Discerning between a file and a folder in the files signal is impossible?

[purpleidea]

Getting the list of a files in a package is wonderfully useful, although in some packages (such as drbd-utils) on Fedora (an arbitrary example that demonstrates the issue) it lists:

/etc/drbd.d"

as a file, which is in fact a directory.

I'm well aware of the similarity in Linux, but it would be particularly useful for certain tools if there was a way to get more information about the file, such as if in fact it is a directory or not.

Other properties might also be useful information, although I have not found a personal use for this information yet. Getting file owner/group/mode and executable status would probably also be quite handy!

I expect this might involve adding a new method and signal to the API as to not break it.

Thanks in advance!

[purpleidea]

To add a bit more credence to this, if you look at:

http://mirror.its.dal.ca/pub/fedora/linux/releases/23/Everything/x86_64/os/repodata/874f220caf48ccd307c203772c04b8550896c42a25f82b93bd17082d69df80db-filelists.xml.gz

This file contains the file lists for all the packages in a particular repository. The file is huge so, I just took out an excerpt from drbd-utils:

<package pkgid="791ee772612b57978589435ad45f39c65aa7a611df2a51cef7d262b4b61b10fb" name="drbd-utils" arch="x86_64">
  <version epoch="0" ver="8.9.4" rel="1.fc23"/>
  <file>/etc/drbd.conf</file>
  <file>/etc/drbd.d/global_common.conf</file>
  <file>/lib/drbd/drbd</file>
  <file>/lib/drbd/drbdadm-83</file>
  <file>/lib/drbd/drbdadm-84</file>
  <file>/lib/drbd/drbdsetup-83</file>
  <file>/lib/drbd/drbdsetup-84</file>
  <file>/usr/lib/drbd/crm-fence-peer.sh</file>
  <file>/usr/lib/drbd/crm-unfence-peer.sh</file>
  <file>/usr/lib/drbd/notify-emergency-reboot.sh</file>
  <file>/usr/lib/drbd/notify-emergency-shutdown.sh</file>
  <file>/usr/lib/drbd/notify-io-error.sh</file>
  <file>/usr/lib/drbd/notify-out-of-sync.sh</file>
  <file>/usr/lib/drbd/notify-pri-lost-after-sb.sh</file>
  <file>/usr/lib/drbd/notify-pri-lost.sh</file>
  <file>/usr/lib/drbd/notify-pri-on-incon-degr.sh</file>
  <file>/usr/lib/drbd/notify-split-brain.sh</file>
  <file>/usr/lib/drbd/notify.sh</file>
  <file>/usr/lib/drbd/outdate-peer.sh</file>
  <file>/usr/lib/drbd/rhcs_fence</file>
  <file>/usr/lib/drbd/snapshot-resync-target-lvm.sh</file>
  <file>/usr/lib/drbd/stonith_admin-fence-peer.sh</file>
  <file>/usr/lib/drbd/unsnapshot-resync-target-lvm.sh</file>
  <file>/usr/lib/systemd/system/drbd.service</file>
  <file>/usr/lib/tmpfiles.d/drbd.conf</file>
  <file>/usr/sbin/drbd-overview</file>
  <file>/usr/sbin/drbdadm</file>
  <file>/usr/sbin/drbdmeta</file>
  <file>/usr/sbin/drbdsetup</file>
  <file>/usr/share/doc/drbd-utils/COPYING</file>
  <file>/usr/share/doc/drbd-utils/ChangeLog</file>
  <file>/usr/share/doc/drbd-utils/README</file>
  <file>/usr/share/doc/drbd-utils/drbd.conf.example</file>
  <file>/usr/share/man/man5/drbd.conf-8.3.5.gz</file>
  <file>/usr/share/man/man5/drbd.conf-8.4.5.gz</file>
  <file>/usr/share/man/man5/drbd.conf-9.0.5.gz</file>
  <file>/usr/share/man/man5/drbd.conf.5.gz</file>
  <file>/usr/share/man/man8/drbd-8.3.8.gz</file>
  <file>/usr/share/man/man8/drbd-8.4.8.gz</file>
  <file>/usr/share/man/man8/drbd-9.0.8.gz</file>
  <file>/usr/share/man/man8/drbd-overview-9.0.8.gz</file>
  <file>/usr/share/man/man8/drbd-overview.8.gz</file>
  <file>/usr/share/man/man8/drbd.8.gz</file>
  <file>/usr/share/man/man8/drbdadm-8.3.8.gz</file>
  <file>/usr/share/man/man8/drbdadm-8.4.8.gz</file>
  <file>/usr/share/man/man8/drbdadm-9.0.8.gz</file>
  <file>/usr/share/man/man8/drbdadm.8.gz</file>
  <file>/usr/share/man/man8/drbddisk-8.3.8.gz</file>
  <file>/usr/share/man/man8/drbddisk-8.4.8.gz</file>
  <file>/usr/share/man/man8/drbdmeta-8.3.8.gz</file>
  <file>/usr/share/man/man8/drbdmeta-8.4.8.gz</file>
  <file>/usr/share/man/man8/drbdmeta-9.0.8.gz</file>
  <file>/usr/share/man/man8/drbdmeta.8.gz</file>
  <file>/usr/share/man/man8/drbdsetup-8.3.8.gz</file>
  <file>/usr/share/man/man8/drbdsetup-8.4.8.gz</file>
  <file>/usr/share/man/man8/drbdsetup-9.0.8.gz</file>
  <file>/usr/share/man/man8/drbdsetup.8.gz</file>
  <file type="dir">/etc/drbd.d</file>
  <file type="dir">/usr/share/doc/drbd-utils</file>
  <file type="dir">/var/lib/drbd</file>
</package>

