[RFC] sysroot and dynamic libraries

[David Daney]

Alan Modra wrote:
> How do people feel about restricting a sysrooted ld to only search for
> dynamic libraries inside the sysroot?  Reason: ld effectively searches
> for any DT_NEEDED libs inside the sysroot.  Thus if you find a dynamic
> library outside the sysroot (due to passing a -L outside the sysroot)
> you may find an incompatible version of a DT_NEEDED dependency inside
> the sysroot.
> 
> Why would anyone want to pass a -L outside the sysroot?  Isn't the
> idea of a sysroot for cross toolchains?  So -L outside the sysroot
> can't find compatible libraries anyway?  Well, yes, but I'm extending
> the idea:  Making a sysroot for a native toolchain so that you can
> install a complete toolchain including a new glibc without danger of
> trashing your system.
> 

Having not tried your patch, I am a little confused.

My libc and other system libraries are indeed in the sysroot, so no problem there.  However hypothetically, I build my application (which includes several dynamic libraries with their own DT_NEEDED) outside of the sysroot and pass -L so the linker can find them.  It sounds like you are saying that with your change, the dependencies of libraries outside of the sysroot may no longer be found and I will now get linker errors unless I manually specify *all* needed libraries.

In any event, as long as most cross toolchains continue resolve dependencies in a somewhat sane manner I would feel neutral about the change.

David Daney





> Index: ld/emultempl/elf32.em
> ===================================================================
> RCS file: /cvs/src/src/ld/emultempl/elf32.em,v
> retrieving revision 1.190
> diff -u -p -r1.190 elf32.em
> --- ld/emultempl/elf32.em	15 Feb 2008 03:35:53 -0000	1.190
> +++ ld/emultempl/elf32.em	7 Aug 2008 00:46:34 -0000
> @@ -1497,6 +1497,9 @@ gld${EMULATION_NAME}_open_dynamic_archiv
>    if (! entry->is_archive)
>      return FALSE;
>  
> +  if (*ld_sysroot && !search->sysrooted)
> +    return FALSE;
> +
>    filename = entry->filename;
>  
>    /* This allocates a few bytes too many when EXTRA_SHLIB_EXTENSION
>