Re: --with-sysroot newbie troubles: "ld: cannot open crt1.o"

[Daniel Jacobowitz <drow at mvista dot com>]

On Tue, Jan 13, 2004 at 01:02:19PM -0800, Daniel Kegel wrote:
> Daniel Jacobowitz wrote:
> >On Mon, Jan 12, 2004 at 05:37:15PM -0800, Daniel Kegel wrote:
> >>The following seem to be important when building a sysroot'd gcc/glibc 
> >>x86 -> x86_64 cross toolchain:
> >>
> >>1. you must configure glibc with --prefix=/usr; there seem to be several
> >>things that rely on this as a special case.
> >
> >In fact, you must almost always configure glibc in this fashion. 
> >Otherwise the /lib - /usr/lib split does not happen.
> 
> I had been happily compiling without that split for many moons...
> this is the first time I've run into a situation where
> something doesn't work without it.

Well, compiling works.  Runtime gets a little strange if you want a
standard LSB/FHS compatible layout.

> >>2. $sysroot/lib/ must exist before building the final gcc, even though it 
> >>is not used, because $sysroot/lib64 is reached via $sysroot/lib/../lib64
> >>
> >>I did those, and I now have a toolchain that can compile hello, world.
> >
> >Ah sorry, these are lib64-specific problems so I wasn't as familiar
> >with them.  I encountered the latter recently on MIPS64.  I tried to
> >fix GCC to make it go away, and it turned out to be... shall we say...
> >bloody hard.
> 
> Oh, it can't be that hard to canonicalize paths before using them, can it?

In fact it's impossible in the general case, and expensive otherwise. 
Think symlinks.

> But no matter, it's easy enough to do "mkdir $sysroot/lib" before building.
> 
> By the way, what's the reasoning for the following hunk (first posted in
> http://gcc.gnu.org/ml/gcc-patches/2003-11/msg00405.html
> but now part of gcc-3.3.3-pre)?
> 
> --- gcc-3.3.1.hj/gcc/configure.in	2003-11-06 15:35:07.000000000 -0500
> +++ gcc-3.3.1.drow/gcc/configure.in	2003-11-06 15:34:23.000000000 -0500
> @@ -2430,7 +2431,7 @@ case "$target" in
>      ;;
>  esac
> 
> -if test x$with_sysroot = x && x$host = x$target \
> +if test x$with_sysroot = x && test x$host = x$target \
>     && test "$prefix" != "/usr" && test "x$prefix" != "x$local_prefix" ; 
>     then
>    AC_DEFINE_UNQUOTED(PREFIX_INCLUDE_DIR, "$prefix/include")
>  fi

Without it is a syntax error? :)  It's the same syntax error I _always_
make when writing shell scripts.  x$host gets interpreted as a command.

> I had been using something like --with-local-prefix=$sysroot/usr when 
> building the
> bootstrap gcc so it could find pthread.h when building the final gcc.
> With this hunk, that no longer works.  I guess that's ok -- I can
> switch to --with-sysroot for gcc-3.3.3 and later -- but it sure was
> confusing until I noticed the change.

The general idea is that $prefix/include for a sysrooted toolchain
contains things like bfd.h for the cross binutils.  The headers are in
$prefix/$target/include or $sysroot/usr/include.  If you use
--with-sysroot then usr/include should get searched automatically, no
hoops jumped through.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer