Return "blank" if text empty due to processing (#13483)

closes #13431 

After processing a text for dictionary substitutions and symbol level, the result might be an empty string, which leads to nothing being spoken at all (not even "blank"). Something should always be spoken when reading text, since nothing being spoken gives the appearance of NVDA freezing.

Description of how this pull request fixes the issue:
When a speech sequence is submitted to speech.speak, if all strings of the sequence become empty due to symbol level and dictionary substitution processing, then the sequence is considered blank and "blank" is appended to it.

The function now also takes an additional argument, suppressBlanks, which is false by default. This parameter allows callers to suppress the "blank" being appended to the sequence, even if the conditions are met. The main intention is to suppress this behavior during say all. Usages of speech.speak have been updated to specify this parameter when appropriate.

source/speech/__init__.py

@@ -4,6 +4,7 @@
 # Copyright (C) 2006-2021 NV Access Limited, Peter Vágner, Aleksey Sadovoy, Babbage B.V., Bill Dengler,
 # Julien Cochuyt
 
+import functools
 from .speech import (
 	_extendSpeechSequence_addMathForTextInfo,
 	_getSpellingSpeechAddCharMode,
@@ -155,7 +156,9 @@ def initialize():
 	synthDriverHandler.setSynth(config.conf["speech"]["synth"])
 	speechInitialize()
 	sayAllInitialize(
-		speak,
+		# override speak func to always have suppressBlanks=True,
+		# since blanks should never be spoken in say all
+		functools.partial(speak, suppressBlanks=True),
 		speakObject,
 		getTextInfoSpeech,
 		SpeakTextInfoState,

source/speech/speech.py

@@ -238,7 +238,7 @@ def speakSpelling(
 		locale=locale,
 		useCharacterDescriptions=useCharacterDescriptions
 	))
-	speak(seq, priority=priority)
+	speak(seq, priority=priority, suppressBlanks=True)
 
 
 def _getSpellingSpeechAddCharMode(
@@ -878,12 +878,14 @@ def getIndentationSpeech(indentation: str, formatConfig: Dict[str, bool]) -> Spe
 def speak(  # noqa: C901
 		speechSequence: SpeechSequence,
 		symbolLevel: Optional[int] = None,
-		priority: Spri = Spri.NORMAL
+		priority: Spri = Spri.NORMAL,
+		suppressBlanks: bool = False
 ):
 	"""Speaks a sequence of text and speech commands
 	@param speechSequence: the sequence of text and L{SpeechCommand} objects to speak
 	@param symbolLevel: The symbol verbosity level; C{None} (default) to use the user's configuration.
 	@param priority: The speech priority.
+	@param suppressBlanks: Whether to not append "blank" to the speech even if considered blank
 	"""
 	logBadSequenceTypes(speechSequence)
 	# in case priority was explicitly passed in as None, set to default.
@@ -943,9 +945,22 @@ def speak(  # noqa: C901
 		if autoLanguageSwitching and isinstance(item,LangChangeCommand):
 			curLanguage=item.lang
 		if isinstance(item,str):
-			speechSequence[index]=processText(curLanguage,item,symbolLevel)
-			if not inCharacterMode:
-				speechSequence[index]+=CHUNK_SEPARATOR
+			text = processText(curLanguage, item, symbolLevel)
+			if not inCharacterMode and text:
+				text += CHUNK_SEPARATOR
+			speechSequence[index] = text
+	# speech sequence should be considered blank if:
+	# 1. it contains strings
+	# 2. all strings are blank after processing
+	if (
+		not suppressBlanks
+		and any(isinstance(i, str) for i in speechSequence)
+		# for checking if blank, just check if empty instead of isBlank(),
+		# since whitespace has been stripped during processing
+		and all(not s for s in speechSequence if isinstance(s, str))
+	):
+		# Translators: This is spoken when the speech sequence is considered blank.
+		speechSequence.append(_("blank"))
 	_manager.speak(speechSequence, priority)
 
 
@@ -964,7 +979,7 @@ def speakPreselectedText(
 	"""
 	seq = getPreselectedTextSpeech(text)
 	if seq:
-		speak(seq, symbolLevel=None, priority=priority)
+		speak(seq, symbolLevel=None, priority=priority, suppressBlanks=True)
 
 
 def getPreselectedTextSpeech(
@@ -1013,7 +1028,7 @@ def speakSelectionMessage(
 ):
 	seq = _getSelectionMessageSpeech(message, text)
 	if seq:
-		speak(seq, symbolLevel=None, priority=priority)
+		speak(seq, symbolLevel=None, priority=priority, suppressBlanks=True)
 
 
 def _getSelectionMessageSpeech(
@@ -1230,7 +1245,7 @@ def speakTextInfo(
 
 	speechGen = GeneratorWithReturn(speechGen)
 	for seq in speechGen:
-		speak(seq, priority=priority)
+		speak(seq, priority=priority, suppressBlanks=suppressBlanks)
 	return speechGen.returnValue
 
