Doesn't contrib/pull complete an unfinished installation?
2011/4/24 <smi...@zenzebra.mv.com>:
> erik quanstrom <quans...@quanstro.net> writes:
>
>>> IIRC it's based on replica(1)
>>> which rather requires ideal circumstances for correct behaviour.
>>
>> partial installs are easy to recover from. i just used it friday to sort
>> out a file system that had a contrib package listed as installed, but it
>> hadn't been.
>
> It *would* be easy if contrib(1) properly indicated the install status
> of a package. As it is right now, contrib/install marks a package as
> installed and THEN goes about installing it. If anything fails, the
> install will still be marked as having been successful. A subsequent
> contrib/install for the same package will snubly claim that the package
> has already been installed and refuse to run. You have to supply the
> '-f' switch to make the install complete. However, doing this requires
> (1) knowing that the install broke, and (2) that all the packages
> dependencies have been met (because -f also overrides dependency
> checking). contrib(1) doesn't provide any way to check if a package is
> up to date, like replica/changes. Having a contrib(1) wrapper for
> replica/changes (say, 'contrib/changes') would be very handy. (And
> simple to write.)
>
> As for prematurely marking packages as successfully installed...
>
> This:
>
> cat > $cfg <<!
> ...
> exec /bin/contrib/pull $name
>
> should probably be replaced with something like:
>
> cat > $cfg.part <<!
> ...
> /bin/contrib/pull $name
> if(~ -$status -) mv $cfg.part $cfg
> if not {
> echo install failed >[1=2]
> rm $cfg.part
> exit 'oh, crap'
> }
>
> --
> +---------------------------------------------------------------+
> |E-Mail: smi...@zenzebra.mv.com PGP key ID: BC549F8B|
> |Fingerprint: 9329 DB4A 30F5 6EDA D2BA 3489 DAB7 555A BC54 9F8B|
> +---------------------------------------------------------------+
>
>
--
- Yaroslav
