[pacman-dev] [PATCH] Allow package to display a brief message before sync install

[Marc - A. Dahlhaus [ Administration | Westermann GmbH ]]

Am Dienstag, den 15.09.2009, 22:53 +1000 schrieb Allan McRae:
> Nagy Gabor wrote:
> > 2009. 09. 15, kedd keltezéssel 21.18-kor James Rayner ezt írta:
> >   
> >> On Tue, Sep 15, 2009 at 9:14 PM, James Rayner <iphitus at iphitus.org> wrote:
> >>     
> >>> On Tue, Sep 15, 2009 at 11:20 AM, Dan McGee <dpmcgee at gmail.com> wrote:
> >>>       
> >>>> I don't really know what to think here. I had looked at that messages
> >>>> one for a long time and thought it was a decent idea, but never went
> >>>> far enough to take it and run with it.
> >>>>
> >>>> @Loui- sure, but this is for extraordinary messages- a lot more
> >>>> exclusive than ChangeLog-worthy stuff, and you have to explicitly
> >>>> request to see that anyway.
> >>>>
> >>>> @Jeff- it isn't exactly straightforward to view an install script
> >>>> beforehand, and the post_install business is a rather hacky reason for
> >>>> needing an install script.
> >>>>
> >>>> -Dan
> >>>>
> >>>>         
> >>> Dan's got the idea...
> >>>
> >>> pacman should not break someone's system without at least telling them
> >>> first. So yes - this is intended for more extraordinary messages.
> >>>
> >>> The current ways of informing the user (homepage/forum news and
> >>> post-install) are broken and non-simple:
> >>>  - both polling based
> >>>       
> >> oh, and post-install is after the fact - when the system is broken, so
> >> it's not a very good way of informing the user that their system "will
> >> break" because it's already broken.
> >>
> >> Anyway, I'm all for a more generalised/ideal setup, but that's been
> >> wanted for a while with no patches coming forward.
> >>
> >>     
> >
> > OK. Here is my staindpont (not closely related to iphitus's patch, but
> > some thoughts about the "problem"):
> >
> > 1. echo lines in install scriplets are stupid. I bet that you also
> > looked into install scriplets in /var/lib/pacman/... many times manually
> > to read that information on an installed package (when something went
> > wrong). I think this requires a new %INFO% field in (local) database,
> > which could be accessed by -Q. Drawback: pre_install, post_install,
> > pre_upgrade etc. is more sophisticated. (It is possible to only print
> > info if we upgrade version older than...)
> > 2. I am not sure about the pre-transaction messages. We ask for user
> > confirmation before downloading packages, so in order to print
> > info/alarm etc. messages then, we _must_ store this info in sync
> > database, or interrupt the transaction once more before actual install.
> > post-transaction messages are easier to implement, see 1. Iphitus
> > chooses putting %ALERT% to syncdb.
> >
> > Overall, I think iphitus's patch is a good compromise, if we want to
> > distinguish important and non-important messages.
> >
> > My problem is that I don't see when the packager should remove %ALERT%
> > from package, in 1.0-2, 1.1-2, 2.0-1? When I've read (and understood)
> > the alert message, printing it again is just a spam.
> >   
> 
> That point is what I have been thinking about all day but there is no 
> easy solution there as far as I can tell.  I have the same issue with 
> deciding when to remove provides lines...
> 
> Allan

How about detaching the alerts from the actual package itself and add it
only to the repo like we do for deltas? We then could add a logic to it
to just print it for versions older than the flagged version by
default...

This of course needs some new tools for the repo to add/modify/remove
alerts.

I think this could be a way around this problem...

Marc