enhanced mouse tracking mode (1006) is set improperly when exiting tcell

[tmatz]

Fixed to properly reset enhanced mouse tracking mode (1006) when calling enableMouse(0).

related #477

Behavior before modification:

If I quit the tcell application with the mouse enabled and then start vim, the cursor doesn't move when I click the mouse.

The symptom was confirmed by _demos/mouse.go in the linux environment on Chromebook and Windows.

[gdamore]

So the problem is that VIM and tcell are fighting over mouse control...

I don't think it's good not to disable mouse tracking by default, because vim aside, most people will not like the result -- you'll move the mouse and it will inject a bunch of garbage into your shell window.

What need to do is figure out if we can query the mouse state when starting up, and restore it to what it was.

Unconditionally not clearing the mouse flags is unacceptable.

(To be honest, I think the problem here is VIM's fault -- when returning from a subshell, it should reenable the mouse behavior it wants, rather than making assumptions that the child process didn't mess with it. This is done for other settings like terminfo flags, and the mouse handling needs to be treated in the precise same way.)

[tmatz]

I found the specification of control sequnces of terminal app in chromebook.

https://chromium.googlesource.com/apps/libapps/+/master/hterm/doc/ControlSequences.md

[tmatz]

@gdamore
It seems that the chromebook uses a terminal made with javascript called hterm.

I checked the specifications and implementation of hterm's control sequence. As a result, I found two things.

First, it turns out that hterm completely ignores her 1003 settings. Therefore, not only motion events but also button and drag events are no longer generated. I suspect that the fix of #477 is a hack to deal with this behavior.

Second, SGR (1006) must be reset at the finalization of tcell. If the SGR setting remains, older applications that do not use SGR (such as vim) will be notified of unexpectedly formatted mouse events and will malfunction. In other words, the current tcell breaks the backward compatibility of the terminal.

I also found that the current implementation of tcell is incomplete. You can give MouseFlags as an argument to EnableMouse(), but if you give only MouseMotionEvents, the # 477 Hack won't work. It will work if you modify it as follows.

func (t *tScreen) enableMouse(f MouseFlags) {
        // Rather than using terminfo to find mouse escape sequences, we rely on the fact that
        // pretty much *every* terminal that supports mouse tracking follows the
        // XTerm standards (the modern ones).
        if len(t.mouse) != 0 {
                // start by disabling all tracking.
                t.TPuts("\x1b[?1000l\x1b[?1002l\x1b[?1003l\x1b[?1006l")
                if f&MouseMotionEvents != 0 {
                        t.TPuts("\x1b[?1000h\x1b[?1002h\x1b[?1003h\x1b[?1006h")
                } else if f&MouseDragEvents != 0 {
                        t.TPuts("\x1b[?1000h\x1b[?1002h\x1b[?1006h")
                } else if f&MouseButtonEvents != 0 {
                        t.TPuts("\x1b[?1000h\x1b[?1006h")
                }
        }
}

[tmatz]

@gdamore
I also checked the source of putty (https://git.tartarus.org/simon/putty.git) and found that putty also completely ignores 1003. Also, I didn't see any particular reason why 1006 had to be enabled when mouse events were disabled.

[gdamore]

I'm going to take a close look at this. My fear is that there is some impedance mismatch here in the intended behavior and the actual... and what happens in the terminals.

It is not surprising to me that some of these flags are ignored by terminals... the design actually depends upon it -- they should enable what they can, and ignore the rest.

The other thing is that my commentary earlier may also have been misunderstanding the context where this occurs.

[tmatz]

@gdamore thank you very much!