The looming Python 3(000) monster

[James Antill]

On Sat, 2008-12-06 at 09:29 +0000, Richard Hughes wrote:
> On Fri, 2008-12-05 at 14:07 -0900, Jeff Spaleta wrote:
> > Do not let rpm transactions fail in the middle. Do not forcibly
> > terminate them...do not cut power to the system...hold your breath and
> > walk away from the console.
> 
> This was one of the problems PackageKit tries to solve.

 In it's usual way, of ignoring the real (but harder to solve) problems
and concentrating on a couple of very simple desktop cases and
implementing the most complicated solution possible ... and then
declaring the problem fixed, after minimal testing.

>  You can start an
> install and do ctrl-alt-backspace and then re-login, and the transaction
> is still running in the background.

 rpm eats SIGHUP when in a transaction, so yum/rpm should do this too.
However rpm had a "feature" where it would abort if the stdout yum was
using went away, which was ... unexpected.
 rpm in Fedora 10 fixed that (not sure which older releases), and
yum-3.2.20+ tries to work around it.

>  It also stops you shutting down
> using a system and session inhibit if the transaction is an a
> non-cancellable state.

 Which isn't really helpful, esp. as the messages I've seen say
something like "Some program is stopping you from doing this" ... pretty
much guaranteeing the next words out of the users mouth will be "die
when I tell you too, scum sucking laptop, die; die; die!".

> If you /sbin/shutdown as root, or suffer a power failure without a UPS,
> then all bets are off :-)

 Or press the power button, or rpm gets OOMed, etc. etc.
 Which is to say it works about as well as yum itself.

-- 
James Antill <james at fedoraproject.org>
Fedora