Re: PATCH: Rebound references don't behave like newly bound ones

X-seq: zsh-workers 54264

From: Philippe Altherr <philippe.altherr@xxxxxxxxx>

To: Zsh hackers list <zsh-workers@xxxxxxx>

Subject: Re: PATCH: Rebound references don't behave like newly bound ones

Date: Sat, 28 Mar 2026 01:02:19 +0000

Arc-authentication-results: i=1; mx.google.com; arc=none

Arc-message-signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20240605; h=to:subject:message-id:date:from:in-reply-to:references:mime-version :dkim-signature; bh=BOXT88MDMadymRvBPFS8WoNxlp9StrQ8AUeJzjT+/mo=; fh=BgAYDYpL6Ne/A5nWEMVJiHiBtrz8Imz3uf26RDwgQX4=; b=Brc/WSP9fi+8ssYHGyg+GO9CeakrC9qTmkwO9SNYHWRxE63gjgGqVCdsb7Upf7tHH2 QeaOd0jAUhihQTUMLkGYacCtpHnSwHeUFX9EZScVfNmDFA7tAsCYRqYdRjKzJoPG30+f Bu0iUBMc6xAPaMunY7ndJfL4j9rBV+PCOsaCj/lo8jYhbcHdKpQwxR1wOY60xGO2wuRj 6gWnLh2OMlr077piS/oybGEvfZ/lHnHngpM4QkM11r6hQ7FV5y5HCeZqwk6Hhzn/srme ICH7wFRp7OJ011CBWGnY+HSCql4PpKvZcoQSfkxFqAl7Z7eEw7QaDonJs4Uy/OaXQYq4 5hFA==; darn=zsh.org

Arc-seal: i=1; a=rsa-sha256; t=1774659750; cv=none; d=google.com; s=arc-20240605; b=DtqeKTPgKLYMoE/cIB6shhsqNH/zplSypDojPi2BlqOT+ZhgXGp86wsHA6HNEkqA0g /m2Uk1jhVFm102rGIgOQ+5dNp2eEmIO+CqGLq/mXmR4bDUMCn4L9iTUW4ZeNcT2ThKqq wWafq15UGW4n1sBUdxgkuCEEbJ1+p+asm/Co1Urpj4l3cjJ3LSgIBaLO3Z6tssaDLAYh JPOBiD2fUia8BQp9y07IHbHE+8HSP4+GpCsZT3idxzBRpnDqBsZ69O2+AxXPwNv2D06G u+FrQCoCyRpxvh/hhC6LP4lcsNbEu2x8I1XclRIox50ANgdJffvq7ZjmJAJHgPpD7qAk lvzg==

Archived-at: <https://zsh.org/workers/54264>

In-reply-to: <CAGdYchtCQmMjJ0sv-dAWfq9uucbikPkuV3zBvNC7viVZVs88Aw@mail.gmail.com>

List-id: <zsh-workers.zsh.org>

References: <CAGdYchvg-3TyCf-+tJafpU1qHgFA1goaTtOzf6ms5-Q5gPu1EQ@mail.gmail.com> <CAGdYchtCQmMjJ0sv-dAWfq9uucbikPkuV3zBvNC7viVZVs88Aw@mail.gmail.com>

Here is the updated patch that I forgot to attach.

Philippe

On Sat, Mar 28, 2026 at 12:52 AM Philippe Altherr <philippe.altherr@xxxxxxxxx> wrote:

Here is an updated version based on workers/54262.

- Fix rebound refs to behave the same as newly bound ones.

Philippe

On Sat, Nov 8, 2025 at 4:55 PM Philippe Altherr <philippe.altherr@xxxxxxxxx> wrote:
Two references referencing the same variable should always behave the same. This is currently not always the case if one of them previously referenced a different variable that went out of scope.

Example
typeset -n ref1 ref2
local var=l1
() {
  () {
    local var=l3
    ref1=var
    echo A1: ref1=$ref1 ref2=$ref2
  }
  ref2=var
  # At this point, "ref1" and "ref2" refer to the same variable
  # "var". Going forward, they should behave the same.
  echo A2: ref1=$ref1 ref2=$ref2
  () {
    local var=l3
    echo A3: ref1=$ref1 ref2=$ref2
  }
  echo A4: ref1=$ref1 ref2=$ref2
  local var=l2
  echo A5: ref1=$ref1 ref2=$ref2
}
echo A6: ref1=$ref1 ref2=$ref2

Current output
Expected output
A1: ref1=l3 ref2=
A2: ref1=l1 ref2=l1
A3: ref1=l3 ref2=l1
A4: ref1=l1 ref2=l1
A5: ref1=l2 ref2=l1
A6: ref1=l1 ref2=l1
A1: ref1=l3 ref2=
A2: ref1=l1 ref2=l1
A3: ref1=l1 ref2=l1
A4: ref1=l1 ref2=l1
A5: ref1=l1 ref2=l1
A6: ref1=l1 ref2=l1

Fixing this discrepancy requires adopting what was described as the "all the way up" strategy for rebound variables. Thankfully, once the fix for hidden references is in (workers/54047), implementing "all the way up" is trivial:

diff --git a/Src/params.c b/Src/params.c
index 5e9387339..c737bf748 100644
--- a/Src/params.c
+++ b/Src/params.c
@@ -5853,7 +5853,8 @@ endparamscope(void)
     for (Param pm; refs && (pm = (Param)getlinknode(refs));) {
if ((pm->node.flags & PM_NAMEREF) && !(pm->node.flags & PM_UPPER) &&
pm->base > locallevel) {
- setscope_base(pm, locallevel);
+ pm->base = 0;
+ setscope(pm);
}
     }
     unqueue_signals();

This change also has for effect to detect and signal incidental reference loops (i.e., reference loops created when a reference is rebound) as soon as the loop is created rather than only when it's first accessed.

