On Mon, Sep 01, 2003, Matthias Kurz wrote:
> [...]
> Solaris SPARC
> -------------
> 8: [...]
> There were some strange warnings, though:
> openpkg-20030830-20030830.sparc64-solaris8-opk.sh: installing into /opkg...
> openpkg-20030830-20030830.sparc64-solaris8-opk.sh: installation done.
> warning: user mlelstv does not exist - using root
> warning: group mlelstv does not exist - using root
> The "user does not exist" - message appears several times.
>
> Later, when packages are added, for every package i see:
> + /opkg/bin/rpm --rebuild /u/openpkg/current/SRC/make-3.80-20030723.src.rpm
> warning: user openpkg does not exist - using root
> warning: group openpkg does not exist - using root
> warning: user rse does not exist - using root
> warning: group rse does not exist - using root
> warning: user rse does not exist - using root
> warning: group rse does not exist - using root
Hmmm... strange. I've to look at this in more detail...
> I also saw something like -L/usr/ucblib -R/usr/ucblib - that is ugly :)
Can you say more exactly on which of the various linker commands during
build?
>
> 2.6: The installation of rpm runs a script "installplatform". This has
> a "#!/usr/bin/env bash" in the beginning. There is no bash on 2.6.
> This did not cause problems on Solaris 8, because there is a
> /bin/bash in the meantime.
> I fixed it by making sure that a bash was found.
> A "regular" solution could change the first line to point to the
> bash that is created during the bootstrap or add the directory where
> this bash resides to $PATH.
We don't need this ugly "installplatform" script in OpenPKG anyway, so
I've disabled its run completely now. This should be fixed now.
> Solaris x86 (8 and 9)
> ---------------------
> Fails in beecrypt.
>
> There is a problem with m4 and aesopt.x86.m4:
> m4:./gas/aesopt.x86.m4:218 more than 4096 chars of argument text
> ifdef(USE_MMX,
> I fixed it by breaking up this section into multiple parts, each starting
> with "ifdef(USE_MMX". The beecrypt author (Bob Deblier) did not like it -
> but this may be my fault, because _i_ called it "ugly" first, when i
> suggested the fix :)
>
> There is a second problem with "mpopt.s". There it is assumed that gas is
> used as assembler. As i told earlier, that this is unusual under Solaris and
> neither the sfw-gcc nor the one from freeware.com does it. And, naturally,
> the one i had compiled for /usr/local also did not use it.
> I think i convinced the author to add an option that disables all assembler
> optimizations. This is always a good thing :) After the bootstrap, GNU m4
> and binutils (gas) can be installed and with a little patch (for ".s.lo:")
> it should be possible to (re-)compile beecrypt with assembler optimizations
> even with a gcc that does not use gas.
> There will probably also be a fix that makes gas unnecessary.
>
> I'll send patches later.
Ok, the "BeeCrypt without assembly stuff" option would be the best final
solution for OpenPKG. In the meantime we can try to workaround it. Just
show us your patches...
Ralf S. Engelschall
[EMAIL PROTECTED]
www.engelschall.com
______________________________________________________________________
The OpenPKG Project www.openpkg.org
Developer Communication List [EMAIL PROTECTED]
