(model, "1", {
+      dataSets: [{ dataRange: "A1:A3", trend: { display: true, type: "polynomial", order: 2 } }],
+    });
+    const runtime = model.getters.getChartRuntime("1") as LineChartRuntime;
+    expect(runtime.chartJsConfig.data.datasets[1].data.every((x) => isNaN(Number(x)))).toBeTruthy();
+  });
+
   test("trend line works with real date values as labels", () => {
     setGrid(model, {
       B1: "1",