NVDA reads duplicate words when doing word navigation

[alexdima]

From @Neurrone

Extracted from #26730 (comment)

I've restarted NVDA with all add-ons disabled and am still encountering this. I'm experimenting with the read-only settings file and the user settings file.

Am just pressing the left/right arrow keys (arrows, ctrl + left/right arrow to move by word, etc). Notice this especially when you go right/left, then change direction.

Not sure what alt+f1 says, since i can't find the message with NVDA.

[alexdima]

@Neurrone

Running NVDA with all add-ons disabled will sometimes result in duplicate lines / letters being read due to the 30ms timeout in NVDA. The combination that works really well for me is NVDA + a single add-on, the one from @derekriemer in #26730 (comment) .

Here is the link for convenience: https://files.derekriemer.com/globalEditorTimer-1.0.nvda-addon

Have you tried this simple combination? Again, I have not tested having multiple add-ons at the same time. If you tell me the add-ons you have installed, I can try to reproduce using the add-on combination you have, but using the single add-on linked above works for me.

Once NVDA merges PR #7201 and produces a build where the timeout is increased from 30ms to 100ms, and it is made configurable, I don't think the plugin will be necessary anymore.

[Neurrone]

@alexandrudima I'm using this file to test:

int main(int argc, char* argv) {
    
}       

I think this issue is only specific to navigating around using the ctrl+left/right arrow keys. From the start of the first line, this is what I hear when using ctrl + right arrow (one line per press):

int 
main(
int 
int 
argc, 
argc, 
char* 
char* 
)

In contrast, this is what I hear in notepad++, which is the expected behaviour:

int
main
(
int
argc
,
char
*
argv
)
{

And this is what I hear for Vs Code pressing ctrl+left arrow from the end of the line:

{
)
a
char* 
char* 
argc, 
argc, 
int 
main(
main(
int 

The duplicate words and unpredictable reading (especially for argv) makes it confusing.
In notepad++, going in the reverse direction also speaks the reverse output, as expected.

[alexdima]

Aha! I think I understand the problem now. We navigate words according to our own fancy rules. This means that we position the cursor at a location which is perhaps unexpected by NVDA, which perhaps has some hard-coded assumptions about how cursor navigation should work.

Our rules are not that fancy, but they are different than Notepad's. For example, when doing ctrl+right Notepad always stops at the beginning of words. When doing ctrl+right, we stop at the end of the words. We also have our own set of word separators which can be customized, while Notepad has them hard-coded.

We have some alternative word navigation commands that can be bound instead of the default ones. I'm going to look into using them and checking if they play better with what NVDA expects. Otherwise, we can look into implementing a set of word navigation commands that match 100% what Notepad does.

[Neurrone]

Even when I try the following change to my keymap file, its still not reading accurately. I think the cursor is moving correctly, but maybe it isn't being tracked properly?

// Place your key bindings in this file to overwrite the defaults
[
    {
        "key": "ctrl+right",
        "command": "cursorWordStartRight",
        "when": "editorTextFocus"
    },
    {
        "key": "ctrl+right",
        "command": "-cursorWordEndRight",
        "when": "editorTextFocus"
    }
    
]

[derekriemer]

@jchadwick Does NVDA have hard coded rules? I think this is just a caret positioning bug.

[derekriemer]

s/jchadwick/@jcsteh/
sorry guys

[alexdima]

I think words would be announced if our word navigation would match NVDA's rules as implemented at https://github.com/nvaccess/nvda/blob/4d792f034d3c56a879b1d7b68fd1a5a7306576ce/source/textInfos/offsets.py#L226