Re: Return value of write() is not properly evaluated

[Thomas Dickey]

On Tue, Dec 30, 2014 at 03:26:57PM +0100, Stian Skjelstad wrote:
> Hello
> 
> I am the maintainer of Open Cubic Player, and the last couple of days I
> have done a couple of tests, using ncurses on lubuntu.
> 
> I stumbled across some graphical issues, and it appeared visible in several
> different terminal emulators when running on higher resolutions. After some
> experimentation I discovered that the problem was not visible if streaming
> stdout via tools like tee: eg
> ocp-curses | tee                  <- no UI noise
> ocp-curses                         <- UI noise every now and then
> 
> Later I did some more testing with strace, and I discovered one point that
> I think is the key issue
> 
> write(1, "\33[H\33[30m\37m\3..........", 8025) = 4096
> --- SIGALRM {si_signo=SIGALRM, si_code=SI_KERNEL} ---
> ioctl(4, SNDRV_PCM_IOCTL_HWSYNC, 0x1)   = 0
> ioctl(4, SNDRV_PCM_IOCTL_HWSYNC, 0x3)   = 0
> ioctl(4, SNDRV_PCM_IOCTL_HWSYNC, 0x20)  = 0
> ioctl(4, SNDRV_PCM_IOCTL_HWSYNC, 0x20)  = 0
> getitimer(ITIMER_REAL, {it_interval={0, 14320}, it_value={0, 9190}}) = 0
> rt_sigreturn()                          = 4096
> rt_sigaction(SIGTSTP, {0x7f2f4f97a0b0, [], SA_RESTORER|SA_RESTART,
> 0x7f2f50e17da0}, NULL, 8) = 0
> select(1, [0], NULL, NULL, {0, 0})      = 0 (Timeout)
> 
> 
> The usage of setitimer() can cause write() to return before a complete
> write has occurred. If piped via software like tee, the receiver will
> swallow bigger chunks of the buffer, making it more likely that a full
> buffer is sent before write() returns.

That sounds like

       The  curses  library does ``line-breakout optimization'' by looking for
       typeahead periodically while updating the screen.  If input  is  found,
       and  it is coming from a tty, the current update is postponed until re‐
       fresh or doupdate is called again.  This allows faster response to com‐
       mands  typed  in  advance.   Normally, the input FILE pointer passed to
       newterm, or stdin in the case that initscr was used, will be used to do
       this typeahead checking.  The typeahead routine specifies that the file
       descriptor fd is to be used to check for typeahead instead.  If  fd  is
       -1, then no typeahead checking is done.

... though an additional complication since 5.9 is that ncurses would behave
different if your program mixes direct writes to stdout (outside ncurses).
A quick read of the source code doesn't show me any of that, but it's something
to keep in mind.

-- 
Thomas E. Dickey <address@hidden>
http://invisible-island.net
ftp://invisible-island.net

signature.asc
Description: Digital signature