When blinking the caret, it looks like a cool idea to only update the
small rectangle containing the caret. Actually it is an awful idea, since
the paint event will paint some rows, and these rows will not be
properly painted outside of the small rectangle. Unfortunately, the
painter will skip those "painted" rows on next paint events.
This leads to painting errors that depend on the ordering of "real"
and "caret" paint events. This is the reason why they only appeared
with split screens.
Quote of the day: ``The real problem is that programmers have spent far
too much time worrying about efficiency in the wrong places and at the
wrong times; premature optimization is the root of all evil (or at
least most of it) in programming.''
-- Donald Knuth, The Art of Computer Programming.
Fixes bug #11354.
If we were not ready to paint the screen, this does not mean that we
should give up on repainting, just potpone it.
I thought that it was bad to call update() in the paint event, but I
cannot find reference to this anymore and everything seems to work as
intended.
We do not know what to do with horizontal scrolling events, so we ignore them.
Note that the code has to be different between Qt4 and Qt5.
Fixes bug #11257.
It is a general problem when doing graphics to know where a line
begins and where it ends pixel-wise. At the instigation of Scott, and
with the use of the kmag magnifier, this commit corrects 3 areas:
* foreign marks were larger than the row element they were supposed to
mark. This could lead to moving lines, depending on paint ordering.
* visible spaces were drawn outside of their box (select a single
space to see this).
* the `L' blinking caret would leave a cursor dropping because the
horizontal part was too wide.
The goal of this patch is to make "self-insert <string>" act as a
sequence of individual self-insert calls. Once it is done, the
insertion of the commit string in GuiWorkArea can use that, which
fixes the issues described in bug 11183.
1/ in InsetMathNest, SELF_INSERT is rewritten to be a proper loop of
interpretChar, instead of inserting the string as characters. This
solves the issue of interpreting ^ properly.
2/ the text version does not need to be changed (the loop is rewritten
using a range, though).
3/ Then {{{GuiWorkArea::inputMethodEvent}}} is changed to invoke
directly self-insert. This removes the update issues.
Fixes bug 11183.
This is detected when an undo group is open and contains at east one
element. This means indeed that changes are in progress. Note that the
group is in general opened in GuiApplication::dispatch. The code there
is changed to ensure that the group is closed before updating the
screen.
This patch is experimental. It is expected to be replaced in master by
a more complete solution. It could in the meantime be backported to 2.3.x.
Fixes bug #11159.
The problem is that popping dialogs during reload can cause paint
events for which we are not ready. If this does not work, then we
can introduce a new flag, besides 'busy', for that case. But busy
does not seem to be used very widely, so hopefully this works.
A brand new event compressor based on Kuba Ober's cleverly simple
solution: <https://stackoverflow.com/a/21006207>.
Fix#9362, #9461, #9933: Lyx suddenly gets keyboard keys wrong, and
deadlocks
Fix#9790: LyX should perform key event compression (for improving the
remote X connections one would also need to implement
Qt::WA_KeyCompression)
Fix#10516: slowness on repeated arrow keys with IBus and Qt5
Patch pulled from
https://github.com/gadmm/lyx-unstable/commit/bf5a1efb0db5bfc2b
Signed-off-by: Juergen Spitzmueller <spitz@lyx.org>
Qt on macOS does not respect the Qt::WA_OpaquePaintEvent attribute and
clears the widget backing store at each update. Therefore, we use our
own backing store in this case.
This restores a simplified version of the code that was removed at 24c29908.
This replaces a showCaret/hideCaret pair and avoids an update.
Also remove an update() call in resizeBufferView: is is not necessary
since we are already in a pintEvent handler.
This avoids endless resize issues on HiDPI systems (e.g. Retina Mac).
Rename pixel_ratio_ to last_pixel_ratio_ to emphasize that this is a
cached value.
Inline needResize method to make the logic clearer in paintEvent.
Remove the old schedule_redraw_ mechanism that was only useful because
of our synchronous drawing code. Now that actual painting is
scheduled instead of forced, it becomes pointless.
Rename WorkArea::redraw(bool) to scheduleRedraw(bool), to show that
the drawing is not done right away.
In GuiView::updateInset, call scheduleRedraw(true), so that metrics
are correctly computed (this was the whole point of the exercise).
This unbreaks input methods by splitting the part of the code that
does the actual drawing to a separate paintPreeditText() method which
is called from paintEvent().
The proper solution would have been to introduce the preedit string in
the Row object, like is done for completion, but this is too complex
to do at this point.
The only change in behavior is that now the commit string is inserted
in one fell swoop, intead of emulating a number of key events.
Rename cursor to caret to in order to avoid ambiguity. The caret is
now the blinking thing only.
