I wonder if each variables with ext magic has its own virtual table pointer.

See SvPVMG & SvPVLV at PerlGuts Illustrated. It's probably somewhat out of date due to the 5.9/5.10 changes, but it is still the clearest guide around.

As I say, for my purposes, the SvPVLV and the associated 'x' type magic would pretty much do everything I want, provided I can adjust the LV_ARGS.

If you're ok with messing with it, here's how:

use strict;
use warnings;

use Inline C => <<'__EOI__';

SV* f(SV* sv, I32 pos, I32 len) {
    SV* targ = newSV_type(SVt_PVLV);
    sv_magic(targ, NULL, PERL_MAGIC_substr, NULL, 0);
    LvTYPE(targ)    = 'x';
    LvTARG(targ)    = SvREFCNT_inc_simple(sv);
    LvTARGOFF(targ) = pos;
    LvTARGLEN(targ) = len;
    return targ;
}

__EOI__

use Devel::Peek qw( Dump );

my $x = "abcde";
$_=uc for f($x,1,3);
print("$x\n");        # aBCDe
[download]

If not, just copy the two functions relating to substr in mg.c

Yep! Bin there, done that. Trouble is, the lvalueness doesn't survive assignment:

#! perl -slw
use strict;
use warnings;

use Inline C => <<'__EOI__';

SV* f(SV* sv, I32 pos, I32 len) {
    SV* targ = newSV_type(SVt_PVLV);
    sv_magic(targ, NULL, PERL_MAGIC_substr, NULL, 0);
    LvTYPE(targ)    = 'x';
    LvTARG(targ)    = SvREFCNT_inc_simple(sv);
    LvTARGOFF(targ) = pos;
    LvTARGLEN(targ) = len;
    return targ;
}

__EOI__

use Devel::Peek qw( Dump );

my $x = "abcde";
my $r = f($x,1,3);
Dump $r;
print $r;

__END__
C:\test>ike2-lv.pl
SV = PVNV(0x410d558) at 0x97890
  REFCNT = 1
  FLAGS = (PADMY,POK,pPOK)
  IV = 0
  NV = 0
  PV = 0x40fa988 "bcd"\0
  CUR = 3
  LEN = 8
bcd
[download]

The effect I am after is persistant lvalueness as demonstrated here (and back up there somewhere):

C:\test>perl -MDevel::Peek=Dump -E"
$s = 'fred'; 
$r = \substr $s, 1, 2; 
$$r = 'oo'; 
say $s; 
Dump $_ for $s, $r; 
$$r = 'eu'; 
say $s;  
Dump $_ for $s, $r
"

food
SV = PV(0x16d6f0) at 0x97b60
  REFCNT = 3
  FLAGS = (POK,pPOK)
  PV = 0x16b3b8 "food"\0
  CUR = 4
  LEN = 8
SV = RV(0x97be8) at 0x97bd8
  REFCNT = 2
  FLAGS = (ROK)
  RV = 0x97c80
  SV = PVLV(0xe2c48) at 0x97c80
    REFCNT = 2
    FLAGS = (PADMY,GMG,SMG,pPOK)
    IV = 0
    NV = 0
    PV = 0x16b478 "oo"\0
    CUR = 2
    LEN = 8
    MAGIC = 0x103708
      MG_VIRTUAL = &PL_vtbl_substr
      MG_TYPE = PERL_MAGIC_substr(x)
    TYPE = x
    TARGOFF = 1
    TARGLEN = 2
    TARG = 0x97b60
    SV = PV(0x16d6f0) at 0x97b60
      REFCNT = 2
      FLAGS = (POK,pPOK)
      PV = 0x16b3b8 "food"\0
      CUR = 4
      LEN = 8
feud
SV = PV(0x16d6f0) at 0x97b60
  REFCNT = 3
  FLAGS = (POK,pPOK)
  PV = 0x16b3b8 "feud"\0
  CUR = 4
  LEN = 8
SV = RV(0x97be8) at 0x97bd8
  REFCNT = 2
  FLAGS = (ROK)
  RV = 0x97c80
  SV = PVLV(0xe2c48) at 0x97c80
    REFCNT = 2
    FLAGS = (PADMY,GMG,SMG,pPOK)
    IV = 0
    NV = 0
    PV = 0x16b478 "eu"\0
    CUR = 2
    LEN = 8
    MAGIC = 0x103708
      MG_VIRTUAL = &PL_vtbl_substr
      MG_TYPE = PERL_MAGIC_substr(x)
    TYPE = x
    TARGOFF = 1
    TARGLEN = 2
    TARG = 0x97b60
    SV = PV(0x16d6f0) at 0x97b60
      REFCNT = 2
      FLAGS = (POK,pPOK)
      PV = 0x16b3b8 "feud"\0
      CUR = 4
      LEN = 8
[download]

But I'd like to avoid the need for taking a reference and using indirection.

And to keep this all in one thread: No to "Are you really asking about making an lvalue function?"

---

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

RIP PCW It is as I've been saying!(Audio until 20090817)

I wonder if each variables with ext magic has its own virtual table pointer.

It does! That means that having only one extension type is not as limiting as it seems!

use strict;
use warnings;

use Inline C => <<'__EOI__';


static const char* test1_str = "Hello, ";

static int test1_mg_get(pTHX_ SV *sv, MAGIC *mg) {
    sv_setpvn(sv, test1_str, (STRLEN)strlen(test1_str));
    return 0;
}

static const MGVTBL vtbl_test1 = {
    test1_mg_get,
    NULL,
    NULL,
    NULL,
    NULL
};

SV* get_test1_var() {
    SV* sv = newSV_type(SVt_PVLV);
    sv_magicext(sv, NULL, PERL_MAGIC_ext, &vtbl_test1, NULL, 0);
    return sv;
}



static const char* test2_str = "World!";

static int test2_mg_get(pTHX_ SV *sv, MAGIC *mg) {
    sv_setpvn(sv, test2_str, (STRLEN)strlen(test2_str));
    return 0;
}

static const MGVTBL vtbl_test2 = {
    test2_mg_get,
    NULL,
    NULL,
    NULL,
    NULL
};


SV* get_test2_var() {
    SV* sv = newSV_type(SVt_PVLV);
    sv_magicext(sv, NULL, PERL_MAGIC_ext, &vtbl_test2, NULL, 0);
    return sv;
}

__EOI__

use Devel::Peek qw( Dump );

my $test1 = get_test1_var();
my $test2 = get_test2_var();
print("$test1$test2\n");
[download]

It does! That means that having only one extension type is not as limiting as it seems!

Actually, there are two. There is also uvar magic:

  U  PERL_MAGIC_uvar           vtbl_uvar       Available for use by ex
+tensions
[download]

You can also chain multiple magics. See the description for SvPVMG in Perlguts:

moremagic is a pointer to another MAGIC and is used to form a single linked list of the MAGICs attached to an SV.

For my purposes, it is the lvalueness associated with PVLVs that is of most interest.

---

Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.

"Science is about questioning the status quo. Questioning authority".

In the absence of evidence, opinion is indistinguishable from prejudice.

RIP PCW It is as I've been saying!(Audio until 20090817)

For my purposes, it is the lvalueness associated with PVLVs that is of most interest.

All of magics with a setter in their vtable have lvalueness. There's nothing special about PERL_MAGIC_substr, and the use of PVLV is not a requirement.

Note that returning a PVLV or something magical doesn't make your function an lvalue function. That's a compile-time determination indicated by the :lvalue attribute.

Are you really asking about making an lvalue function?