Increase default caret move timeout from 30 to 100 ms and make it configurable #7201

jcsteh · 2017-05-24T03:57:11Z

Link to issue number:

Fixes #6424.
Probably also addresses #5725, #6699.
Should address #4481 for terminal connections with <= 100 ms latency. It doesn't directly handle latency beyond 100 ms, but the ability to configure the timeout means this can be handled with user intervention.
Probably addresses microsoft/vscode#26730 (at least in some cases).

Summary of the issue:

NVDA has a very conservative cursor movement timeout of 30 ms. That is, when the user presses a cursor key, backspace, etc., we check the caret position, send the key press and then wait about 30 ms for the caret to move. 30 ms was chosen to ensure maximum responsiveness, but this just isn't enough in the wild; e.g. in complex web editors (particularly in Chrome) or when connecting to remote terminals with connection latency.

Description of how this pull request fixes the issue:

Previously, the default was hard-coded to 30 ms in editableText.EditableText._hasCaretMoved.
This is now configurable via config.conf["editableText"]["caretMoveTimeout"]. It is not exposed in the GUI, as this should only be changed by advanced users. The value is in ms, not seconds, as that's a better unit for this purpose. However, the function continues to use seconds for its arguments for backwards compatibility.
The default is now 100 ms.
Previously, _hasCaretMoved slept at the end of the loop but didn't check for caret movement again. This meant that it could return False even if the caret did actually move during the last sleep. Now, the timeout check happens after the caret movement check, ensuring the correct result.
_hasCaretMoved now works with ms rather than seconds to avoid floating point precision errors. These precision errors could result in an extraneous additional retry.
Added some debug logging to help diagnose caret tracking problems in future.

Testing performed:

In Notepad, tested cursoring through text, cursoring when already at the edge (e.g. left arrow at start of document) and backspacing. All worked as expected. Delay when already at edge was longer, which is expected.
Tested in Chrome with the test case from related Chromium issue 725051. Observed that cursor tracking was more reliable with this PR. Saw some cases where debug logging shows the caret took longer than 30 ms to move, which explains the increased accuracy with this PR.
Tested in Firefox as per point 1 and received expected results.
Tested backspacing while composing a message in Chrome, particularly when there's only one character in the message. In NVDA next, this often says nothing. With this PR, this works as expected. Again, observed debug logging shows the caret took longer than 30 ms to move in some cases.

Known issues with pull request:

None known.

Change log entry:

In Bug Fixes:

- In editable text, when moving the caret (e.g. with the cursor keys or backspace), NVDA's spoken feedback is now more accurate in many cases, particularly in Chrome and terminal applications. (#6424)

jcsteh · 2017-05-24T03:57:40Z

Ug. Accidentally clicked submit before I was finished. I've updated the initial comment.

…figurable. - 30 ms was chosen to ensure maximum responsiveness, but this just isn't enough in the wild; e.g. in complex web editors (particularly in Chrome) or when connecting to remote terminals with connection latency. - This is configurable via config.conf["editableText"]["caretMoveTimeout"]. It is not exposed in the GUI, as this should only be changed by advanced users. The value is in ms, not seconds, as that's a better unit for this purpose. However, the function continues to use seconds for its arguments for backwards compatibility. - Previously, _hasCaretMoved slept at the end of the loop but didn't check for caret movement again. This meant that it could return False even if the caret did actually move during the last sleep. Now, the timeout check happens after the caret movement check, ensuring the correct result. - _hasCaretMoved now works with ms rather than seconds to avoid floating point precision errors. These precision errors could result in an extraneous additional retry. - Added some debug logging to help diagnose caret tracking problems in future.

alexdima · 2017-05-24T07:44:36Z

@jcsteh Thank you very much!

feerrenrut

Looks good. Another PR with a great description!

feerrenrut · 2017-05-25T09:15:19Z

source/config/configSpec.py

@@ -192,6 +192,9 @@

 [upgrade]
 	newLaptopKeyboardLayout = boolean(default=false)
+
+[editableText]
+	caretMoveTimeout = integer(min=0, max=2000, default=100)


