Re: Bug#619275: Perl Policy change to document major version upgrade trigger
On Fri, Mar 25, 2011 at 12:38:19PM +0200, Niko Tyni wrote:
> While I do think this is a nice solution, I've got a couple of concerns:
> - is this overkill? Would it be enough for the long running daemons to just
> register an interest in a file trigger on /usr/bin/perl ? This means
> minor perl upgrades will activate the trigger too, but that may well
> be a good thing - think of security fixes and the like. (OTOH, this
> approach doesn't help daemons embedding libperl...)
As a consumer of the interface (ie the package wanting to be triggered)
I would prefer an explicit declaration of intent for the trigger, rather
than just watching /usr/bin/perl, for the specific instances where I can
know (or have a good hunch) that my software will break. The more
general problem of ensuring that all changes are reflected in running
processes is not something that I think we can tackle alone (it's not at
all perl-specific, unlike the problem this change is intended to solve).
The principle of least surprise probably applies here too. We shouldn't
(in an adhoc way, without distribution-wide take-up) be encouraging
daemons to stop and start when they would actually carry on working
without that action.
I would love to see some more general project infrastructure for ensuring
that the right daemons are given the chance to be restarted after *any*
upgrade of packages they depend on, but that's a separate issue.
> - is it too early to put this in policy? Generally policy documents existing
> practice, but no package is using this yet. Should we wait for at least some
> level of adoption, probably by filing wishlist bugs on known affected daemons
> like spamassassin, and see how it works out first?
Yes, absolutely. I was anticipating having already filed a wishlist
bug at this point, but I was away from my main development system
which had my spamassassin patch on it, so ended up filing this bug
first :) Although I'd now like to be confident that we're doing the
right thing before filing bugs on other packages, so will wait for the
above discussion to pan out a bit first.
One of the reasons I worked on this was because I've been hit by
exactly this issue with spamassassin in the past; I'd appreciate
pointers for other software that would benefit from this change.
Dominic Hargreaves | http://www.larted.org.uk/~dom/
PGP key 5178E2A5 from the.earth.li (keyserver,web,email)