Re: [patch] Matsushita AM33/2.0 port

[Roland McGrath <roland at redhat dot com>]

> --enable-add-ons=am33,linuxthreads wasn't enough to get
> am33/sysdeps/.../linuxthreads added to the search dirs.  Adding a
> linuxthreads Implies to linuxthreads/sysdeps didn't add them either.
> Am I missing anything?

I was suggesting that we consider changing the algorithm so that would work
and then use that style.  I didn't say that it would already work now.  If
this indeed seems to be the best way to address the nptl/linuxthreads
conflict issue, then I will make it work that way.

> 	* Makeconfig (shlib-versions.v.i): Move top-level
> 	shlib-versions last.

I did this one.

> 	* sysdeps/unix/sysv/linux/configure.in (arch_minimum_kernel):
> 	Leave it alone if it's already set.
> 	* sysdeps/unix/sysv/linux/configure: Rebuilt.

I did this one a little differently, so a higher-priority fragment can
override the ones in the case statement too.

> 	* configure.in (base_machine): Let add-on's mach.sh set it.
> 	* configure: Rebuilt.

I don't want to have a plethora of little scripts add-ons provide.  All
existing add-ons have no use for the subdir-style configure script.  My
inclination is to punt that and make the one canonical required thing be a
fragment that is included early on so it can set base_machine.  If an
add-on needs configure hooey to run later in the process, it can provide an
add-on/sysdeps/something/configure fragment for that.  If an add-on wants
any subdir-style configure runs done, it can do AC_CONFIG_SUBDIRS.

Does that seem reasonable?


Thanks,
Roland