As you can see, the "directory" files, have the:

file type="dir"

parts which is very useful to know.

I actually wrote a function to figure out which files must be dirs (because they have other listed files in the contents), but unfortunately some cases such as /var/lib/drbd can't be figured out since there are no files in the dir.

Thanks again!

[ximion]

At least when using the aptcc backend, all directories have a / suffix, which makes it easy to tell whether a file is a file or dir.

[purpleidea]

At least when using the aptcc backend, all directories have a / suffix, which makes it easy to tell whether a file is a file or dir.

That is very funny, because my https://github.com/purpleidea/mgmt/ actually uses the exact same convention to distinguish between a file and a directory. So that is a perfectly acceptable way to represent this. @ximion Very useful information, thanks!

For anyone using my tool under the aptcc backend, will not experience any issues, although I've tried to workaround some of them anyways.

This is an issue with the libhif (dnf) backend AFAICT. Can packagekit add the trailing slashes, since it has this information? This would be a perfectly acceptable way to represent this for me, and wouldn't require any API changes.

[hughsie]

Fwiw, I agree that a trailing backslash should indicate a directory. The issue is sometimes when packages own the files in a directory, and explicitly not the directory itself, e.g. in the bash completion case. If the directory-owned-by-package == trailing backslash is the behaviour you want, that should be pretty easily fixable.

[purpleidea]

On Mon, Mar 7, 2016 at 3:25 AM, Richard Hughes [email protected]
wrote:

Fwiw, I agree that a trailing backslash should indicate a directory. The
issue is sometimes when packages own the files in a directory, and
explicitly not the directory itself, e.g. in the bash completion case. If
the directory-owned-by-package == trailing backslash is the behaviour you
want, that should be pretty easily fixable.

Yeah this sounds great. If the package doesn't own the directory, then it
won't be listed as an independent item anyways. Which is fine :)

So is it fair to count the current behaviour of sometimes not returning the
trailing slashes as a mini bug? (And can you patch this?)

[lukegb]

I looked into this briefly, and tried to work out what all the various pieces are doing:

• PackageKit uses libdnf's dnf_sack_add_repos
• libdnf's dnf_sack_add_repos delegates to dnf_sack_add_repo
• libdnf's dnf_sack_add_repo delegates to dnf_sack_load_repo
• libdnf's dnf_sack_load_repo delegates to load_filelists_cb for loading the filelists
• libdnf's load_filelists_cb delegates to libsolv's repo_add_rpmmd
• libsolv's repo_add_rpmmd parses the filelist and, for STATE_FILE, doesn't even look at the type attribute

As far as I can tell from looking into this (and I could well be wrong), this would mean a patch to libsolv to store information about what type of thing each <file> entry is, and then exposing this somehow in libdnf, so PackageKit can add trailing slashes to things if it deems they need them?