in reply to Re^2: XS/Inline::C concat *any* two SVs.
in thread XS/Inline::C concat *any* two SVs.

Drawing on the code that has so far been presented, it looks to me that you might want: 
SV* test( SV *a, SV *b ) {
    SV * c;
    if(SvREADONLY(a)) {
      c = newSVsv(a);
      sv_catsv(c, b);
      return c;
      }

    sv_catsv(a,b);
    c = newSVsv(a);
    return c;
}

[download]
I don't know how to avoid the "uninitialized value in subroutine entry" warning (apart from the obvious). Predicting whether you're going to get it or not is a little tricky. Of the 2 following subs, only test5() produces the warning for me: 
SV * test4(SV * a, SV * b){return newSVuv(42);}
SV * test5(char * a, char * b){return newSVuv(42);}

[download]
And if I comment out the sv_catsv() call in the code that vkon posted, then the warning disappears. Obviously the exact nature of what the xsub does has a bearing on whether we get that warning.

Cheers,
Rob

Replies are listed 'Best First'.
Re^4: XS/Inline::C concat *any* two SVs.
by BrowserUk (Patriarch) on May 28, 2006 at 14:15 UTC

    The code I posted in the OP now does what I need it to do. Basically the XS equivalent of Perl's 

    $scalar1 .= $scalar2;

    [download]

    I don't want to create a new scalar, except in the specific instance of a readonly input. The routine will be called many times and I need to accumulate the results in the scalar.

    Duplicating the (already appended to), scalar in order to avoid using SvREFCOUNT_inc() is extremely wasteful and completely negates the benefits of perl's dynamically allocated string management. You have a scalar with n bytes, you call sv_catsv() to add m bytes to it. Having done that, you make a complete copy of m+n bytes in order to avoid the unreferenced scalar warning? This makes no sense to me, especially as the scalar of n bytes originates from Perl and has never gone out of scope.

    My question still remains, why am I getting that warning?

    The question is are there any other things that I need to test for and handle in order that I can handle any pair of SVs as input.

    And if I comment out the sv_catsv() call ...

    That is the (other) problem. If you call an SV that is undef as the first parameter to sv_catsv(), (which it usually will be on the first call for a particular SV), you get the warning. Hence the need for 

        if( !SvPOK( a ) )      // Still nothing, must be undef?
        sv_setpv( a, "" ); // Make it the null string to stop (one pos
+sible)
        // Use of uninitialized value in subroutine entry from sv_cats
+v

    [download]
    Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
    Lingua non convalesco, consenesco et abolesco. -- Rule 1 has a caveat! -- Who broke the cabal?
    "Science is about questioning the status quo. Questioning authority".
    In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]
[select]
      You can't concatenate onto a readonly scalar, in either Perl or XS. 
      #!/usr/bin/perl
use strict;
use warnings;

use Inline C => <<'END_C';

void 
readonly(SV *sv) {
    SvREADONLY_on(sv);
}

END_C

my $foo = "foo";
readonly($foo);

# This dies with: 
# "Modification of a read-only value attempted at cat_readonly.plx lin
+e 16."
$foo .= "bar";

# unreachable:
print "$foo\n";

      [download]

      So, if you have to deal with the possibility that your first scalar might be readonly, you have to do $a = $a . $b or its equivalent in XS rather than $a .= $b. It might be wasteful, but it's the only way.

      --
      Marvin Humphrey
      Rectangular Research ― http://www.rectangular.com
[reply]
[d/l]
[select]

        If you look at the OP, you'll see that is exactly what I do.

        Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
        Lingua non convalesco, consenesco et abolesco. -- Rule 1 has a caveat! -- Who broke the cabal?
        "Science is about questioning the status quo. Questioning authority".
        In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]

        If you look at the OP, you'll see that is exactly what I do. But only for readonly scalars.

        Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
        Lingua non convalesco, consenesco et abolesco. -- Rule 1 has a caveat! -- Who broke the cabal?
        "Science is about questioning the status quo. Questioning authority".
        In the absence of evidence, opinion is indistinguishable from prejudice.
[reply]
[d/l]

In Section Seekers of Perl Wisdom