 

tests/system/robot/chromeTests.py

@@ -205,21 +205,16 @@ def test_mark_aria_details_role():
 		# The role "ROLE_LIST_ITEM" is used instead
 		"has details",
 		"doc-endnote,",
-		"",  # space between spans
 		"has foot note",
 		"doc-footnote,",
-		"",  # space between spans
 		"has comment",
 		"comment,",
-		"",  # space between spans
 		# the role definition is unsupported as an IA2 role
 		# The role "ROLE_PARAGRAPH" is used instead
 		"has details",
 		"definition,",
-		"",  # space between spans
 		"has details",
 		"definition,",
-		"",  # space between spans
 		# The role "form" is deliberately unsupported
 		"has details",
 		"form",
@@ -862,7 +857,7 @@ def test_pr11606():
 	actualSpeech = _chrome.getSpeechAfterKey("tab")
 	_asserts.strings_match(
 		actualSpeech,
-		"section  multi line  editable  list  bullet  link  A    link  B"
+		"section  multi line  editable  list  bullet  link  A  link  B"
 	)
 	# move past the end of the first link.
 	# This should not be affected due to pr #11606.
@@ -886,7 +881,7 @@ def test_pr11606():
 	actualSpeech = _chrome.getSpeechAfterKey("NVDA+upArrow")
 	_asserts.strings_match(
 		actualSpeech,
-		"bullet  link  A    link  B"
+		"bullet  link  A  link  B"
 	)
 
 
@@ -1214,14 +1209,14 @@ def test_ariaRoleDescription_inline_contentEditable():
 	actualSpeech = _chrome.getSpeechAfterKey("downArrow")
 	_asserts.strings_match(
 		actualSpeech,
-		"Start  drawing  Our logo    End"
+		"Start  drawing  Our logo  End"
 	)
 	# When reading the line by word,
 	# Both entering and exiting the custom role should be reported.
 	actualSpeech = _chrome.getSpeechAfterKey("control+rightArrow")
 	_asserts.strings_match(
 		actualSpeech,
-		"drawing  Our logo    out of drawing"
+		"drawing  Our logo  out of drawing"
 	)
 	actualSpeech = _chrome.getSpeechAfterKey("control+rightArrow")
 	_asserts.strings_match(

tests/system/robot/symbolPronunciationTests.py

@@ -147,10 +147,10 @@ def test_moveByWord():
 			"don't",  # Expected: mid-word symbol
 			'right-pointing arrow', 't-shirt',
 			# todo: There should not be any "empty" words. Expect 'bar bar', and 'at  caret  star  line'
-			'1', 'bar', '2', '', '3', '', '4',
+			'1', 'bar', '2', 'blank', '3', 'blank', '4',
 			# end of first line
 			'blank',  # single space and newline
-			'',  # tab and newline  todo: There should not be any "empty" words.
+			'blank',  # tab and newline  todo: There should not be any "empty" words.
 			'blank',  # 4 spaces and newline
 			'right-pointing arrow',
 			't-shirt',
@@ -195,14 +195,14 @@ def test_moveByLine():
 		symbolLevel=SymLevel.NONE,
 		expectedSpeech=[
 			'Say', '(quietly)', 'Hello,', 'Jim .', "don't",  # Expect:
-			'',  # todo: Expect 'right-pointing arrow'
+			'blank',  # todo: Expect 'right-pointing arrow'
 			't-shirt',
-			'',  # todo: Expect 'right-pointing arrow'
+			'blank',  # todo: Expect 'right-pointing arrow'
 			't-shirt',
 			't-shirt',  # todo: Expect 'right-pointing arrow t-shirt'
 			'1   2    3      4',  # todo: Should symbols be passed to synth, i.e. "1 | 2 || 3 etc"?
 			'blank',  # single space
-			'',  # tab  # todo: There should not be any "empty" lines.
+			'blank',  # tab  # todo: There should not be any "empty" lines.
 			'blank',  # four spaces
 			'blank',  # end of doc
 		]

user_docs/en/changes.t2t

@@ -101,6 +101,7 @@ Instead NVDA reports that the link has no destination. (#14723)
 - Reporting of object shortcut keys has been improved. (#10807)
 - When rapidly moving through cells in Excel, NVDA is now less likely to report the wrong cell or selection. (#14983, #12200, #12108)
 - NVDA now generally responds slightly faster to commands and focus changes. (#14928)
+- Text sequences that are empty due to dictionary or symbol level processing are spoken as "blank" instead of not being spoken at all (#13431).
 -