Skip to Content.
Sympa Menu

sm-discuss - Re: [SM-Discuss] Bug 16031 - Cannot upgrade from stable 0.62-10 to test 0.63 due to gmp 4.3.2 -> 6.1.2 update

sm-discuss AT lists.ibiblio.org

Subject: Public SourceMage Discussion List

List archive

Chronological Thread  
  • From: Remko van der Vossen <wich AT yuugen.jp>
  • To: Eric Sandall <eric AT sandall.us>
  • Cc: sm-discuss AT lists.ibiblio.org
  • Subject: Re: [SM-Discuss] Bug 16031 - Cannot upgrade from stable 0.62-10 to test 0.63 due to gmp 4.3.2 -> 6.1.2 update
  • Date: Sat, 5 Aug 2017 09:51:16 +0200

On Fri, Aug 04, 2017 at 06:47:03PM -0700, Eric Sandall wrote:
> On 8/4/17 5:45 AM, Thomas Orgis wrote:
> > Am Thu, 3 Aug 2017 18:08:41 -0700
> > schrieb Eric Sandall <eric AT sandall.us>:
> >
> >> How did those of you on test grimoire get past Bug 16031-
> >> <https://bugs.sourcemage.org/show_bug.cgi?id=16031><https://bugs.sourcemage.org/show_bug.cgi?id=16031>Cannot
> >> upgrade from stable 0.62-10 to test 0.63 due to gmp 4.3.2 -> 6.1.2
> >> update <https://bugs.sourcemage.org/show_bug.cgi?id=16031> ?
> > Thanks for pointing that out. Glad that I didn't jump into that yet. As
> > discussed recently with the ncurses6 jump, we really, really need at
> > least library preservation for these cases. Some sorcery support to
> > remove the respective library files from the install log of gmp/ncurses
> > before dispel and put them into a list. Or create local spells for the
> > vintage version to dispel after all dependers got rebuild.
> >
> > Any ABI break like that needs to be handled. Currently, there is only
> > manual workarounds, like Pol did with ncurses locally.
>
> We should revert 7ca504a370305392ee739b9586c7ef9ad57b1a88 until a tested
> fix is in place. At no point should upgrades need manual intervention
> nor break a box because the breakage is unknown.
>
> I tried doing a `/sorcery hold gmp/` (stable Sorcery), but even then a
> `/cast --queue/` (with gmp removed) still triggered gmp to be cast and
> broke my VM (huzzah for snapshots ;)). It seems Sorcery no longer holds
> spells, though this worked in the past with some of the gcc updates I
> put off for a bit.

For me a simple ln -sf libgmp.so.10 /lib/libgmp.so.3 solved the issue
when testing a stable to test upgrade. But still we need a general
solution for this as Thomas mentioned.

In my opinion we need to be able to "split" a spell into a reduced
original and a preservation spell that "owns" the legacy shared objects.
The preservation spell should have a mechanism to prevent dispelling
until all dependees on the old shared object have been upgraded or
removed.

Regards, Remko




Archive powered by MHonArc 2.6.24.

Top of Page