Date: Sun, 6 Jul 1997 18:33:37 -0700 (PDT)
From: David Rivas <jdr@shorter>
Subject: dlsym resolution in 1.1.3
To: mr@shorter, lindholm@shorter, thomasb@shorter

Didn't know who to talk to on this one, so=20
excuse me for the distribution.

I think the=041.1.3 symbol resolving logic, which seems
to have been totally rewritten in 1.2, has a little
bug that I found and need fixed.

Symbol resolution seems to come down to the function,
in src/solaris/java/runtime/linker_md.c:

long
sysDynamicLink(char *sym_p)
{
    void *f =3D 0;
    int i;

    /*
     * we need this kludge, because the java interpreter internals need
     * to access this before the Linker class is loaded and calls
     * the static initializers
     */
    if (!linkerinitialized) {
=09(void)sysInitializeLinker();
    }
   =20
    if ((f =3D dlsym(RTLD_NEXT, sym_p)) =3D=3D 0) {
=09for (i =3D useddlsegments; --i >=3D 0;) {
=09 if ((f =3D dlsym(dlsegment[i].handle, sym_p))) {
=09=09return((long)f);
=09 }
=09}
    }
    return 0;
}


The problem is that if the first dlsym succeeds it returns 0
instead of the symbol address.=20

A possible fix is:

long
sysDynamicLink(char *sym_p)
{
    void *f =3D 0;
    int i;

    /*
     * we need this kludge, because the java interpreter internals need
     * to access this before the Linker class is loaded and calls
     * the static initializers
     */
    if (!linkerinitialized) {
=09(void)sysInitializeLinker();
    }

    if ((f =3D dlsym(RTLD_NEXT, sym_p)) =3D=3D 0) {
=09for (i =3D useddlsegments; --i >=3D 0;) {
=09 if ((f =3D dlsym(dlsegment[i].handle, sym_p))) {
=09=09return((long)f);
=09 }
=09}
    }=20

    /* Return the symbol (which may be null) */
    return f;
}


For a variety of reasons I'm trying to get the AuidoEngine to
work on pre 1.2 platforms, and this bug gets tickled.

Who can review and put this (or another) fix back? I can
put it back if that's approprate.


david