[codeofdusk]

Link to issue number:

Supersedes #11722. Fixes #11740 (originally filed upstream as microsoft/terminal#5481). Blocking #10964.

Summary of the issue:

In an old implementation of NVDA's UIA console support, we originally used the standard _getTextLines implementation, which called getTextInChunks to get each line from the console according to UIA. This was extremely slow to run over the entire buffer, so we switched to getting all text and calling splitlines, which was much faster.

In pre-21H1 UIA console, every line ended with a newline character. In 21H1 console, new text can wrap across lines, so every line doesn't necessarily end in a newline. This makes splitlines inadequate for 21H1: when text wraps, NVDA starts reading from the beginning of the visible text for every new line of output!

Description of how this pull request fixes the issue:

This PR:

• Disables the _getTextLines override on the 21H1 UIA console. Pre-21H1 is entirely unaffected.
• Clarifies a comment in initialization at POSITION_LAST based on a conversation with @carlos-zamora.

Testing performed:

Re-tested the case in microsoft/terminal#5481 and verified that it is no longer reproducible.

Known issues with pull request:

• Please prioritize this PR for 2020.4.
• There may be some loss of performance on the 21H1 UIA console, see UI Automation in Windows Console: remove optimized _getTextLines to fix autoread in 21H1 console #11722 (comment). I haven't observed this in practice, but any performance loss would definitely be fixed by using Diff Match Patch (as only one cross-process call is needed for that character-based algorithm).

Change log entry:

None needed.

[AppVeyorBot]

• Build (for testing PR)
• System test failure

See test results for failed build of commit 7370b21590

[feerrenrut]

Thanks @codeofdusk