Open /dev/tty in non-blocking mode #454
Conversation
|
See https://code.rocketnine.space/tslocum/cview/issues/58 for additional discussion. |
|
Thank you very much. |
|
Can confirm this fixes the issue introduced by 7694d90 However, it seems to add a slight delay when quitting an application. I'll try and make a minimal example to show this. |
| for { | ||
| select { | ||
| case <-stopQ: | ||
| return | ||
| default: | ||
| } | ||
|
|
||
| err = t.in.SetReadDeadline(time.Now().Add(250 * time.Millisecond)) |
There was a problem hiding this comment.
Well, 250ms feels like the delay I had.
There was a problem hiding this comment.
This causes us to wake up every 250ms. Which is vile on some platforms. I was hoping to avoid period polling.
There was a problem hiding this comment.
...Okay? What do you propose instead, considering the issues regarding suspend and resume (#452)?
| @@ -102,15 +102,19 @@ func (t *tScreen) disengage() { | |||
| // so that it can be restored when the application terminates. | |||
| func (t *tScreen) initialize() error { | |||
| var err error | |||
| t.out = os.Stdout | |||
| if t.in, err = os.Open("/dev/tty"); err != nil { | |||
| t.inFd, err = syscall.Open("/dev/tty", syscall.O_RDONLY|syscall.O_NONBLOCK, 0644) | |||
There was a problem hiding this comment.
This should not be necessary -- I believe os.Open() will do what you need, and you can call SetNonBlocking() if need be.
There was a problem hiding this comment.
I already tried this (on Linux) to no effect.
| if err == nil { | ||
| return nil | ||
| t.saved, err = term.GetState(t.inFd) | ||
| if err != nil { |
There was a problem hiding this comment.
This was a bug in the old code ... thanks for noticing it. :-)
|
In this comment you state:
Are you planning on returning to STDIN? This PR's only purpose is to work around issues introduced by 7694d90, so it can be closed if that is the case. |
|
I think I'll probably return to stdin, and offer an API for people to supply a file descriptor of their own creation. I just haven't done it yet. |
|
I've taken a different approach entirely -- doing some more experimentation. Please see my "tty" branch for an example. This still uses /dev/tty, but it solves the blocking on disengage, and provides a pluggable interface for folks who want to build a different I/O layer underneath entirely (e.g. for a websocket terminal or something.) |
|
My PR for this is #459 -- would appreciate test feedback. |
|
I've merged #459 -- after making some additional fixes for darwin (which has a buggy tty driver in how it handles close()). Those changes make this PR obsolete. |
tcell 2.2.1 has a bug [0][1][2] that made senpai need a keypress after Ctrl-C was hit. [0] gdamore/tcell#452 [1] gdamore/tcell#454 [2] gdamore/tcell@7694d90
Resolves #452.