ANother attempt to use moveToCodePointOffset for cursor routing

source/NVDAObjects/UIA/wordDocument.py

@@ -270,6 +270,18 @@ def _getTextFromUIARange(self, textRange):
 			t = t.replace(END_OF_ROW_MARK, "")
 		return t
 
+	def _getTextForCodepointMovement(self) -> str:
+		"""
+		#8576, #10960: In Word, list bullets are exposed in text but are ignored when moving by character.
+		Therefore in `getTextWithFields`, the bullets are stripped from the text and exposed in the `line-prefix` field.
+		To stay compatible with this, we can't simply use the `text` property,
+		as it can potentially contain bullets that should be stripped.
+		"""
+		t = super()._getTextForCodepointMovement()
+		if not t:
+			return t
+		return "".join(f for f in self.getTextWithFields(formatConfig=dict()) if isinstance(f, str))
+
 	def _isEndOfRow(self):
 		"""Is this textInfo positioned on an end-of-row mark?"""
 		info = self.copy()
@@ -648,7 +660,7 @@ def _caretMoveBySentenceHelper(self, gesture, direction):
 		description=_(
 			# Translators: a description for a script that reports the comment at the caret.
 			"Reports the text of the comment where the system caret is located."
-			" If pressed twice, presents the information in a browsable message"
+			" If pressed twice, presents the information in a browsable message",
 		),
 		category=SCRCAT_SYSTEMCARET,
 		speakOnDemand=True,

source/braille.py

@@ -1426,14 +1426,39 @@ def update(self):
 			self._brailleInputIndStart = None
 
 	def getTextInfoForBraillePos(self, braillePos: int) -> textInfos.TextInfo:
-		"""Fetches a collapsed TextInfo at the specified braille position in the region."""
+		"""Fetches a collapsed TextInfo at the specified braille position in the region.
+		:param braillePos: The braille position.
+			If no textInfo could be found at braillePos,
+			try to find one at braillePos - 1 until a position has been found.
+		"""
 		pos = self._rawToContentPos[self.brailleToRawPos[braillePos]]
 		# pos is relative to the start of the reading unit.
-		# Therefore, get the start of the reading unit...
+		maxIterations = 10
+		startTime = time.time()
+		for i, curPos in enumerate(range(pos, max(-1, pos - maxIterations), -1)):
+			if curPos == 0:
+				# Not necessary to find offset.
+				break
+			# Move curPos code points from the start.
+			# Note that, as liblouis uses 32 bit encoding internally,
+			# it is really safe to assume that one code point offset is equal to one character within liblouis.
+			# If an attempt fails, we try to move to the previous character
+			try:
+				return self._readingInfo.moveToCodepointOffset(curPos)
+			except RuntimeError:
+				msg = f"Error in moveToCodepointOffset in iteration {i + 1} (position {curPos}"
+				if i + 1 >= maxIterations or (exceeded := time.time() - startTime > 0.5):
+					logFunc = log.exception
+					curPos = pos
+					if exceeded:
+						msg += ", exceeded time limit of 0.5 seconds"
+				else:
+					logFunc = log.debug
+				logFunc(msg)
 		dest = self._readingInfo.copy()
 		dest.collapse()
-		# and move pos characters from there.
-		dest.move(textInfos.UNIT_CHARACTER, pos)
+		if curPos > 0:
+			dest.move(textInfos.UNIT_CHARACTER, curPos)
 		return dest
 
 	def routeTo(self, braillePos: int):

source/textInfos/__init__.py

@@ -711,6 +711,10 @@ def getMathMl(self, field):
 		"""
 		raise NotImplementedError
 
+	def _getTextForCodepointMovement(self) -> str:
+		"""Gets the text as used in moveToCodepointOffset."""
+		return self.text
+
 	def moveToCodepointOffset(
 		self,
 		codepointOffset: int,
@@ -803,7 +807,7 @@ def moveToCodepointOffset(
 			we reduce the count of characters in order to make sure
 			the algorithm makes some progress on each iteration.
 		"""
-		text = self.text
+		text = self._getTextForCodepointMovement()
 		if codepointOffset < 0 or codepointOffset > len(text):
 			raise ValueError
 		if codepointOffset == 0 or codepointOffset == len(text):
@@ -845,7 +849,7 @@ def moveToCodepointOffset(
 					moveCharacters = codepointOffsetLeft
 				code = tmpInfo.move(UNIT_CHARACTER, moveCharacters, endPoint="end")
 				lastMove = moveCharacters
-				tmpText = tmpInfo.text
+				tmpText = tmpInfo._getTextForCodepointMovement()
 				actualCodepointOffset = len(tmpText)
 				if not text.startswith(tmpText):
 					raise RuntimeError(
@@ -865,7 +869,7 @@ def moveToCodepointOffset(
 					moveCharacters = -codepointOffsetRight
 				code = tmpInfo.move(UNIT_CHARACTER, moveCharacters, endPoint="start")
 				lastMove = moveCharacters
-				tmpText = tmpInfo.text
+				tmpText = tmpInfo._getTextForCodepointMovement()
 				actualCodepointOffset = totalCodepointOffset - len(tmpText)
 				if not text.endswith(tmpText):
 					raise RuntimeError(