Remove unused header contents, and some not so useful methods.
No intended change of behavior.
The trick is to remember in BufferView what has been done at the
previous draw, so that the row that contained the caret can be
repainted if needed.
To this end, add an argument paint_caret to BufferView, although
painting the caret is not the job of the BufferView (at this point).
BufferView::needRepaint will act as an interface with
TextMetrics::drawParagraph to know whether the painting of a given
row should be forced.
Currently everything is done at the top row level, so that, if the
caret is in a large table, the whole table will have to be repainted.
It is not clear yet that this is necessary.
Historically, because of two-stage drawing, LyX has been painting on a
Pixmap, and this pixmap is copied to screen at paint event time.
Now that we have three-stage drawing, it is possible to delay the
painting to actual paint event and avoid the intermediate Pixmap.
Known bug: the cursor is never erased.
Normally the two stages of drawing are
1/ compute metrics of insets/rows/paragraphs/mathrow...
2/ draw the elements and cache their positions
Now the three stages are
1/ metrics
2/ nodraw: do not draw the elements, but cache their position
3/ draw the elements (and store again their position; it does not
seems to hurt performance).
Revive the NullPainter: this replaces the setDrawingEnabled mechanism
with a painter that does nothing. The advantage is that updatePosCache
(renamed from setPosCache) does not need anymore to be invoked from
the frontend.
updatePosCache (the nodraw stage) is called at the end of
BufferView::updateMetrics.
Allow a LaTeX backend to run external commands after user confirmation.
This is a per document and per machine setting. The authorization has
to be given through the document settings pane, but is not recorded in
the document itself. Moving the document to either another computer or
another directory on the same computer revokes the authorization.
This can also be done by right clicking the red icon that appears in
the status bar when a document is marked as one requiring shell escape.
The patch also checks whether the user has added the -shell-escape
option to a LaTeX converter and nags the user to remove the option
(which would be used for all documents) in favor of the (per document)
support offered by LyX.
QSignalBlocker in updateScrollbar is too strong and prevents the scroll bar from
communicating with its scroll area. The only solution to block signals between
specifically between two objects is to disconnect. This makes sense in this
case, by making updateScrollbar responsible for managing the connection in the
first place.
This is a consequence of e91572a0, although it is not clear why. The
contextMenuEvent code of the workarea should not delegate context
menu when it cannot honor it.
Also fix a small memory leak.
Fixes bug #10616.
In some (not yet understood) situations, the paragraph metrics cache
is empty in generateSyntheticMouseEvent(). We just avoid a crash in
this case, but there is probably an underlying problem that deserves
being fixed.
Fixes bug #10324.
Starting at 61b2bd5e, boost::bind was progressively replaced with
std::bind. They are not interchangeable though. boost::bind implements
the tracking of boost::signals{,2}::trackable objects. Now that
std::bind has completely replaced boost::bind, tracking never occurred.
This commit replaces boost::signals2::trackable with the new preferred
boost::signals2 methods: scoped_connections or slot::track_foreign. The
support::Trackable class introduced is less safe but easier for transitioning
old code.
Fixes the crash at #8261.
- Adds a currentZoom variable which holds the current zoom level.
- The zoom stored in preferences is used as default zoom level (default binding:
M+0).
- The currentZoom is saved and restored via QSettings.
- Adds LFUN buffer-zoom for (re)setting zoom.
Introduce a distinction between being read-only and having the read-only flag. A
buffer is read-only if either it has been externally modified or if it has the
read-only flag.
Instead of using a coverity annotation (that does not work) it is
better to really iinitialize the members of the object, just in case
somebody decides to actually use the DisplayPath() constructor.
Starting with Qt 5.6, setting the environment variable QT_SCALE_FACTOR
makes everything accordingly bigger. So, if QT_SCALE_FACTOR=1.2, all
text and GUI elements are rendered 20% bigger. However, if an application
does not account for this, everything will also look "blocky".
With this commit, all text and images will be scaled remaining sharp.
This works whether a HiDpi screen is used or not, but is mostly useful
with a HiDpi screen, as all GUI elements are more spaced apart and one
can use the mouse for selecting things without requiring a high precision.
It turns out that Cursor::getFont() does not return a fully realized
font in some cases. This patch uses real_current_font instead, which
seems better anyway.
I suspect that all uses of getFont() should be removed, but this is
not code that I know well.
Fixes bug #10478.
This should avoid performance problems related to the window update machinery.
Moreover this fixes a crash introduced by 82808fea when closing a file.
Note that GuiWorkArea::Private already had a read_only_ member, but it
was unused.
Also rename LyXVC::vcname() to LyXVC::vcstatus() since it now contains
directly the UI string to be shown.
