Re: Bogus self reference error with -nu

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

On Wed, May 21, 2025 at 1:03 AM Philippe Altherr
<philippe.altherr@xxxxxxxxx> wrote:
>
> function () {
>   typeset -nu ref1=foo
>   echo ref1=$ref1
>   typeset -nu ref2=ref2
>   echo ref2=$ref2
> }
>
> There is no ref2 in the upper scope, so there should be no self reference error.

It took more head-scratching for this than it should have, but there
is in fact a reference error here -- it's just that the error message
could stand to be clarified.

The problem here is that "local -nu ref2=ref2" creates a reference not
just to a global that doesn't exist, but to a global that CAN NEVER
exist, because the local is in the way.  The following changes the
error message.  (I'll include it in a larger patch as an attachment
later.)

diff --git a/Src/params.c b/Src/params.c
index 7b515515e..2f4658f7a 100644
--- a/Src/params.c
+++ b/Src/params.c
@@ -6436,7 +6436,10 @@ setscope(Param pm)
        }
        if (refname && upscope(pm, pm->base) == pm &&
            strcmp(pm->node.nam, refname) == 0) {
-           zerr("%s: invalid self reference", refname);
+           if (pm->node.flags & PM_UPPER)
+               zerr("%s: local reference hides nonexistent global", refname);
+           else
+               zerr("%s: invalid self reference", refname);
            unsetparam_pm(pm, 0, 1);
        }
     } while (0);

This will have to be reconciled with Philippe's upscope_upper() patch,
which may have some merit, but at this point the patch relies on
upscope() not caring about upper.

> Once that bug is fixed, it's worth testing the following where the two references should again behave the same.
>
> function () {
>   typeset -nu ref1=foo
>   typeset foo=init
>   ref1=assign
>   echo ref1=$ref1
>   typeset -nu ref2=ref2
>   ref2=assign
>   echo ref2=$ref2
> }
> typeset -p foo ref2

They do behave the same post-patch in that both produce the "hides
nonexistent" error.