I would be tempted to mention the units as part of the name for this config option, something like caretMoveTimeoutMS or caretMoveTimeoutMilliSec

feerrenrut · 2017-05-25T09:15:19Z

source/editableText.py

+			timeout = config.conf["editableText"]["caretMoveTimeout"]
+		else:
+			# This function's arguments are in seconds, but we want ms.
+			timeout *= 1000


Since timeout is documented on the function to be in seconds, I think it could be easy to miss that its converted into milliseconds. Perhaps store the milliseconds in a new var, timeoutMs

feerrenrut · 2017-05-25T09:15:19Z

source/editableText.py

@@ -42,21 +43,30 @@ class EditableText(ScriptableObject):
 	#: Whether or not to announce text found before the caret on a new line (e.g. auto numbering)
 	announceNewLineText=True

-	def _hasCaretMoved(self, bookmark, retryInterval=0.01, timeout=0.03):
+	def _hasCaretMoved(self, bookmark, retryInterval=0.01, timeout=None):


In the description of this PR you mention backwards compat for this function I think? Changing timeout to None will risk breaking that.

It's certainly a valid point, but I think it's a negligible/acceptable risk.

It's most likely that a caller would call this function either accepting the defaults or passing its own timeout, both of which will still work as expected.

This doesn't call super, so there's no chance that some method will receive None that wasn't expecting it.

The only risk I can think of is that a subclass might have hard-coded the old 0.03 default, in which case they won't get the user's configured value. However, at worst, this means some subclass won't behave consistently with the rest of NVDA, and this would be pretty clear from the debug log if caret movement started timing out. I also think the likelihood of someone having done this is unlikely.

Since it's a private method, maintaining backwards compat is a courtesy anyway, not an obligation. :)

bhavyashah · 2017-05-25T14:09:58Z

I am a little uncertain here. Am I correct in assuming that this pull request and its corresponding code is part of the latest Next snapshot of NVDA?

…

On 5/25/17, Reef Turner ***@***.***> wrote: feerrenrut requested changes on this pull request. Looks good. Another PR with a great description! > @@ -192,6 +192,9 @@ [upgrade] newLaptopKeyboardLayout = boolean(default=false) + +[editableText] + caretMoveTimeout = integer(min=0, max=2000, default=100) I would be tempted to mention the units as part of the name for this config option, something like `caretMoveTimeoutMS` or `caretMoveTimeoutMilliSec` > @type timeout: float @return: a tuple containing a boolean denoting whether this method timed out, and a TextInfo representing the old or updated caret position or None if interupted by a script or focus event. @rtype: tuple - """ + """ + if timeout is None: + timeout = config.conf["editableText"]["caretMoveTimeout"] + else: + # This function's arguments are in seconds, but we want ms. + timeout *= 1000 Since timeout is documented on the function to be in seconds, I think it could be easy to miss that its converted into milliseconds. Perhaps store the milliseconds in a new var, `timeoutMs` > @@ -42,21 +43,30 @@ class EditableText(ScriptableObject): #: Whether or not to announce text found before the caret on a new line (e.g. auto numbering) announceNewLineText=True - def _hasCaretMoved(self, bookmark, retryInterval=0.01, timeout=0.03): + def _hasCaretMoved(self, bookmark, retryInterval=0.01, timeout=None): In the description of this PR you mention backwards compat for this function I think? Changing timeout to `None` will risk breaking that. -- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: #7201 (review)

-- Best Regards Bhavya Shah Blogger at Hiking Across Horizons: https://bhavyashah125.wordpress.com/ Contacting Me E-mail Address: bhavya.shah125@gmail.com Follow me on Twitter @BhavyaShah125 or www.twitter.com/BhavyaShah125 Mobile Number: +91 7506221750

jcsteh · 2017-05-25T23:47:58Z

@bhavyashah No, it's not in next. It's still undergoing code review. Once review is approved, it will be merged to next. For your information, a pull request that is in next will have the "incubating" label applied.

jcsteh · 2017-06-01T12:29:06Z

@feerrenrut, this is ready for review round 2. Thanks! :)

…figurable. Incubates #7201 (issue #6424).

