Re: precmd: write error: interrupted

[Frank Terbeck <ft@bewatermyfriend.org>]

Hi Bart,

Bart Schaefer wrote:
[...]
> Here's a stab at a more comprehensive patch.  This follows the pattern of
> child_block()/child_unblock() which we use elsewhere to reap children only
> when safe to do so.
>
> I'm not 100% confident that this does the right thing for non-interactive
> shells/sourcing of scripts, but I *think* the change in shingetline()
> should cover those cases.

I'm afraid these changes break SIGWINCH handling of child processes.

If you start an editor or mail-reader or terminal-multiplexer (or
whatever really), that child will not react to any terminal-resizes
anymore.

It took a while to find out, why this:

  TRAPWINCH() { print $COLUMNS $LINES; }

didn't trigger in shells inside the terminal-multiplexer I use, but does
trigger in a naked xterm without multiplexer. The fact that the terminal
multiplexer itself didn't react to SIGWINCHes anymore steered me into
zsh's direction (well, not directly but eventually).

Building a version of the shell before these changes¹ makes everything
work again.  Sorry to be the bearer of bad news. ;)


Regards, Frank

¹ * cbf6f14 <Jun T> 31357: _cp: add support for Mac OS X (5 days ago)