FAQ title bar answer using chpwd()

FAQ title bar answer using chpwd()

[Andrew Janke]

Hi, ZSH workers,

I have a question about an answer in the ZSH FAQ. Is this the right list 
to ask about it?

http://zsh.sourceforge.net/FAQ/zshfaq03.html#l23

Answer 3.6 suggests using chpwd() to emit terminal control sequences to 
update the title bar for Apple's Terminal.app. This probably isn't the 
right place to do it, because cd's can happen inside functions, 
pipelines, and so on whose output is going to be captured for use 
elsewhere. The terminal control sequences will contaminate the captured 
output instead of going to the terminal.

There's at least one place in the wild where this causes an issue, with 
Android development scripts: 
https://github.com/robbyrussell/oh-my-zsh/issues/3524

The prompt or other known user-interaction output points are probably 
the only safe place to emit terminal control sequences like this. So 
precmd() is probably a better choice than chpwd() for this. It seems 
like chpwd() should probably not be used to do anything that emits to 
stdout, in case it's being captured.

Should the FAQ answer be updated to reflect this?

Cheers,
Andrew

Re: FAQ title bar answer using chpwd()

[Daniel Shahaf]

Andrew Janke wrote on Fri, Mar 06, 2015 at 22:40:11 -0500:
> Hi, ZSH workers,
> 
> I have a question about an answer in the ZSH FAQ. Is this the right
> list to ask about it?
> 
> http://zsh.sourceforge.net/FAQ/zshfaq03.html#l23
> 

Personally I'd have addressed zsh-users.

> Answer 3.6 suggests using chpwd() to emit terminal control sequences
> to update the title bar for Apple's Terminal.app. This probably
> isn't the right place to do it, because cd's can happen inside
> functions, pipelines, and so on whose output is going to be captured
> for use elsewhere. The terminal control sequences will contaminate
> the captured output instead of going to the terminal.
> 
> There's at least one place in the wild where this causes an issue,
> with Android development scripts:
> https://github.com/robbyrussell/oh-my-zsh/issues/3524
> 

We had a similar issue at Subversion: http://svn.apache.org/r1643793

> The prompt or other known user-interaction output points are
> probably the only safe place to emit terminal control sequences like
> this. So precmd() is probably a better choice than chpwd() for this.
> It seems like chpwd() should probably not be used to do anything
> that emits to stdout, in case it's being captured.
> 
> Should the FAQ answer be updated to reflect this?

Why is a chpwd() hook set within the context of the Android scripts in
the first place?  Is it because they get sourced instead of executed?
If so, they really should switch to the ssh-agent/gpg-agent model, of
[eval `script`] instead of [source 'script'].

In the meantime, they can do something like:
[ -n "$ZSH_VERSION" ] && cd() { builtin cd -s "$@" }

>

Incidentally, another possible fix would be to have the FAQ recommend to
have the control sequences be printed to the controlling terminal,
rather than to stdout.  That would cause a different problem: instead of
contaminating the output, the terminal title might be wrong if a script
did 'cd' and then exited without changing back to the interactive user's
directory, as in '(cd ..)'.

Cheers,

Daniel

Re: FAQ title bar answer using chpwd()

[Bart Schaefer]

On Mar 6, 10:40pm, Andrew Janke wrote:
}
} http://zsh.sourceforge.net/FAQ/zshfaq03.html#l23
} 
} Answer 3.6 suggests using chpwd() to emit terminal control sequences to 
} update the title bar for Apple's Terminal.app. This probably isn't the 
} right place to do it, because cd's can happen inside functions, 
} pipelines, and so on whose output is going to be captured for use 
} elsewhere. The terminal control sequences will contaminate the captured 
} output instead of going to the terminal.

If you've actually examined the FAQ example, you'll see

  [[ -t 1 ]] || return

which means to do nothing if the standard output is not a terminal.

That's a perfectly reasonable solution to the issue described here:

} https://github.com/robbyrussell/oh-my-zsh/issues/3524

In short, the FAQ answer is correct and whoever created the chpwd to
which that issue refers, has coded it wrong.

Re: FAQ title bar answer using chpwd()

[Bart Schaefer]

On Mar 7,  4:19am, Daniel Shahaf wrote:
}
} We had a similar issue at Subversion: http://svn.apache.org/r1643793

OK, so the real problem is that

- there exists some software written for the user's shell to execute

- that software needs to change working directory

- the user is an idiot who has installed a chpwd function that always
  prints to stdout

- this breaks the software and is reported as a bug to the developer.

Frankly I don't think any update to the zsh FAQ is going to eliminate
idiocy.  The developer needs to either use "cd -q ..." if the shell is
known to be zsh, or to use "{ cd ... } >/dev/null" if zsh-ness for some
reason is not testable.

Re: FAQ title bar answer using chpwd()

[Lawrence Velázquez]

On Mar 7, 2015, at 2:20 AM, Bart Schaefer <schaefer@brasslantern.com> wrote:

> If you've actually examined the FAQ example, you'll see
> 
>  [[ -t 1 ]] || return
> 
> which means to do nothing if the standard output is not a terminal.

Would the FAQ code clean up properly after executing a subshell that
switched directories? It doesn't seem to, in my experiments.

    % (cd /)

vq