…g valueChange events as caret moves. Also, treat caret events as caret moves as well, as this should be more reliable than comparing bookmarks in some cases.

jcsteh · 2017-06-21T01:32:25Z

@feerrenrut, would you remind reviewing the latest commit? @tspivey reported that this change was making the delete key feel laggy, which I've now fixed for most cases.

#7201 (issue #6424).

jcsteh · 2017-06-21T06:29:50Z

Arrrg. This latest change breaks Chrome again. Looks like caret events get fired even though Chrome hasn't updated the caret yet. I'm guessing they're MSAA system caret events, rather than IA2 caret events.

… uses the system caret, but this moves before IA2 is updated. Ignore system caret events for IA2 implementations and rely only on IA2 caret events.

jcsteh · 2017-06-22T05:43:18Z

@feerrenrut, it's not my week. Review latest commit pretty please? :)

#7201 (issue #6424).

…sed by c967b21. Firefox fires caret events even when you use caret movement commands when already at the edge of the control. That means we don't fire caretMovementFailed events in these cases. Furthermore, it seems quite a few things (e.g. Wordpad, Notepad++) don't fire valueChange or caret events when you press delete. To work around this, drop the checks for caret and valueChange events (and revert the shouldAllowSystemCaretEvent stuff). Instead, when pressing delete, compare the word at the caret (in addition to bookmarks) to detect movement.

jcsteh · 2017-06-26T03:02:10Z

@feerrenrut, yet another regression fix for you to review when you have a chance. Thanks.

… Incubates #7201 (issue #6424). Fixes #7323.

…if the word at the caret has changed once _hasCaretMoved_minWordTimeoutMs has elapsed. I.e. 30 ms must elapse before starting to check the word. This fixes a regression introduced by #7201 where deleting characters in Mozilla Gecko editable controls would cause the character being deleted to be spoken rather than the character that followed.

…if the word at the caret has changed once _hasCaretMoved_minWordTimeoutMs has elapsed. I.e. 30 ms must elapse before starting to check the word. (#7496) This fixes a regression introduced by #7201 where deleting characters in Mozilla Gecko editable controls would cause the character being deleted to be spoken rather than the character that followed.

amaniahmad · 2018-06-07T15:26:33Z

Dear
Please could help me
how make website accessibility
step by step
thanks for all

jcsteh requested a review from feerrenrut May 24, 2017

jcsteh force-pushed the i6424CaretMoveTimeout branch from 963c3c8 to a65c5b9 May 24, 2017

feerrenrut requested changes May 25, 2017

View changes

alexdima mentioned this pull request May 31, 2017

Accessibility: NVDA reading wrong or duplicate content in editor microsoft/vscode#26730

Closed

Review actions.

5da54fe

jcsteh requested a review from feerrenrut Jun 1, 2017

feerrenrut approved these changes Jun 7, 2017

View changes

jcsteh added a commit that referenced this pull request Jun 7, 2017

Increase default caret move timeout from 30 to 100 ms and make it con…

Loading status checks…

c81024b

…figurable. Incubates #7201 (issue #6424).

nvaccessAuto added the incubating label Jun 7, 2017

nvaccessAuto assigned jcsteh Jun 7, 2017

alexdima mentioned this pull request Jun 9, 2017

NVDA reads duplicate words when doing word navigation microsoft/vscode#28306

Closed

Avoid full caret move timeout when pressing the delete key by treatin…

c967b21

…g valueChange events as caret moves. Also, treat caret events as caret moves as well, as this should be more reliable than comparing bookmarks in some cases.

jcsteh requested a review from feerrenrut Jun 21, 2017

feerrenrut approved these changes Jun 21, 2017

View changes

jcsteh added a commit that referenced this pull request Jun 21, 2017

Fix lag when pressing delete key due to caret timeout change. Incubates

Loading status checks…

4d6c7be

#7201 (issue #6424).

