Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

replaced with gdk_cursor_new_for_display()

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Ideally we would handle this differently. See the TODO entry in the render
function. But the patch is a start.

Thanks to carlos for the patch.

Closes: #398

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Also emit the D-Bus signal

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Thanks to MaximeC for the initial patch.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

They are not used anywere, and do not seem to become useful anymore in
the light of the new separation of document and view.

Otherwise the document object gets updated with wrong position values,
because the adjustment belong to the index, not the document view.

The adjustment callbacks act as an interface between position data in
the document object, and the adjustments.

We remove the horizontal centering code, as now it is done by
position_set. Those callbacks should not change the position read from
the document object in any way.

Also, we split the adjustment_value_changed callback into a vertical and
an horizontal version. Previously a single callback was reused for both,
horizontal and vertical. That lead to a subtle problem when coming out
of index mode. What happened was the following:

  1. horizontal adjustment bounds change coming out of index mode. This
     triggers an hadjustment changed signal.

  2. the hadjustment_changed callback handles it, and resets the
     hadjustment value, as the bound may have changed. This triggers a
     value_changed event.

  3. the value_changed callback handles the event, and captures the
     position for *BOTH*, horizontal and vertical adjustments, saving
     them to the document object.

  1..3 is repeated for the vertical adjustment.

  Now, if in 3. the horizontal adjustment bounds were not yet updated
  after the index mode, we got ourselves at the wrong vertical position.

This race condition is avoided now because both value_changed callbacks
*ONLY* handle their own direction, either vertical or horizontal, not
 both.

This new function adjust_view is in charge of recomputing the scale
according to adjustment settings and trigger a render_all.

adjust_view contains the old sc_adjust_window code, slightly simplified
thanks to the availability of the document_get_viewport_size.

Then it is used by sc_adjust_window, document_open and the
cb_view_resized callback. Makes slightly more sense this way than
calling the shortcut sc_adjust_window directly.

Now we can trigger a gtk page refresh calling refresh_view. This
function triggers a custom signal refresh-view, whose handler copies the
position from the document object to the adjustments.

All of those callbacks are conceptually related (change the page
layout), and depend from one another.

Now the single callback page_layout_value_changed defers to
page_widget_set_mode to change whatever is needed in the GTK widgets.

  - fix the computation of the number of rows in the table. Doing a ceil
    of an integer division has no effect...

  - set the page_padding in page_widget_set_mode function, instead of
    doing it independently.

  - call zathura_document_set_layout after calling page_widget_set_mode
    to save the page layout settings into the document object.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

This should fix #347.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Add two more signals to ZathuraRenderRequest to notify pages if they are cached
or not. This allows us to move some logic away from
cb_view_vadjutment_value_changed to more appropriate places.

ZathuraPageWidget will now release the surface if
* it gets the signal that the page is no longer cached and the page is
  invisible,
* the page is not cached and the render request is aborted.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

The page widget shouldn't have to care what should be done with selected images
and text.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Marwan Tanager authored Jun 21, 2013 and Sebastian Ramacher committed Jun 21, 2013

Move calls to zathura_jumplist_add surrounding zathura_link_evaluate to zathura_link_evaluate itself

This is just in case that zathura_link_evaluate returns without evaluating the
link, which would result in a new jump being added for the current position,
but without adding a corresponding jump for the target position (actually a
second one would be added but the check in zathura_jumplist_add would prevent
it from being appended on the linked list since it's position would be the same
as the current one, because the link haven't been evaluated).

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Marwan Tanager authored Jun 09, 2013 and Sebastian Ramacher committed Jun 10, 2013

	- Don't delete the elements on the right of the current one, when
	  appending a new jump to the jumplist, because this makes no sense at
	  all; the point of the jumplist in the first place is to remember
	  previously jumped-to positions in the document, so there is no need
	  to delete anythings except to trim the oldest entries from the
	  beginning to maintain the maximum size. This also makes us compatible
	  with the Vim way of doing things.

	- Make the jumplist mechanism functional on the same page; if we
	  followed a link to a target on the same page, remember the
	  adjustments before and after following the link. The same holds for
	  navigating search results on the same page.

	- Implement position_set and use it instead of position_set_delayed
	  when following links in order to give zathura_jumplist_save a chance
	  to record the exact adjustments of the link target. Otherwise, it
	  will always record the adjustments after going to the target page,
	  but before going to the exact position within it.

	- Don't consider movements with ^i and ^o as jumps :)

	- Don't use page_set followed by setting the adjustments in
	  sc_jumplist, because this is redundant and causes clutter when using
	  ^i and ^o, as the adjustments is set twice this way (once in page_set
	  and again in position_set_delayed).  It's enough to only update the
	  page number on the statusbar and then set the adjustments.

	- Hide implementation details (zathura_jumplist_save and
	  zathura_jumplist_append), and make things more consistent by
	  exporting and using only zathura_jumplist_add for adding new entries.

The end result: A more slick jumping experience :-)

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Marwan Tanager authored Jun 08, 2013 and Sebastian Ramacher committed Jun 10, 2013

This patch activates the last aborted search when pressing the search shortcuts
('n' or 'N').

To avoid confusion, and to make things more predictable, I've chosen to always
reactivate an aborted search starting from the beginning (or end, in case of
'N' or '?') of the current page, as opposed to Vim which continues from the
next search term each time the search is reactivated.

Searching using '/' or '?' doesn't center the view at the current search term
like when using 'n' or 'N', so we fix this here.

Also, I managed to work around the issue of the thin rectangular margins that
show around the previously-highlighted search terms after the search is aborted
(either explicitly or as a result of following links), by redrawing the page
widget (only if it's visible) instead of redrawing the rectangles covering the
highlighted search terms.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Marwan Tanager authored Mar 27, 2013 and Sebastian Ramacher committed Mar 29, 2013

This also has the side effect of not flooding the console with debug messages
on every change of the vertical adjustment as a result of redundant cache
lookups.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>

Marwan Tanager authored Mar 26, 2013 and Sebastian Ramacher committed Mar 26, 2013

This patch implements a page cache that is invalidated in a LRU fashion.

Pages are added to the cache as soon as they become visible. When the cache is
full and a new page that isn't in the cache becomes visible, the least recently
viewed page in the cache is evicted from memory and the new one takes it's
place.

The cache size is configurable using the page-cache-size configuration
variable, with a default value of 15 pages. Very large values for the cache
size are not recommended, though, as it will stress the system memory out.

The old periodic page reclaiming code is no longer necessary with this patch,
so I removed it.

Special thanks to Ignas Anikevičius, and Sebastian Ramacher for the
inspirations.

Signed-off-by: Sebastian Ramacher <sebastian+dev@ramacher.at>