Thanks, Bart.  That fixes four of the five failures.  All except this one:

../../../src/zsh-2020-07-07/Test/E01options.ztst: starting.
This test hangs the shell when it fails...
Pattern match failed, line 1:
-<zsh: sure you want to delete all 15 files in
/opt/build/zsh-2020-07-07/Test/options.tmp \[yn\]\?
 <zsh: sure you want to delete all <-> files in / \[yn\]\?
+>nuc2% zsh: sure you want to delete all 15 files in
/opt/build/zsh-2020-07-07/Test/options.tmp [yn]?
 >nuc2% zsh: sure you want to delete all 28 files in / [yn]?

On Sun, Jul 5, 2020 at 1:23 PM Bart Schaefer <schaefer@brasslantern.com>
wrote:

> On Sun, Jul 5, 2020 at 9:57 AM Bart Schaefer <schaefer@brasslantern.com>
> wrote:
> >
> > On Sun, Jul 5, 2020 at 9:28 AM Bart Schaefer <schaefer@brasslantern.com>
> wrote:
> > >
> > > It's the change to supposedly subvert symlink attacks.
> >
> > Hmm, no, I'm getting random failures even after I revert.  Sorry for
> > the red herring.
>
> Ahem.  Those failures were something else in my environment.
>
> It does have to do with the "always" block.  All the tests pass if I
> just put a "return 0" after the "always".
>
> A bit of further digging indicates that
>   zpty_flush After comptesteval
> does not always succeed, but it doesn't need to.  The "always" is now
> preserving that failure status.
>
> So perhaps this?
>
> diff --git a/Test/comptest b/Test/comptest
> index b14d4c9..a36e301 100644
> --- a/Test/comptest
> +++ b/Test/comptest
> @@ -128,6 +128,7 @@ comptesteval () {
>         return 1
>       }
>       zpty_flush After comptesteval
> +     : Suppress error from zpty_flush
>     } always {
>       rm $tmp
>     }
>