Fix regression of cursor tracking in Chrome caused by c967b21. Chrome…

e526792

… uses the system caret, but this moves before IA2 is updated. Ignore system caret events for IA2 implementations and rely only on IA2 caret events.

jcsteh requested a review from feerrenrut Jun 22, 2017

feerrenrut approved these changes Jun 22, 2017

View changes

jcsteh added a commit that referenced this pull request Jun 22, 2017

Fix delete key lag without breaking caret tracking in Chrome. Incubates

Loading status checks…

104d2ff

#7201 (issue #6424).

jcsteh mentioned this pull request Jun 25, 2017

Next snapshot ignoring browse mode settings in search edit fields like google or startpage #7323

Closed

jcsteh requested a review from feerrenrut Jun 26, 2017

feerrenrut approved these changes Jun 26, 2017

View changes

jcsteh added a commit that referenced this pull request Jun 26, 2017

Fix regression of Automatic focus mode for caret movement in Firefox.…

Loading status checks…

1102092

… Incubates #7201 (issue #6424). Fixes #7323.

leonardder mentioned this pull request Jun 27, 2017

Regression: pressing delete in terminal causes error #7327

Closed

jcsteh merged commit e45c36a into master Jul 10, 2017

nvaccessAuto removed the incubating label Jul 10, 2017

nvaccessAuto added this to the 2017.3 milestone Jul 10, 2017

jcsteh deleted the i6424CaretMoveTimeout branch Jul 10, 2017

This was referenced Jul 10, 2017

Chrome app "line": speech reports next / previous character incorrectly #5725

Closed

NVDA does not navigate the text in edit field while browsing in google chrome #6699

Closed

NVDA Unreliable in Applications via SSH #4481

Closed

abdullah5490 mentioned this pull request Jul 10, 2017

NVDA read all text while reading word by word, in an edit field under chrome #7372

Closed

jcsteh mentioned this pull request Jul 18, 2017

KeyboardInputGesture.send should use SendInput rather than Keybd_event #4294

Closed

leonardder mentioned this pull request Jul 27, 2017

Cursor jumps twice in consoles #4890

Open

jcsteh mentioned this pull request Aug 7, 2017

Randomly switching from focus to browse mode in multi line edit boxes. #1958

Closed

michaelDCurran mentioned this pull request Aug 15, 2017

Fix regression from #7201 where delete would announce the wrong character in Mozilla Gecko edit controls #7496

Merged

nvaccess / nvda

Increase default caret move timeout from 30 to 100 ms and make it configurable #7201

Increase default caret move timeout from 30 to 100 ms and make it configurable #7201

jcsteh commented May 24, 2017 •

edited

jcsteh commented May 24, 2017

alexdima commented May 24, 2017

feerrenrut left a comment

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

bhavyashah commented May 25, 2017

jcsteh commented May 25, 2017

jcsteh commented Jun 1, 2017

jcsteh commented Jun 21, 2017

jcsteh commented Jun 21, 2017 •

edited

jcsteh commented Jun 22, 2017

jcsteh commented Jun 26, 2017

amaniahmad commented Jun 7, 2018

nvaccess / nvda

Sponsor nvaccess/nvda

Increase default caret move timeout from 30 to 100 ms and make it configurable #7201

Increase default caret move timeout from 30 to 100 ms and make it configurable #7201

Conversation

jcsteh commented May 24, 2017 • edited

Link to issue number:

Summary of the issue:

Description of how this pull request fixes the issue:

Testing performed:

Known issues with pull request:

Change log entry:

jcsteh commented May 24, 2017

alexdima commented May 24, 2017

feerrenrut left a comment

This comment has been minimized.

feerrenrut May 25, 2017 Contributor

This comment has been minimized.

feerrenrut May 25, 2017 Contributor

This comment has been minimized.

feerrenrut May 25, 2017 Contributor

This comment has been minimized.

jcsteh Jun 1, 2017 Author Contributor

bhavyashah commented May 25, 2017

jcsteh commented May 25, 2017

jcsteh commented Jun 1, 2017

jcsteh commented Jun 21, 2017

jcsteh commented Jun 21, 2017 • edited

jcsteh commented Jun 22, 2017

jcsteh commented Jun 26, 2017

amaniahmad commented Jun 7, 2018

jcsteh commented May 24, 2017 •

edited

feerrenrut May 25, 2017
Contributor

feerrenrut May 25, 2017
Contributor

feerrenrut May 25, 2017
Contributor

jcsteh Jun 1, 2017
Author Contributor

jcsteh commented Jun 21, 2017 •

edited