Front page | perl.perl5.porters |
Postings from November 2008
On Tue, Nov 18, 2008 at 01:11:27PM +0000, Ben Morrow wrote:
>
> Quoth Mordechai.T.Abzug@nasa.gov (Morty Abzug):
> > On Mon, Nov 17, 2008 at 10:38:05AM -0800, Chip Salzenberg via RT wrote:
> > > As for restoring the partially read file back to its original name
> > > on error: I actually think that's a bad idea. There may be a LOT of
> > > processing involved, and partial output is definitely better than no
> > > output in some circumstances.
> >
> > In my use case, I was looking to use Perl to in-place edit system
> > configs. The processing is minimal. I would much rather that failure
> > automatically revert the config rather than leave a corrupt config
> > (neither the old nor the new) in place.
>
> perl -pi~ -e'... END { $? and rename "$ARGV~", $ARGV }'
That's a clever idiom that will work in many cases. Kinda ugly, but oh
well. I wonder if we already ensure that $ARGV is undef after the entire
while (<>) {} loop is complete. If not, we should. Otherwise a die or
nonzero exit or setting of $? in another END block could trigger a spurious
rename. If we do appropriately unset $ARGV, then:
perl -pi~ -e'... END { $? and $ARGV and rename "$ARGV~", $ARGV }'
I wonder whether we could make this a standard but optional feature of
inplace editing. How about:
${^INPLACE_SAFE}++; # nop in old Perls, forces rename on error in new Perls
> > And what about the case where one has an error when using -i without a
> > backup [...]
>
> But that's not easy, since the original file has been unlinked and Unix
> doesn't have a 'flink' syscall. [...] Just use a backup if you need to
> recover from errors.
Indeed.
--
Chip Salzenberg twitter:chipsalz
"UTOPIA PLANITIA IS IN ORBIT is the new HAN SHOT FIRST" - Crisper Than Thou
Thread Previous
|
Thread Next