fix(web): transcription-construction from contexts with selections

common/web/keyboard-processor/src/index.ts

@@ -39,6 +39,7 @@ export { default as KeyMapping } from "./text/keyMapping.js";
 export { default as OutputTarget } from "./text/outputTarget.js";
 export * from "./text/outputTarget.js";
 export { default as RuleBehavior } from "./text/ruleBehavior.js";
+export * from "./text/stringDivergence.js";
 export * from "./text/systemStores.js";
 
 export * from "@keymanapp/web-utils";

common/web/keyboard-processor/src/text/outputTarget.ts

@@ -1,6 +1,7 @@
 ///<reference types="@keymanapp/models-types" />
 
 import { extendString } from "@keymanapp/web-utils";
+import { searchStringDivergence } from "./stringDivergence.js";
 
 extendString();
 
@@ -120,85 +121,24 @@ export default abstract class OutputTarget {
    * @param from An output target (preferably a Mock) representing the prior state of the input/output system.
    */
   buildTransformFrom(original: OutputTarget): Transform {
-    let to = this.getText();
-    let from = original.getText();
+    const toLeft = this.getTextBeforeCaret();
+    const fromLeft = original.getTextBeforeCaret();
 
-    let fromCaret = original.getDeadkeyCaret();
-    let toCaret = this.getDeadkeyCaret();
+    const leftDivergenceIndex = searchStringDivergence(fromLeft, toLeft, false)[0];
+    const deletedLeft = fromLeft.substring(leftDivergenceIndex)._kmwLength();
+    // No need for our specialized variant here.
+    const insertedText = toLeft.substring(leftDivergenceIndex);
 
-    // Step 1:  Determine the number of left-deletions.
-    let maxSMPLeftMatch = fromCaret < toCaret ? fromCaret : toCaret;
+    const toRight = this.getTextAfterCaret();
+    const fromRight = original.getTextAfterCaret();
+    const rightDivergence1 = searchStringDivergence(fromRight, toRight, true)[0];
 
-    // We need the corresponding non-SMP caret location in order to binary-search efficiently.
-    // (Examining code units is much more computationally efficient.)
-    let maxLeftMatch = to._kmwCodePointToCodeUnit(maxSMPLeftMatch);
+    // Right insertions aren't supported, but right deletions will matter in some scenarios.
+    // In particular, once we allow right-deletion for pred-text suggestions applied with the
+    // caret mid-word..
+    const deletedRight = fromRight.substring(0, rightDivergence1 + 1)._kmwLength();
 
-    // 1.1:  use a non-SMP-aware binary search to determine the divergence point.
-    let start = 0;
-    let end = maxLeftMatch;  // the index AFTER the last possible matching char.
-
-    // This search is O(maxLeftMatch).  1/2 + 1/4 + 1/8 + ... converges to = 1.
-    while(start < end) {
-      let mid = Math.floor((end+start+1) / 2); // round up (compare more)
-      let fromLeft = from.substr(start, mid-start);
-      let toLeft   = to.substr(start, mid-start);
-
-      if(fromLeft == toLeft) {
-        start = mid;
-      } else {
-        end = mid - 1;
-      }
-    }
-
-    // At the loop's end:  `end` now holds the non-SMP-aware divergence point.
-    // The 'caret' is after the last matching code unit.
-
-    // 1.2:  detect a possible surrogate-pair split scenario, correcting for it
-    //       (by moving the split before the high-surrogate) if detected.
-
-    // If the split location is precisely on either end of the context, we can't
-    // have split a surrogate pair.
-    if(end > 0 && end < maxLeftMatch) {
-      let potentialHigh    = from.charCodeAt(end-1);
-      let potentialFromLow = from.charCodeAt(end);
-      let potentialToLow   = to.charCodeAt(end);
-
-      // if potentialHigh is a possible high surrogate...
-      if(potentialHigh >= 0xD800 && potentialHigh <= 0xDBFF) {
-        // and at least one potential 'low' is a possible low surrogate...
-        let flag = potentialFromLow >= 0xDC00 && potentialFromLow <= 0xDFFF;
-        flag = flag || (potentialToLow >= 0XDC00 && potentialToLow <= 0xDFFF);
-
-        // Correct the split location, moving it 'before' the high surrogate.
-        if(flag) {
-          end = end - 1;
-        }
-      }
-    }
-
-    // 1.3:  take substring from start to the split point; determine SMP-aware length.
-    //       This yields the SMP-aware divergence index, which gives the number of left-deletes.
-    let newCaret = from._kmwCodeUnitToCodePoint(end);
-    let deletedLeft = fromCaret - newCaret;
-
-    // Step 2:  Determine the other properties.
-    // Since the 'after' OutputTarget's caret indicates the end of any inserted text, we
-    // can easily calculate the rest.
-    let insertedLength = toCaret - newCaret;
-    let delta = to._kmwSubstr(newCaret, insertedLength);
-
-    let undeletedRight = to._kmwLength() - toCaret;
-    let originalRight = from._kmwLength() - fromCaret;
-    let deletedRight = originalRight - undeletedRight;
-
-    // May occur when reverting a suggestion that had been applied mid-word.
-    if(deletedRight < 0) {
-      // Restores deleteRight characters.
-      delta = delta + to._kmwSubstr(toCaret, -deletedRight);
-      deletedRight = 0;
-    }
-
-    return new TextTransform(delta, deletedLeft, deletedRight);
+    return new TextTransform(insertedText, deletedLeft, deletedRight);
   }
 
