[Rpm-ecosystem] file trigger design: opt out of file list?

[Colin Walters]

Hi, so we just completed an implementation of %transfiletriggerin for rpm-ostree:
https://github.com/projectatomic/rpm-ostree/pull/869

Quick aside: why did we need to do that?  Because rpm-ostree implements "offline"
updates (while still keeping your system running untouched), which really requires
rewriting a lot of what librpm does.    (We also have experimental support for "livefs"
or online updates, which is mostly experimental because we haven't worked through
what the UI/UX is mixing offline/online updates; even live updates are transactional/atomic
in the sense that if they're interrupted in the middle, rebooting will take you back
to the non-live-updated state)

An (IMO) cool example of the value of what rpm-ostree is doing is this:
<https://github.com/projectatomic/rpm-ostree/pull/888>
(We survive a %post that does rm -rf /)
I make no claims that rpm-ostree significantly increases security if you're installing RPMs:
<https://blog.verbum.org/2017/06/12/on-dm-verity-and-operating-systems/>
But our protection against *accidental* damage is quite good; I'm not aware of
another package system doing something like this.

Anyways so on to the topic at hand.  I've been looking at the specs in Fedora using
that same tarball I was looking at in the previous thread, and it looks to me
like the vast majority of triggers don't need the list of files piped on stdin, they
simply walk the filesystem on their own.

Again using `rpm-specs-20170518.tar.xz`:

$ rep transfiletriggerin *.spec
desktop-file-utils.spec:%transfiletriggerin -- %{_datadir}/applications
fontconfig.spec:%transfiletriggerin -- /usr/share/fonts /usr/share/X11/fonts/Type1 /usr/share/X11/fonts/TTF /usr/local/share/fonts
gap.spec:%transfiletriggerin -- %{gapdir}/pkg
gdk-pixbuf2.spec:%transfiletriggerin -- %{_libdir}/gdk-pixbuf-2.0/2.10.0/loaders
glib2.spec:%transfiletriggerin -- %{_libdir}/gio/modules
glib2.spec:%transfiletriggerin -- %{_datadir}/glib-2.0/schemas
gtk2.spec:%transfiletriggerin -- %{_libdir}/gtk-2.0/immodules/ %{_libdir}/gtk-2.0/%{bin_version}/immodules/
gtk3.spec:%transfiletriggerin -- %{_libdir}/gtk-3.0/3.0.0/immodules
man-db.spec:%transfiletriggerin -- %{_mandir}
php-pear.spec:%transfiletriggerin -- %{pecl_xmldir}
shared-mime-info.spec:%transfiletriggerin -- %{_datadir}/mime
texlive.spec:%transfiletriggerin kpathsea -- %{_texdir}
texlive.spec:%transfiletriggerin kpathsea -- %{_texdir}/texmf-dist/fonts/map/dvips/
vagrant.spec:%transfiletriggerin -p %{_bindir}/ruby -- %{dirname:%{vagrant_plugin_spec}}
vim.spec:%transfiletriggerin common -- %{_datadir}/%{name}/vimfiles/doc
$ 

Looking at the code, it looks like only `vagrant` and `php-pear`actually
use rpm's file list.  And while I don't know offhand how many `php-pear` plugins
there are, there are a grand total of 9 vagrant plugins in Fedora today - it'd
be easy to change the vagrant spec to have the tool simply walk the filesystem
itself.

So perhaps something like an opt-in method for triggers to say they don't
need the file list?

`%transfiletriggerin -q -- %{_datadir}/mime`?

(The shared-mime info trigger is one of the bigger ones; we pass over 1000 filenames
 to it that it ignores)