The housekeeping plugin's D-Bus name changed in GNOME 3.24, not 3.26,
which was shipped in Fedora 26. Confusing!

Fallout from de9bb935

The /org/gnome/SettingsDaemon/Housekeeping D-Bus object got moved to
the org.gnome.SettingsDaemon.Housekeeping bus name when the
gnome-settings-daemon plugins were split into their own processes in
GNOME 3.26.

Support for older GNOME versions have been retained to allow new
Flatpaks to run on older host OSes.

Fallout from 23919376

There are no child BaseManagers for the EDIT and PREVIEW modes. Hence
it lead to:
  CRITICAL **: photos_item_manager_get_for_mode: assertion
    'mode != PHOTOS_WINDOW_MODE_PREVIEW' failed

Fallout from d5d28be0

ItemManager::items-changed is only emitted if a BaseItem was added or
removed from the union of all the child BaseManagers (ie. the 0th
child). Therefore, due to a mode change, if the current child is
cleared and populated with new items, the addition of new items won't
lead to ItemManager::items-changed if they were already present in one
of the other children. In that case, various GSimpleActions that were
disabled by the clearing of the child BaseManager won't be re-enabled
after the addition of new items.

This is easily observed while searching because entering SEARCH clears
the corresponding BaseManager. eg., if the first 50 items in OVERVIEW
are all LocalItems, and searching for local content would lead to the
same 50 LocalItems to be shown in SEARCH.

Single precision floating point is all that's needed. It is wasteful to
promote calculations to double precision through the use of such
literals. This can have significant performance impact on architectures
like ARM.

The brightness and contrast of the green and blue channels were not
being adjusted for 8 bits per channel inputs.

Single precision floating point is all that's needed. It is wasteful to
promote calculations to double precision through the use of such
literals. This can have significant performance impact on architectures
like ARM.

It broke the string freeze. Since this isn't part of the actual
bug-fix, it's better to just back it out.

This reverts commit c0b6077a.

There is no guarantee that the GtkRevealer will asynchronously animate
before emitting notify::child-revealed. There is no known breakage,
since there is no LoadMoreButton anymore. This is a pre-emptive fix.

Fallout from 470c05cd

https://bugzilla.gnome.org/show_bug.cgi?id=785376

There is no guarantee that the GtkRevealer will asynchronously animate
before emitting notify::child-revealed. When it doesn't, the SpinnerBox
remains visible, and obstructs the hyperlink in EmptyResultsBox from
receiving the mouse event, or interferes with ToolCrop's GdkCursor.

Fallout from 73e32473

https://bugzilla.gnome.org/show_bug.cgi?id=785376

This will make it easier to develop and test the "empty state" logic.

https://bugzilla.gnome.org/show_bug.cgi?id=785376

Just because photos_utils_file_query_info has succeeded it doesn't mean
that a thumbnail is present. If it's absent, the path is NULL and it
led to:
  GLib-GIO-CRITICAL **: g_file_new_for_path: assertion 'path != NULL'
    failed

Wget reports several TLS errors with exiv2.org:
  Resolving exiv2.org (exiv2.org)... 54.72.225.196
  Connecting to exiv2.org (exiv2.org)|54.72.225.196|:443... connected.
  GnuTLS: A TLS warning alert has been received.
  GnuTLS: received alert [112]: The server name sent was not recognized
  ERROR: The certificate of ‘exiv2.org’ is not trusted.
  ERROR: The certificate of ‘exiv2.org’ hasn't got a known issuer.
  ERROR: The certificate of ‘exiv2.org’ was signed using an insecure
    algorithm.
  The certificate's owner does not match hostname ‘exiv2.org’

Hence it still uses plain HTTP.