Example
typeset -n ref1
typeset -n ref2=ref1;
() {
  typeset ref2=foo
  ref1=ref2
  echo ref1=$ref1
}
echo ref1 and ref2 now form a loop
: $ref1
echo NOT REACHED

Current output
New output
ref1=foo
ref1 and ref2 now form a loop
zsh: ref1: invalid self reference
ref1=foo
zsh: ref2: invalid self reference

- Fix rebound refs to behave the same as newly bound ones.

Philippe

diff --git a/Src/params.c b/Src/params.c index 4f5454abb..461e02acf 100644 --- a/Src/params.c +++ b/Src/params.c @@ -5905,7 +5905,8 @@ endparamscope(void) for (Param pm; refs && (pm = (Param)getlinknode(refs));) { if ((pm->node.flags & PM_NAMEREF) && !(pm->node.flags & PM_UNSET) && !(pm->node.flags & PM_UPPER) && pm->base > locallevel) { - setscope_base(pm, locallevel); + pm->base = 0; + setscope(pm); } } unqueue_signals(); diff --git a/Test/K01nameref.ztst b/Test/K01nameref.ztst index 82ccbfb89..0b4475827 100644 --- a/Test/K01nameref.ztst +++ b/Test/K01nameref.ztst @@ -1258,6 +1258,81 @@ F:previously this could create an infinite recursion and crash >h2: ref1=f1 ref2=f1 ref3=f1 >f1: ref1=f1 ref2=f1 ref3=f1 + () { + typeset -n ref1 ref2 + local var=l1 + () { + () { + local var=l3 + ref1=var + echo A1: ref1=$ref1 ref2=$ref2 + } + ref2=var + # At this point, "ref1" and "ref2" refer to the same variable + # "var". Going forward, they should behave the same. + echo A2: ref1=$ref1 ref2=$ref2 + () { + local var=l3 + echo A3: ref1=$ref1 ref2=$ref2 + } + echo A4: ref1=$ref1 ref2=$ref2 + local var=l2 + echo A5: ref1=$ref1 ref2=$ref2 + } + echo A6: ref1=$ref1 ref2=$ref2 + } + () { + typeset -n ref1 ref2 + () { + () { + local var=l3 + ref1=var + echo B1: ref1=$ref1 ref2=$ref2 + } + ref2=var + # At this point, "ref1" and "ref2" refer to the same undefined + # variable "var". Going forward, they should behave the same. + echo B2: ref1=$ref1 ref2=$ref2 + () { + () { + local var=l4 + echo B3: ref1=$ref1 ref2=$ref2 + } + local var=l3 + echo B4: ref1=$ref1 ref2=$ref2 + } + local var=l2 + echo B5: ref1=$ref1 ref2=$ref2 + () { + () { + local var=l4 + echo B6: ref1=$ref1 ref2=$ref2 + } + local var=l3 + echo B7: ref1=$ref1 ref2=$ref2 + } + echo B8: ref1=$ref1 ref2=$ref2 + } + local var=l1 + echo B9: ref1=$ref1 ref2=$ref2 + } +0:rebound nameref behaves the same as newly bound one +>A1: ref1=l3 ref2= +>A2: ref1=l1 ref2=l1 +>A3: ref1=l1 ref2=l1 +>A4: ref1=l1 ref2=l1 +>A5: ref1=l1 ref2=l1 +>A6: ref1=l1 ref2=l1 +>B1: ref1=l3 ref2= +>B2: ref1= ref2= +>B3: ref1=l4 ref2=l4 +>B4: ref1=l3 ref2=l3 +>B5: ref1=l2 ref2=l2 +>B6: ref1=l2 ref2=l2 +>B7: ref1=l2 ref2=l2 +>B8: ref1=l2 ref2=l2 +>B9: ref1=l1 ref2=l1 + # # The following two tests are linked, do not separate # @@ -1500,39 +1575,12 @@ F:$$[1] reference should print the first digit of $$ but prints nothing () { typeset ref2=foo ref1=ref2 + echo reached } - echo reached - echo $ref1 - echo NOT REACHED -1:expansion of incidental reference loop triggers error ->reached -*?*: ref1: invalid self reference - - typeset -n ref1 - typeset -n ref2=ref1; - () { - typeset ref2=foo - ref1=ref2 - } - echo reached - ref1=foo - echo NOT REACHED -1:assignment to incidental reference loop triggers error ->reached -*?*: ref1: invalid self reference - - typeset -n ref1 - typeset -n ref2=ref1; - () { - typeset ref2=foo - ref1=ref2 - } - echo reached - typeset -n ref3=ref1 echo NOT REACHED -1:reference to incidental reference loop triggers error +1:incidental reference loop triggers error >reached -*?*: ref1: invalid self reference +*?*: ref2: invalid self reference typeset -A -g VAR0=(aa AA) typeset -n -g REF0=VAR0

Reverse Date

Date

Thread

Author