Subject: Re: RFC: migration to a fully dynamically linked system
To: None <martin@duskware.de>
From: None <cgd@broadcom.com>
List: tech-userlevel
Date: 12/21/2001 10:41:49
At Fri, 21 Dec 2001 10:17:42 +0000 (UTC), "Martin Husemann" wrote:
> When I was hit by the -current sparc ld.elf_so bug (i.e. everything linked
> dynamically suddenly stopped working after installing new binaries), I could
> bring my DSL router (which happens to be the only sparc system around) [...]


Hmm, so, what morals should we take from this story?

I'd say they should go something like:

* Do not run/develop -current on systems which you depend on for
  production use.

  How is this any different than, say, some random VFS bug that
  happened to trash your FS?

* If you really must do so, try to stick to more tested versions
  (snapshots) which theoretically shouldn't have _such_ catastrophic
  problems.

  If necessary, test the code elsewhere first before migrating it to
  production systems.

* Before upgrading _any_ system whose data you care about, make sure
  you have a working backup.

* Do definitely provide a useful set of statically-linked
  system-restore utilities, if you're providing a
  fully-dynamically-linked system.



cgd