return doesn't allow an indirect object

[tonycoz]

Since return isn't actually a function, I didn't think the "function" part of the original message applied.

Fixes #21716

[leonerd]

As this change makes a few things become compiletime errors that previously had been either runtime errors or potentially fine if subs were defined late, it's probably worth at least a mention in the perldelta so if people find code breaks they know why.

[tonycoz]

As this change makes a few things become compiletime errors that previously had been either runtime errors or potentially fine if subs were defined late, it's probably worth at least a mention in the perldelta so if people find code breaks they know why.

If the sub is defined late the indirect object has already been compiled into an rv2gv op:

# system perl without this patch
$ perl -MO=Concise,xx -e 'sub xx { return sum map { $_ } 1..5 } sub sum(@); '
main::xx:
a  <1> leavesub[1 ref] K/REFC,1 ->(end)
-     <@> lineseq KP ->a
1        <;> nextstate(main 2 -e:1) v ->2
-        <@> return KS ->-
-           <0> pushmark s ->2
3           <1> rv2gv sKR/1 ->4
2              <#> gv[*sum] s ->3
8           <|> mapwhile(other->9)[t4] K ->a
7              <@> mapstart K ->8
4                 <0> pushmark s ->5
-                 <1> null lK/1 ->5
-                    <1> null lK/1 ->8
-                       <@> scope lK ->-
-                          <;> ex-nextstate(main 3 -e:1) v ->9
-                          <1> ex-rv2sv sK/1 ->-
9                             <#> gvsv[*_] s ->8
6                 <1> rv2av lKPM/1 ->7
5                    <$> const[AV ARRAY] s ->6
-e syntax OK

It even runs and returns a GV:

# system perl without this patch
$ perl -MDevel::Peek -e 'sub xx { return sum map { $_ } 1..5 } sub sum(@); Dump([xx()])'
SV = IV(0x561df481e1a8) at 0x561df481e1b8
  REFCNT = 1
  FLAGS = (TEMP,ROK)
  RV = 0x561df481d750
  SV = PVAV(0x561df47f2f50) at 0x561df481d750
    REFCNT = 1
    FLAGS = ()
    ARRAY = 0x561df482da10
    FILL = 5
    MAX = 5
    FLAGS = (REAL)
    Elt No. 0
    SV = PVGV(0x561df48497b0) at 0x561df481d738
      REFCNT = 1
      FLAGS = (FAKE,MULTI)
      NAME = "sum"
      NAMELEN = 3
      GvSTASH = 0x561df47f1518  "main"
      FLAGS = 0x2
      GP = 0x561df482e180
        SV = 0x0
        REFCNT = 3
        IO = 0x0
        FORM = 0x0  
        AV = 0x0
        HV = 0x0
        CV = 0x561df481e3e0
        CVGEN = 0x0
        GPFLAGS = 0x0 ()
        LINE = 1
        FILE = "-e"
        EGV = 0x561df481e0c8    "sum"
    Elt No. 1
    SV = IV(0x561df481d7d0) at 0x561df481d7e0
      REFCNT = 1
      FLAGS = (IOK,pIOK)
      IV = 1
    Elt No. 2
    SV = IV(0x561df481d7e8) at 0x561df481d7f8
      REFCNT = 1
      FLAGS = (IOK,pIOK)
      IV = 2
    Elt No. 3
    SV = IV(0x561df481d878) at 0x561df481d888
      REFCNT = 1
      FLAGS = (IOK,pIOK)
      IV = 3

So perhaps it should be a deprecation instead.

That would make me sad for such obviously broken code, since it seems more likely that the intent is to call a function rather than return a glob.

[tonycoz]

That would make me sad for such obviously broken code, since it seems more likely that the intent is to call a function rather than return a glob.

which my perldelta illustrates.

[leonerd]

LGTM

I think this looks fine, though we should keep an eye out for BBC reports in case it turns out there's modules around that break on it - or even that actually rely on it to work.

[tonycoz]

I also considered a parser approach to fixing this (splitting LSTOP into list ops that take an indirob and those that don't), but that's more complex and riskier.