make a hash from .xs file

[Jeff Walker]

I am trying to build a hash from an .xs file, then return a ref to that hash
to the perl code.  I figured out how pass the hash back using a reference,
but when I get the hash back, all the keys exist, but all the values are
undefined.  Here is some stripped down code that illustrates this:

SV *
Blah()
        PREINIT:
        HV * h;
        int i;
        char name[100];
        SV **ssv;
        SV *svval;
        CODE:
        
        h = newHV(); /* allocate new hash */
        RETVAL = newRV_inc( (SV *) h ); /* return a ref to the hash */
        SvREFCNT_dec( h ); /* free the hash whenever the ref is gone */

        for ( i = 0 ; i < 100 ; i++ ) {
                /* make the new scalar value which will be the value */
                svval = newSViv( i );

                /* make the string that will be the key */
                sprintf( name, "%d", i );

                ssv = hv_store( h, name, strlen( name ), svval, 0 );
                if ( ssv == NULL ) {
                        croak( "oops: key %s not stored", name );
                }
        }

        OUTPUT:
        RETVAL

The perl program (which of course uses the module) get the ref returned by
Blah.  I then print the keys of this hash and everything is okay.  Then I
try to see the values (foreach $thing ( keys( %$hash ) ) ) but they are all
not defined (via ! defined($hash{$thing} ))

I tried incrementing the ref count of svval (I thought maybe adding it to
the hash didn't increase the ref count for the scalar).

I also have a different problem that isn't exposed in this example.  I am
making a module from a c++ object, so am having to statically link a new
perl.  When I do this, stuff like the above happens (undefined hash values)
but then perl never exits at the end of the script.  and perfmeter tells me
that CPU usage goes to 100% on a CPU.  I kill perl with SIGINT and
everything is fine.  I assume that this has something to do with garbage
collection, maybe some bad ref counts?  I can't reproduce it with a small
program (although it happens every time with the big one), so I don't expect
too much, but maybe someone else has heard of that?

Thanks in advance.

-- 
Jeff Walker                   MatchLogic, Inc.
[EMAIL PROTECTED]        7233 Church Ranch Blvd.
Voice 1 (303) 222-2105        Westminster, CO  80021
Fax   1 (303) 222-2001        www.matchlogic.com