[pacman-dev] Huge cleanup of add.c and my attempt at fixing FS#7484

[Dan McGee]

On 7/14/07, Andrew Fyfe <andrew at neptune-one.net> wrote:
> Dan McGee wrote:
> > Ok, I've been slaving away at this for a while and libalpm/add.c is
> > finally getting a bit more manageable. Showing a patch is worthless
> > because it is huge and is much harder to follow due to the refactoring
> > stuff. Instead I've attached a drop-in add.c that compiles cleanly
> > with no other changes needed. I'm looking for some feedback on this.
> >
> > Some highlights:
> > 1. All pactests still pass, so no regressions on that front.
> > 2. *Much* cleaner layout. add_commit has now been split across four
> > functions (upgrade_remove, extract_single_file, commit_single_pkg, and
> > _alpm_add_commit).
> > 3. No more super functions. The heaviest function
> > (extract_single_file) now weighs in at 350 lines, and a lot of that is
> > comments.
> > 4. Most of what Andrew started to do in his 7484 patch has been
> > brought into this. However, I rewrote it a bit for clarity (at least I
> > thought so). We may need a bit more meat added to this section.
> >
> > Comments/concerns/questions appreciated. I still want to cleanup the
> > huge if(needbackup) section a bit, that is still quite long but hard
> > to refactor.
> >
> > -Dan
> >
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > pacman-dev mailing list
> > pacman-dev at archlinux.org
> > http://archlinux.org/mailman/listinfo/pacman-dev
> A HUGE improvement :) The only thing missing is the directory permission
> checks and warning (see the latest patch on my FS7484 branch).

I'm not too keen on having to add that strmode function just for this-
I think we can get away with printing numeric style modes, such as
4777, 0644, etc.

I should add some warning stuff in there though when permissions
differ. Did we make a call yet on whether we should change the
permissions to those of the installing package, or leave it for the
user to do?

> I've got a couple of ideas on how pacman should install a package, your
> break up of add_commit matches up fairly close to what I was planning.
> With the exception of the backup code, I'm thinking that all the
> conflict checks and checks introduced for FS7484 should be done before
> we start adding/removing files from the users system. But this is a post
> 3.1 task :)

Remember that all this conflict checking stuff does in fact get done
before this add_commit code runs, so we could do all this then as
well. It shouldn't be too hard to drop it in, although that will
entail doing a second run though the archive unless we extract once to
/tmp and copy later.

My biggest worry about extracting to /tmp would be filling it up for
those that use a tmpfs and are installing a very large package (some
things in AUR, openoffice, etc).

-Dan