   buildTranscriptionFrom(original: OutputTarget, keyEvent: KeyEvent, readonly: boolean, alternates?: Alternate[]): Transcription {

common/web/keyboard-processor/src/text/stringDivergence.ts

@@ -0,0 +1,68 @@
+/**
+ * Returns the index for the code point divergence point in code unit coordinates.
+ * @param str1
+ * @param str2
+ * @param commonRight If false or undefined, asserts a common prefix to the strings.  If true, asserts a common suffix.
+ * @returns The code unit indices within each string for the start of the code point not common to both.
+ */
+export function searchStringDivergence(str1: string, str2: string, commonRight?: boolean): [number, number] {
+  let maxInterval = Math.min(str1.length, str2.length) - 1;
+  const commonLeft = !commonRight;
+
+  let index: number;
+  let end: number;
+
+  /**
+   * Index shift per loop iteration.
+   */
+  let inc: number;
+  /**
+   * Difference in index for comparison between strings.
+   * Mostly matters when assuming a common right-hand side.
+   */
+  let offset: number;
+
+  if(commonLeft) {
+    index = 0;
+    end = maxInterval;
+    inc = 1;
+    offset = 0;
+  } else {
+    index = str1.length - 1;
+    end = index - maxInterval;
+    inc = -1;
+    offset = str2.length - str1.length;
+  }
+
+  for(; commonLeft ? index <= end: index >= end; index += inc) {
+    if(str1.charAt(index) != str2.charAt(index + offset)) {
+      break;
+    }
+  }
+
+  // `index` corresponds to the first char that is different _in the direction indicated by inc_.
+
+  // if commonLeft, high surrogate; if commonRight, low surrogate.
+  const commonPotentialSurrogate = str1.charCodeAt(index - inc);
+  // Opposite surrogate type from the previous variable.
+  const divergentChar1 = str1.charCodeAt(index);
+  const divergentChar2 = str2.charCodeAt(index + offset);
+
+  const isHigh = (charCode: number) => charCode >= 0xD800 && charCode <= 0xDBFF;
+  const isLow =  (charCode: number) => charCode >= 0xDC00 && charCode <= 0xDFFF;
+  const commonChecker = commonLeft ? isHigh : isLow;
+  const divergentChecker = commonLeft ? isLow : isHigh;
+
+  // If the last common char qualifies as a direction-appropriate SMP surrogate...
+  if(commonChecker(commonPotentialSurrogate)) {
+    // And one of the two divergent chars is a qualifying match - a surrogate
+    // of the opposite type...
+    if(divergentChecker(divergentChar1) || divergentChecker(divergentChar2)) {
+      // Our current index would split a surrogate pair; decrement the index to
+      // preserve the pair.
+      return [index - inc, index - inc + offset];
+    }
+  }
+
+  return [index, index + offset];
+}

common/web/keyboard-processor/tests/node/transcriptions.js

@@ -1,20 +1,129 @@
 import { assert } from 'chai';
 
-import { Mock } from '@keymanapp/keyboard-processor';
+import { Mock, searchStringDivergence } from '@keymanapp/keyboard-processor';
 import { extendString } from '@keymanapp/web-utils';
 
 extendString();  // Ensure KMW's string-extension functionality is available.
 
 String.kmwEnableSupplementaryPlane(false);
 