tests/unit/test_braille/test_routing.py

@@ -1,16 +1,16 @@
 # A part of NonVisual Desktop Access (NVDA)
 # This file is covered by the GNU General Public License.
 # See the file COPYING for more details.
-# Copyright (C) 2023 NV Access Limited, Leonard de Ruijter
+# Copyright (C) 2023-2024 NV Access Limited, Leonard de Ruijter
 
-"""Unit tests for the move system caret when routing review cursor braille setting."""
+"""Unit tests for braille cursor routing."""
 
 import config
 import braille
 import textInfos
 import api
 import controlTypes
-from ..textProvider import CursorManager
+from ..textProvider import CursorManager, BasicTextProvider
 import unittest
 import time
 from config.featureFlagEnums import ReviewRoutingMovesSystemCaretFlag
@@ -147,3 +147,58 @@ def test_moveCaret_always_instantActivate(self):
 		self.assertGreaterEqual(self.cm.lastActivateTime, curTime)
 		caret = self.cm.makeTextInfo(textInfos.POSITION_CARET)
 		self.assertEquals(caret, review)
+
+
+class TestTextInfoRegionRouting(unittest.TestCase):
+	"""A test for TextInfoRegion.getTextInfoForBraillePos, which is used in braille cursor routing.
+	This test ensures that braille routes to the expected character when dealing with emoji
+	or other composites.
+	These glyphs are threated as one character by uniscribe, however they span multiple characters
+	on a braille display.
+	Note that due to the nature of this test, it relies on uniscribe to be available.
+	"""
+
+	def test_routeToEmoji(self):
+		testText = "⚠️test"
+		obj = BasicTextProvider(text=testText)
+		ti = obj.makeTextInfo(textInfos.POSITION_CARET)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[:2])
+		ti.collapse(end=True)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[2])
+		region = braille.TextInfoRegion(obj)
+		region.update()
+		index = 3  # Position of e
+		pos = region.rawToBraillePos[index]
+		region.routeTo(pos)
+		ti = obj.makeTextInfo(textInfos.POSITION_CARET)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[index])
+
+	def test_routeToComposite(self):
+		testText = "רבְּר"
+		obj = BasicTextProvider(text=testText)
+		ti = obj.makeTextInfo(textInfos.POSITION_CARET)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[0])
+		ti.collapse(end=True)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[1:4])
+		ti.collapse(end=True)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[4])
+		region = braille.TextInfoRegion(obj)
+		region.update()
+		index = 1  # Position of ב
+		pos = region.rawToBraillePos[index]
+		region.routeTo(pos)
+		ti = obj.makeTextInfo(textInfos.POSITION_CARET)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[1:4])
+		index = 3  # Position of ּ (\u5bc)
+		pos = region.rawToBraillePos[index]
+		region.routeTo(pos)
+		ti = obj.makeTextInfo(textInfos.POSITION_CARET)
+		ti.expand(textInfos.UNIT_CHARACTER)
+		self.assertEqual(ti.text, testText[1:4])

tests/unit/textProvider.py

@@ -18,8 +18,6 @@
 
 
 class BasicTextInfo(NVDAObjectTextInfo):
-	# NVDAHelper is not initialized, so we can't use Uniscribe.
-	useUniscribe = False
 	# Most of our code use UTF-16 as internal encoding.
 	# Mimic this behavior, so we can also implicitly test textUtils module code
 	encoding = textUtils.WCHAR_ENCODING

user_docs/en/changes.md

@@ -19,6 +19,7 @@
   * UIA in Windows Terminal. (#16873, @codeofdusk)
 * When accessing Microsoft Word without UI Automation, NVDA no longer outputs garbage characters in braille in table headers defined with the set row and column header commands. (#7212)
 * The Seika Notetaker driver now correctly generates braille input for space, backspace and dots with space/backspace gestures. (#16642, @school510587)
+* Braille cursor routing is now much more reliable when a line contains one or more Unicode variation selectors or decomposed characters. (#10960, @mltony, @LeonarddeR)
 * In on-demand speech mode, NVDA does not talk anymore when a message is opened in Outlook, when a new page is loaded in a browser or during the slideshow in PowerPoint. (#16825, @CyrilleB79)