Re: [RESEND][PATCH] ld: -rpath must search under sysroot

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Thu, 6 Sep 2012, Andy Ross wrote:

> diff --git a/ld/emultempl/elf32.em b/ld/emultempl/elf32.em
> index de51050..35e0e7e 100644
> --- a/ld/emultempl/elf32.em
> +++ b/ld/emultempl/elf32.em
> @@ -1263,9 +1263,13 @@ fragment <<EOF
>  EOF
>  if [ "x${USE_LIBPATH}" = xyes ] ; then
>  fragment <<EOF
> -	  if (gld${EMULATION_NAME}_search_needed (command_line.rpath,
> -						  &n, force))
> -	    break;
> +	  if (command_line.rpath) {
> +	    char *tmprp = gld${EMULATION_NAME}_add_sysroot (command_line.rpath);
> +	    found = gld${EMULATION_NAME}_search_needed (tmprp, &n, force);
> +	    free(tmprp);
> +	    if (found)
> +	      break;
> +	  }

This looks like it may change what happens when there is no RPATH option 
at all.  Previously, gld${EMULATION_NAME}_search_needed would still be 
called even if command_line.rpath is NULL, and it in turn has logic:

  if (name[0] == '/')
    return gld${EMULATION_NAME}_try_needed (n, force);

  if (path == NULL || *path == '\0')
    return FALSE;

which handles the name[0] == '/' case *before* returning in the NULL rpath 
case.  Is this an intentional change?  If so, I think you need to give 
more explanation of why it's OK to change what the function does in this 
case (no -rpath option, name[0] == '/').

Note for coding style that the opening '{' goes on the line after the 
"if", appropriately indented, and that a space is needed after "free".

-- 
Joseph S. Myers
joseph@codesourcery.com