Have TentativeBoundary trigger rollback

[TylerLeonhardt]

The problem was, certain shells (like PowerShell) timeout when you hit backspace/left arrow at the far left of the prompt, rather than return something. This meant that the cursor is never invalidated because invalidation only happens when the response from the shell is different than what we predict.

This fix makes sure that the TentativeBoundary prediction also rolls back it's internal prediction which will make sure that our internal cursor receives the update that the internal state has changed.

Additionally this adds a couple tests.

[TylerLeonhardt]

Also, are there tests for this code somewhere? Or perhaps a place I could add new tests to?

[connor4312]

Ahh, right. Good catch. Although nothing is emitted we'd want to roll back in order to correct the cursor position.

why does the apply run the inner apply

The idea in the tenative boundary is to "buffer" the predictions and keep them from being emitted unless the tenatively proposed prediction is validated. If validation is successful, all the buffered predictions in the next generation will be instantly applied. So we still want to call apply() so that the cursor position is updated for use by those predictions.

Also, if you didn't guess by osmosis, the reason we maintain our own cursor position is that xterm writes are asynchronous, so if we looked up the cursor from the buffer each time that would cause issues since things we just wrote or predicted might not be reflected yet.

Also, are there tests for this code somewhere? Or perhaps a place I could add new tests to?

Here: https://github.com/microsoft/vscode/blob/master/src/vs/workbench/contrib/terminal/test/browser/terminalTypeahead.test.ts Tests are not super thorough right now, but fleshing them out is always good.

[TylerLeonhardt]

Let me take a stab at writing a test or two before this goes in.