Re: [xml] HP-UX compilation warnings for 2.6.23

[Rick Jones <rick jones2 hp com>]

> > "xpath.c", line 11113: warning #2128-D: loop is not reachable from preceding
> >          code
> >      do {
> >      ^
>
>
>   looks buggy to me, there is a goto next_node; below the loop, and 
> next_node is declared inside the loop, not very nice, but I think the compiler
> is in error, probably worth pointing to your compiler engineers...

I'll see what I can do.

> > "xmlmodule.c", line 284: warning #2167-D: argument of type "void **" is
> >          incompatible with parameter of type "shl_t *"
> >      rc = shl_findsym(&handle, name, TYPE_UNDEFINED, symbol);
>
>
>   IIRC it's a problem of portability of the shared libs loading APIs 
> between platforms. If you can generate an HP-UX specific patch that would be
> good !

HP-UX has dlopen since 11.11.  From what I can see of the config.log, it looks 
like the configure script isn't looking for it - at least not on HP-UX, or 
perhaps finding shl_load is overriding it.  I'm a little leary of messing with 
configure.in - especially since my HP-UX system does not have autoconf on it at 
the moment.

However, I can (and really should) add a check to my netperf4 configure script 
to check for dlopen - just plain old AC_CHECK_FUNCS and bring that over to my 
HP-UX 11.23 system and indeed, it does find dlopen.

So, it would appear that a bit of configure manipulation might suffice, and then 
on those HP-UX systems with dlopen it will take the dlopen path rather than the 
shl_load path.

As a proof of concept I tried altering libxml2's config.h after configure to set 
HAVE_DLOPEN but make rebuild config.h on me, so I had to hardcode it in 
config.h.in.  A subsequent recompile appears to have been happy:

       source='xmlmodule.c' object='xmlmodule.lo' libtool=yes \^M
        DEPDIR=.deps depmode=none /bin/sh ./depcomp \^M
        /bin/sh ./libtool --mode=compile --tag=CC cc -DHAVE_CONFIG_H  -I. -I. 
-I. -I./include -I./include  -D_REENTRANT
      -g -c -o xmlmodule.lo xmlmodule.c^M
 cc -DHAVE_CONFIG_H -I. -I. -I. -I./include -I./include -D_REENTRANT -g -c 
xmlmodule.c  -DPIC -o .libs/xmlmodule.o^M
 cc -DHAVE_CONFIG_H -I. -I. -I. -I./include -I./include -D_REENTRANT -g -c 
xmlmodule.c -o xmlmodule.o >/dev/null 2>&1^M


> > "runtest.c", line 3857: warning #2177-D: function "thread_specific_data" was
> >          declared but never referenced
> >  thread_specific_data(void *private_data)
> >  ^
>
>    Seems testThread() is not defined in the case of HP-UX, either extend
> #if defined(linux) || defined(solaris) around line 3900 or define a specific
> test for the OS. Patch welcome !

Any history as to why it was #if defined(linux) || defined (solaris)?

Since the code in that block is including pthread.h, how about changing the #if 
to #ifdef HAVE_PTHREAD_H  ?

# diff runtest.c.orig runtest.c
3901c3901
< #if defined(linux) || defined(solaris)
---
> #ifdef HAVE_PTHREAD_H
# diff -c runtest.c.orig runtest.c
*** runtest.c.orig      Sat Dec 10 02:48:08 2005
--- runtest.c   Thu Feb 23 11:12:26 2006
***************
*** 3898,3904 ****
      return ((void *) Okay);
  }

! #if defined(linux) || defined(solaris)

  #include <pthread.h>

--- 3898,3904 ----
      return ((void *) Okay);
  }

! #ifdef HAVE_PTHREAD_H

  #include <pthread.h>

FWIW that seems to compile cleanly on my system.


> > "tree2.c", line 30: warning #2550-D: variable "dtd" was set but never used
> >      xmlDtdPtr dtd = NULL;       /* DTD pointer */
> >                ^
>
>
>    Right, but as an example dropping the result of a function returning a
> node would be of bad taste, keeping it is better.

Is it good taste to not check the return value at all ?-)  Or free the dtd 
pointer at the end? Or did freeing the doc at the end take care of free'ing the dtd?

# diff -c ../libxml2-2.6.22/doc/examples/tree2.c doc/examples/tree2.c
*** ../libxml2-2.6.22/doc/examples/tree2.c      Sun Jan  2 02:25:03 2005
--- doc/examples/tree2.c        Thu Feb 23 11:24:05 2006
***************
*** 45,50 ****
--- 45,56 ----
       */
      dtd = xmlCreateIntSubset(doc, BAD_CAST "root", NULL, BAD_CAST "tree2.dtd");

+     /* this is just to suppress any overly enthusiastic compiler warnings
+        about the dtd not being referenced. */
+     if (NULL == dtd) {
+       i = 0;
+     }
+
      /*
       * xmlNewChild() creates a new node, which is "attached" as child node
       * of root_node node.

(I forgot to do the mv foo.c foo.c.orig; cp foo.c.orig foo.c bit so the diff is 
against 2.6.22)

Making all in examples
        source='tree2.c' object='tree2.o' libtool=no \
        DEPDIR=.deps depmode=none /bin/sh ../../depcomp \
        cc -DHAVE_CONFIG_H -I. -I. -I../.. -I../../include -I../../include 
-I./include  -D_REENTRANT      -g -c tree2.c
        /bin/sh ../../libtool --mode=link --tag=CC cc  -g    -o tree2  tree2.o 
../../libxml2.la -lpthread   -lm
cc -g -o .libs/tree2 tree2.o  ../../.libs/libxml2.so -lpthread -lm
creating tree2

rick jones