save and restore TTY attributes #725
Conversation
rolandwalker
force-pushed
the
termios
branch
2 times, most recently
from
279a7ff to
0f4987e
Compare
add TCSAFLUSH action on exit so tig can't leave behind pending keystrokes
* restore tty attributes from init-time termios * make fseek ... SEEK_END unconditional after command * use TCSAFLUSH when setting tty attributes to block leakage of keystrokes across tig/external boundary - in the "after" case, this is probably more robust than fseek() on the file pointer
| void | ||
| init_display(void) | ||
| { | ||
| bool no_display = !!getenv("TIG_NO_DISPLAY"); | ||
| const char *term; | ||
| int x, y; | ||
|
|
||
| init_tty(); |
There was a problem hiding this comment.
This will open /dev/tty for tests. I think it would be better to skip this if no_display is true.
There was a problem hiding this comment.
Intentionally so — from the point of view of the OS, a TTY is much more than a display.
The TTY is also the medium for the line discipline, in rough terms the interactive input.
A TTY is also inseparable from process-group leadership via tcsetpgrp(), which among other things determines the disposition of signals, which is relevant under the test harness.
Also, regardless of whether a display is visible, tig is still linked with ncurses, and ncurses maintains a thoroughly TTY-oriented understanding of the world in which it runs. The goal I was driving for here is for ncurses to keep an input-only TTY under no_display. Tig still does not display anything, but the input "side" of ncurses remains connected to a proper ioctlable TTY.
The separate matter of process-group leadership becomes relevant when there are long-lived child processes as proposed in #301/#734. The simplest and most robust way to handle child cleanup is to
- be process-group leader
SIGHUPyour own process group at exit.
In the most ordinary case, the interactive shell makes tig the process-group leader transparently, and we don't have to think about it. However, that magic does not happen in non-interactive cases, such as when driven by tig-pick or the test harness (where no_display happens to be set). So it is a good idea, when you know you may be invoked from a script, to grab a TTY as a step toward becoming process-group leader.
Congrats on the release!
|
Thanks a lot for writing down the detailed motivation. |
|
From: Jonas Fonseca <[email protected]>
To: jonas/tig <[email protected]>
Cc: Subscribed <[email protected]>
Sent: Saturday, September 30, 2017 7:30 PM
Subject: Re: [jonas/tig] save and restore TTY attributes (#725)
Merged #725.—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.
|
main program:
opt_ttybecomes a structtcgetattr()at init timetcsetattr()indone_display()TCSAFLUSHaction indone_display()so that tig can't leave keystrokes behindopen_external_viewer(), after running external command:TCSAFLUSHaction when restoring tty attributes to block keystrokes leaking in from external command phase - probably more robust thanfseek()on the file pointerfseek … SEEK_END, which does no harm, make it unconditionaledit: also set tty attributes and flush before running external command