'exit' inside zshz() kills the interactive shell when the datafile is misconfigured

[ahjota]

Summary

Three exit calls in zsh-z.plugin.zsh execute in the user's current shell instead of just aborting the plugin:

1. The source-time version check (is-at-least 4.3.11) — kills the shell during startup on old zsh.
2. In zshz(): a custom datafile set without a directory path (ZSHZ_DATA not containing /).
3. In zshz(): the datafile being a directory.

Because z is an alias for zshz 2>&1 running in the current shell, cases 2 and 3 terminate the interactive session. With terminals that close when the shell exits (e.g. Ptyxis' default exit-action=close), the tab or window disappears — same failure mode as the well-known gretzky/n.zsh exit 1 chpwd bug.

Reproducer

% mkdir /tmp/zdir
% export ZSHZ_DATA=/tmp/zdir   # or: ~/.z somehow becomes a directory
% z foo
ERROR: Zsh-z's datafile (/tmp/zdir) is a directory.
# shell exits; terminal tab closes

Non-interactive confirmation (zsh 5.9, macOS arm64; plugin @ acd0e19):

% zsh -c 'source zsh-z.plugin.zsh; ZSHZ_DATA=/tmp/zdir; zshz foo; print SURVIVED'
ERROR: Zsh-z's datafile (/tmp/zdir) is a directory.
%                        # <- SURVIVED never prints; the shell is gone (exit status 0)

The no-directory-path case behaves the same: ZSHZ_DATA=z z foo.

Expected behavior

Print the error and return 1, leaving the shell alive. The precmd path is unaffected either way — (zshz --add "$PWD" &) already runs in a subshell — so this only changes the interactive-alias and source-time behavior.

I have a patch ready on a branch in my fork (the three exits become return 1) and can open a PR if you'd like.

[agkozak]

@ahjota Thank you for your thorough and thoughtful analysis. I had noticed the same problem myself and fixed it on my develop branch. The two datafile checks now return 1, and the version gate uses return 1 2> /dev/null || exit 1 so it still exits correctly if the file is run as a standalone script.

All that's holding up the release is some documentation work, which I should finish in the next couple of days.

Would you do me the favor of confirming that that branch fixes the problem? Thank you.

[ahjota]

Sure thing, I'll try it out today.

Also, to clarify "well known gretzky/n.zsh bug" -- well known to me, maybe not the zsh plugins community. :D I've just been vetting the zsh plugins I use since I've had a lot of issues since moving to Ptyxis.

[agkozak]

Thank you for your help. I just added a little cosmetic code to prevent the error message from being displayed twice in a row.

[ahjota]

     ################################################################
     # zsh-z develop branch: c04860f
     # zsh version: zsh 5.9 (x86_64-redhat-linux-gnu)
     ################################################################

     ################################################################
     # CASE 1: Datafile is a directory
     ################################################################

     $ mkdir /tmp/zdir_test
     $ export ZSHZ_DATA=/tmp/zdir_test
     $ z foo

     ERROR: Zsh-z's datafile (/tmp/zdir_test) is a directory.
     SURVIVED (exit=1)

     ----------------------------------------------------------------
     Result: shell survived, error printed, exit code 1
     ----------------------------------------------------------------

     ################################################################
     # CASE 2: Datafile set without a directory path
     ################################################################

     $ export ZSHZ_DATA=justafile
     $ z foo

     ERROR: You configured a custom Zsh-z datafile (justafile), but have not specified its directory.
     SURVIVED (exit=1)

     ----------------------------------------------------------------
     Result: shell survived, error printed, exit code 1
     ----------------------------------------------------------------

     ################################################################
     # CASE 3: Version gate (is-at-least 4.3.11)
     ################################################################

     Source-time check uses: return 1 2>/dev/null || exit 1
       - When sourced:  return 1 (shell stays alive)
       - When run as script: exit 1 (correct, script should exit)

     95:  return 1 2> /dev/null || exit 1

     ----------------------------------------------------------------
     Result: sourced plugin returns instead of killing the shell;
             standalone script still exits as expected
     ----------------------------------------------------------------

     ################################################################
     # BONUS: Quiet mode — precmd background add is silent
     ################################################################

     $ export ZSHZ_DATA=/tmp/zdir_test
     $ (zshz --add "$PWD" &!)   # runs in precmd subshell

     SURVIVED (no errors above = quiet fix works)

     ----------------------------------------------------------------
     Result: no error spam at every prompt; foreground 'z foo' still
             reports the error (see Case 1)
     ----------------------------------------------------------------

[ahjota]

Also tested with v2.0.0, thanks!

[agkozak]

I just released v2.0.0. Thanks for your input, @ahjota.