[pacman-dev] Scripts bombing with no args

[Dave Reisner]

On Sat, Jul 30, 2011 at 10:58:46AM +1000, Allan McRae wrote:
> On 29/07/11 22:38, Dave Reisner wrote:
> >On Fri, Jul 29, 2011 at 07:01:29PM +1000, Allan McRae wrote:
> >>On 29/07/11 03:14, Dave Reisner wrote:
> >>>On Thu, Jul 28, 2011 at 11:18:15AM -0500, Dan McGee wrote:
> >>>>Allan or Dave, mind taking a look at this? We should show usage info,
> >>>>as rankmirrors and pacman-key do without args, instead of this silly
> >>>>behavior.
> >>>>
> >>>>-Dan
> >>>>
> >>>>dmcgee at galway ~/projects/pacman (master)
> >>>>$ ./scripts/repo-add
> >>>>==>   ERROR: '' does not have a valid archive extension.
> >>>
> >>>We never check for $1 being unset before proceeding. Probably my
> >>>mistake when I did some refactoring of repo-add.
> >>>
> >>>>dmcgee at galway ~/projects/pacman (master)
> >>>>$ ./scripts/pkgdelta
> >>>>./scripts/pkgdelta: line 147: $1: unbound variable
> >>>>
> >>>
> >>>'set -u' with 'set -e'? Ugly...
> >>
> >>Care to explain why?  Using this in our cleanup scripts would have
> >>prevented the Arch repos disappearing on more than one occasion....
> >>
> >
> >Because the combination of these two flags causes scripts to fail in
> >mysterious and unexpected ways. Also, it won't save you from set but
> >empty vars, which I suspect would be just as dangerous as unset vars in
> >the situation you allude to. Just validate the user inputs. It makes
> >things more obvious to an outside reader of the code, as well as
> >consumers of the script.
> >
> 
> FYI, this is the story I am referring to:
> 
> http://archlinux.me/brain0/2009/08/16/shit-happens-when-you-party-naked-or-use-crappy-shell-scripts/
> 
> I'm still a fan of keeping set -u and -e...  It may not save you
> from everything, but it does stop some things.
> 
> Allan
> 

I think that story only reinforces my point that you should be
validating what you take in as input. Should a script bomb out in the
middle of operation due to an unbound var triggering errexit, you may be
left in some unknown or inconsistant state. It then means you have to
add an ERR trap, figure out where you are, and take the appropriate
action, if any, to restore sanity. It's unwieldy and vague, and explicit
validation at key checkpoints will always be a more robust (and more
readable) solution.

I'll leave it up to you to decide which direction to go in. We really
just wanted to fix the lack of usage on bad command line args. imo,
pkgdelta needs a lot more help than just validation of input, but that's
for another patch(set).

dave