UIA: more intelligent word boundaries

[codeofdusk]

Windows Terminal version (or Windows build number)

OpenConsole commit 5deb332

Other Software

NVDA 2021.3

Steps to reproduce

Move by word in a UIA console containing the following text:

Loading file c:\test.json

Expected Behavior

In Microsoft Word, NVDA splits the word boundaries as follows:

['Loading ', 'file ', 'c', ':\\', 'test', '.', 'json']

Actual Behavior

In conhost, NVDA splits the words as follows:

['Loading ', 'file ', 'c:\\test.json                                                                                               \r\n                                                                                                                        ']

[codeofdusk]

NVDA's legacy console support (and the version I implemented for pre #4018 builds) uses the Uniscribe API to determinbe word boundaries. This sort of approach, if possible in conhost, would have two benefits:

• The boundaries are closer to those found in Microsoft Word and other applications and are therefore more aligned with user expectations.
• Some languages do not separate words with spaces or do so inconsistently. Is this of concern in conhost?

Otherwise, the word navigation in this commit is flawless (very fast and otherwise accurate). Great work @carlos-zamora!

[zadjii-msft]

IIRC we're using the wordDelimiters for splitting words, which are configurable by the user, even in conhost (though, much harder to set for conhost). I could be wrong about that though.

The "actual behavior" here though does sound wrong 😨