It's broken, see:

https://bugs.webkit.org/show_bug.cgi?id=147822#c3

I realized I was being stupid, and have corrected this infraction.

This is not tested at all since I don't want to context switch to my
laptop. It probably works. Users should complain if not.

Just like we do notification permission requests

This is more scalable and is needed to avoid code duplication in
EphyWebView

Previously, deleting your history didn't even cause the size of the
history database to shrink on disk. There's only one proper way to
delete an SQLite database, and that's to remove it from the filesystem.

This reverts commit 34a55be4.

This reverts commit cbd0f2b4.

This reverts commit ed8685d3.

Really not sure whether we want to do this or not.

We forgot to link to nettle and hogweed. It only worked because WebKit
was pulling them in via GnuTLS, but WebKit has now switched to gcrypt.

Also, use the pkg-config variables in the proper place in tests.

https://bugzilla.gnome.org/show_bug.cgi?id=773830

We should not listen to changes in the username input field while editing
its contents by using webkit_dom_html_input_element_set_editing_values()
under the risk of ending up recursively calling fill_form_cbi() forever.

Ideally WebKitGtk+ should provide a mechanism similar to
g_signal_handlers_block_by_func() to address this kind of issues. We could
surround the set_editing_values() call by a pair of
webkit_dom_event_target_remove_event_listener() and
webkit_dom_event_target_add_event_listener() but that would force us to
somehow pass the webpage to fill_form_cb, and what's worst, duplicate the
multiple credentials detection code.

https://bugzilla.gnome.org/show_bug.cgi?id=773830

Tests don't run the embed shell application instance.

Hence, this is a great place to create the UriTester. Right before the
D-Bus server is created.

Speculative fix for a reported crasher

It's not needed anymore after the EphyUriTester rework, I just failed to
notice.

This is much less efficient, but it also makes the code much simpler and
more robust. Using EphyBookmarksManager as the list model was a good
choice at the time, when we were not thinking about smart bookmarks, but
now that we need to handle both it doesn't work so well. My initial
implementation was complex and buggy. Simplify this.

I broke this when trying to filter smart bookmarks out of the model.

The container owns the reference; this fixes a memory corruption issue.

We need to ref all the bookmarks that we put into this container. It
wasn't crashing because we accidentally leaked the container.

There's no public API anymore and no extensions. Having private headers
years later serves no purpose and is just confusing.