[Perl/perl5] 3364ec: find_lexical_cv: fix following the parent indexes ...

  Branch: refs/heads/blead
  Home:   https://github.com/Perl/perl5
  Commit: 3364ec30b83a92c7135acbc44b80f7fd8858131d
      https://github.com/Perl/perl5/commit/3364ec30b83a92c7135acbc44b80f7fd8858131d
  Author: Tony Cook <[email protected]>
  Date:   2026-03-25 (Wed, 25 Mar 2026)

  Changed paths:
    M op.c
    M t/op/lexsub.t

  Log Message:
  -----------
  find_lexical_cv: fix following the parent indexes after a gap

Normally padvars inherited from outer scopes have a PARENT_PAD_INDEX()
value that is that names index in the outer scope, found by
CvOUTSIDE().

So imagine these pads, each for an enclosing scope:

  [&true] [&reftype] [&foo] [&bar]  # outer most scope
  [&true PPI=1] [&reftype PPI=2]    # A scope
  [&true PPI=1]                     # B scope

where PPI is the PARENT_PAD_INDEX() for that entry, so the B scope
inherits "&true" from index 1 in the A scope, and A inherits "&true"
from index 1 and "&reftype" from index 2 in the outermost scope.

This might correspond to code like:

  # definitions of true, reftype, foo, bar
  use builtin qw(true reftype);
  my sub foo { ... }
  my sub bar { ... }

  sub A {
    # inherited references to reftype and true
    if (reftype($somevar) eq "HASH" && ...)
       return true;

     my sub B {
       eval "... reftype(...) ...";
       return true;
     }
  }

The pad cannot be modified after the sub has finished compilation.

So if we do an eval within the B scope and reference &reftype:

  [&true] [&reftype] [&foo] [&bar]  # outer most scope
  [&true PPI=1] [&reftype PPI=2]    # A scope
  [$y] [&true PPI=2]                # B scope
  [&reftype]                        # eval scope

we can no longer modify scope B to build the PARENT_PAD_INDEX chain up
to the original imported definition.

Before b0bc598140d4 this could result in an assertion or segmentation
fault, since find_lexical_cv() didn't handle the gap in the chain at
all.

The fix in b0bc598140d4 handled the case where the name was found
after the gap in the defining scope, this allows find_lexical_cv() to
correctly find &foo or &bar starting from the eval scope, but it
didn't correctly adjust the working pad index when that name was
inherited from some parent scope, such as &reftype in A being
inherited from the outer most scope.

So for &reftype, the starting index is 1, find_lexical_cv() sees the
missing index, does an exhaustive search up the CvOUTSIDE() chain and
finds it in the A scope, but didn't adjust the index to the found
entry 2, this meant it started following the chain from the &true
entry, resulting it finding the CV for &true and failing to compile
since the prototype for &true allows no arguments, unlike &reftype.

This change fixes that, so when the name is found after the break in
the PARENT_PAD_INDEX chain, the index starts from the correct pad
entry.

Fixes #24131
Fixes #24056


  Commit: a26628aaf68b64f792c1f02d8400cd2f7ce93084
      https://github.com/Perl/perl5/commit/a26628aaf68b64f792c1f02d8400cd2f7ce93084
  Author: Tony Cook <[email protected]>
  Date:   2026-03-25 (Wed, 25 Mar 2026)

  Changed paths:
    M pod/perldelta.pod

  Log Message:
  -----------
  perldelta for find_lexical_sv() finding the wrong sub


Compare: https://github.com/Perl/perl5/compare/a76201055098...a26628aaf68b

To unsubscribe from these emails, change your notification settings at https://github.com/Perl/perl5/settings/notifications