[gentoo-alt] IRIX

[Fabian Groffen]

----- Forwarded message from Fabian Groffen <grobian@g.o> -----

I replied to the gentoo-alt list, as what you're doing is quite
interesting, and more people that may help or are interested are hanging
out there.

On 19-12-2007 17:23:56 +0000, Stuart Shelton wrote:
> With LDFLAGS and CPPFLAGS set for the initial build stage, wget
> doesn't compile as USE="ssl", but OpenSSL hasn't been installed within
> prefix-portage yet.  It works fine with USE="-ssl".

The bootstrap guide for Solaris has the same.

> texinfo fails because 'help2man' (which it claims is optional) isn't
> available, so this must be installed first.

Hmmm... ok :(

> The next issue is with building coreutils, seemingly caused by the
> automake utilities (since the same package build without problems when
> using non-portage automake tools):

coreutils has an eautoreconf in it, apparently it doesn't like you.

>  * Done with patching
>  * Applying 6.9-iswblank.patch ...                                        [ ok ]
>  * Running eautoreconf in
> '/usr/opt/portage/var/tmp/portage/sys-apps/coreutils-6.9-r1/work/coreutils-6.9'
> ...
>  * Running aclocal -I m4 ...                                              [ !! ]
>
>  * Failed Running aclocal !
>  *
>  * Include in your bugreport the contents of:
>  *
>  *   /usr/opt/portage/var/tmp/portage/sys-apps/coreutils-6.9-r1/temp/aclocal-1944129.out

> $ cat /usr/opt/portage/var/tmp/portage/sys-apps/coreutils-6.9-r1/temp/aclocal-1944129.out
> ***** aclocal *****
>
> aclocal-1.10: autom4te failed with exit status: 1

Hmmm... might be your /bin/sh is too crappy for it to work, or the
wrapper isn't fully working yet.

> On 19/12/2007, Stuart Shelton <srcshelton@×××××.com> wrote:
> > Whist I remember, I also had to add "mips-irix" to portage/profiles/arch.list

Isn't it already in there?
It should, as it is in SVN.

> > If I interrupt ncurses, change those two Makefiles, and resume then it
> > builds successfully - and the only other problem is /bin/sh reporting
> > 'too many args' (event though ncargs is 20480) when installing
> > manpages.  Commenting this out and running that command manually
> > (using bash rather than sh) fixed this.

I guess using bash for everything on IRIX is a good idea in any case.

> > bash then installed without issues.
> >
> > The next thing I'm stuck on is that '=autoconf-2.6*' and
> > '=automake-1.5*' want perl >= 5.005, whereas SGI's default is 5.00405.
> >  There is perl-5.8.8 in my $PATH. but the package's configure scripts
> > aren't seeing it - does 'emerge' filter PATH?

Yep.  It sanitises it, uses it's own path from the prefix environment
and adds /usr/bin and /bin.

> > symlinking perl-5.8.8 into $EPREFIX/bin solved this issue.

Beware that this will give you a collision once you install perl.

> > On 19/12/2007, Stuart Shelton <srcshelton@×××××.com> wrote:
> > > Hi Fabian,
> > >
> > > After a long break, I've again picked up prefix-portage on IRIX.

Nice to see!

> > > I'm building from the latest prefix overlays, etc., and I'll let you
> > > know how I get on.
> > >
> > > Can sys-apps/sed-4.1.4-r1 be marked as ~mips-irix?

I think no packages have the mips-irix keyword in the tree.  I prefer to
wait with keywording once we know that we can do "system".

> > > sys-apps/sed-4.1.5 doesn't compile (with the platform compilers) as
> > > the following syntax isn't accepted:
> > >
> > > cc-1017 c99: ERROR File = regcomp.c, Line = 561
> > >   An unmatched left bracket ("[") appears in an expression.
> > >
> > >     [0 ... 0x80 / BITSET_WORD_BITS - 1] = BITSET_WORD_MAX
> > >        ^

I have no idea what this actually is supposed to mean (the code), so I
can't come up with a fix.
I wonder if it really is a good idea to stay with MIPSpro... from my
perspective gcc is preferred, even though it doesn't have as good
performance.  It may very well be like Sun Studio compilers that we
should come up with the ability to use a native compiler after
bootstrapping at your own risk or something.

> > > I've also noticed that, using the portage 'emerge' command built from
> > > bootstrap-prefix.sh, it hangs on completion (either after outputting
> > > build errors or the ">>> Completed installing sed-4.1.4-r1 into
> > > /usr/opt/portage/var/tmp/portage/sys-apps/sed-4.1.4-r1/image/opt/portage/"
> > > message - but not if package collisions are detected!).
> > >
> > > Update: After rebuilding with 'FEATURES="-collision-protect"', emerge
> > > completes - perhaps it was building a cache of some form?  If this is
> > > the case, some notification would be useful: It really did look to
> > > have hung.

Hmmm... I recal seeing a hang, but we never really tackled it.  Not much
I can say about it.

> > > Finally, and I guess there's no easy answer to this one, many packages
> > > override CFLAGS with "-O2 -g", which is insufficient for properly
> > > building on IRIX with MIPSpro compilers... I'm not sure if portage
> > > offers any mechanism to override this?

No packages should override CFLAGS.  If they do, they are broken from
Gentoo's point of view, and should be fixed.  You probably see Portage
setting CFLAGS to this value (it is its default, IMHO).  Set CFLAGS in
etc/make.conf and Portage will obey your wishes.

> > > (For example, findutils resets CFLAGS to '-O2 -g', which drops the
> > > '-c99' option to enable C99-mode compilation, and enabling extended
> > > headers.  The simplest solution to this is for set CC to c99 - but
> > > it's not a particularly elegant solution)

It should not be necessary, see above.

> > > A similar issue may occur with bash, which doesn't work properly on
> > > IRIX/MIPSpro if compiled above -O1.  I think that an IRIX-specific
> > > 'replace-flags' in the ebuild would resolve this, though.

Yeah.  [[ ${CHOST} == mips-sgi-irix* ]] && replace-flags -O? -O1 should
do the trick.

> > > Infact, bash relies on sys-libs/ncurses, and sys-libs/ncurses-5.6-r2
> > > appears to have CC hard-coded as 'gcc', despite the configure script
> > > correctly detecting 'c99' as the compiler.
> > > (And, after this failure, emerge has hung again - so it does look like
> > > a bug/incompatibility in Portage 2.2.00.8470-prefix.  Additionally,
> > > running 'emerge --version' complains about 'gcc' not being found... is
> > > this a valid test on prefix-portage?)

We really assume a Gentoo toolchain, with exceptions only there where it
must (e.g. if binutils doesn't work, or doesn't produce working code).
So far all platforms use GCC of some form, with Darwin using the Apple
branch of GCC.

> > > With ncurses, work/narrowc/ncurses/Makefile contains BUILD_CC=gcc and
> > > work/narrowc/misc/Makefile contains HOSTCC=gcc - both obviously wrong
> > > <sigh>

Probably need to patch that.

> > > More as I work through it ;)

Keep the work going!


----- End forwarded message -----

-- 
Fabian Groffen
Gentoo on a different level
-- 
gentoo-alt@g.o mailing list