+const toSupplementaryPairString = function(code){
+  var H = Math.floor((code - 0x10000) / 0x400) + 0xD800;
+  var L = (code - 0x10000) % 0x400 + 0xDC00;
+
+  return String.fromCharCode(H, L);
+}
+// A unicode-coding like alias for use in constructing SMP strings.
+const u = toSupplementaryPairString;
+
+/**
+ * Returns the "Mathematical Sans-Serif Small" SMP encoding for
+ * a passed-in lowercase char between 'a' and 'z', inclusive.
+ * @param {*} char
+ * @returns
+ */
+const ss = (char) => {
+  const charCodeOffset = char.charCodeAt(0) - 'a'.charCodeAt(0);
+  return u(0x1d5ba + charCodeOffset);
+}
+
+describe("String divergence calculations", function() {
+  describe("Common prefix", () => {
+    it("BMP text", () => {
+      const result1 = searchStringDivergence("apple", "applause", false);
+      assert.deepEqual(result1, [4, 4]);
+
+      const result2 = searchStringDivergence("applesauce", "applause", false);
+      assert.deepEqual(result2, [4, 4]);
+
+      const result3 = searchStringDivergence("applesauce", "applesauce", false);
+      assert.deepEqual(result3, [10, 10]);
+    });
+
+    it("SMP text", () => {
+      const smp_ify = (str) => str.split('').map(ss).join('');
+
+      const result1 = searchStringDivergence(
+        smp_ify('apple'),
+        smp_ify('applause'),
+        false
+      );
+
+      // 2 per SMP char; is in code-unit... units.
+      // Will avoid splitting code points, though.
+      assert.deepEqual(result1, [8, 8]);
+
+      const result2 = searchStringDivergence(
+        smp_ify('applesauce'),
+        smp_ify('applause'),
+        false
+      );
+
+      assert.deepEqual(result2, [8, 8]);
+
+      const result3 = searchStringDivergence(
+        smp_ify('applesauce'),
+        smp_ify('applesauce'),
+        false
+      );
+
+      assert.deepEqual(result3, [20, 20]);
+    });
+  });
+
+  describe("Common suffix", () => {
+    it("BMP text", () => {
+      //    att|endance
+      // transc|endance
+      const result1 = searchStringDivergence("attendance", "transcendance", true);
+      assert.deepEqual(result1, [2, 5]);
+
+      // transcend|ance
+      //  happenst|ance
+      const result2 = searchStringDivergence("transcendance", "happenstance", true);
+      assert.deepEqual(result2, [8, 7]);
+
+      // And if the two are equal...
+      const result3 = searchStringDivergence("post-caret text", "post-caret text", true);
+      assert.deepEqual(result3, [-1, -1]);
+    });
+
+    it("SMP text", () => {
+      const smp_ify = (str) => str.split('').map(ss).join('');
+
+      //   att|endance
+      // trans|endance
+      const result1 = searchStringDivergence(
+        smp_ify("attendance"),
+        smp_ify("transcendance"),
+        true
+      );
+
+      // 2 per SMP char; is in code-unit... units.
+      // Will avoid splitting code points; is odd b/c we get the index of the LAST char of the pair.
+      assert.deepEqual(result1, [5, 11]);
+
+      // transcend|ance
+      //  happenst|ance
+      const result2 = searchStringDivergence(
+        smp_ify("transcendance"),
+        smp_ify("happenstance"),
+        true
+      );
+      assert.deepEqual(result2, [17, 15]);
+
+      // And if the two are equal...
+      const result3 = searchStringDivergence(
+        smp_ify("post-caret text"),
+        smp_ify("post-caret text"),
+        true
+      );
+      assert.deepEqual(result3, [-1, -1]);
+    });
+  })
+});
+
 describe("Transcriptions and Transforms", function() {
-  var toSupplementaryPairString = function(code){
-    var H = Math.floor((code - 0x10000) / 0x400) + 0xD800;
-    var L = (code - 0x10000) % 0x400 + 0xDC00;
-
-    return String.fromCharCode(H, L);
-  }
-
   // Built in-line via function.  Looks functionally equivalent to "apple", but with SMP characters.
   let u = toSupplementaryPairString;
   let smpApple = u(0x1d5ba)+u(0x1d5c9)+u(0x1d5c9)+u(0x1d5c5)+u(0x1d5be);
@@ -298,7 +407,7 @@ but not himself.`;  // Sheev Palpatine, in the Star Wars prequels.
       }
     });
 
-    it.only('from targets with existing selection', () => {
+    it('from targets with existing selection', () => {
       //                              |            |
       const target = new Mock("testing testing one two three");
       target.setSelection(8, 20)
@@ -313,7 +422,7 @@ but not himself.`;  // Sheev Palpatine, in the Star Wars prequels.
       });
     });
 
-    it.only('to targets with existing selection', () => {
+    it('to targets with existing selection', () => {
       //                              |            |
       const target = new Mock("testing testing one two three");
       target.setSelection(8, 20)

web/src/app/browser/src/keymanEngine.ts

@@ -205,9 +205,6 @@ export default class KeymanEngine extends KeymanEngineBase<BrowserConfiguration,
     // Automatically performs related handler setup & maintains references
     // needed for related cleanup / shutdown.
     this.pageIntegration = new PageIntegrationHandlers(window, this);
-
-    // Initialize supplementary plane string extensions
-    String.kmwEnableSupplementaryPlane(true);
     this.config.finalizeInit();
 
     if(this.ui) {

web/src/engine/main/src/keymanEngine.ts

@@ -196,6 +196,9 @@ export default class KeymanEngine<
 
     config.initialize(optionSpec);
 
+    // Initialize supplementary plane string extensions
+    String.kmwEnableSupplementaryPlane(true);
+
     // Since we're not sandboxing keyboard loads yet, we just use `window` as the jsGlobal object.
     // All components initialized below require a properly-configured `config.paths` or similar.
     const keyboardLoader = new KeyboardLoader(this.interface, config.applyCacheBusting);