Inside a math inset when completing macro names, it could lead to crashes.
Note that this processUpdateFlags is present when outside of this if() branch.
When inserting inset over a selection, the layout is transferred if
the inset covers full paragraphs.
This is not the right policy in the case of environments. Until proper
code is written and tested, limit ourselves to command layouts.
Fixes bug #12251.
Typical example is a Labeling layout which label is an inline equation
larger than the screen. Before this commit, the row would not get
broken at all.
Two parts in the patch:
1/ when breaking the row in shortenIfNeeded, mark the last element as
AlwaysBreakAfter instead of BreakAfter, in case the next element is
NoBreakBefore.
2/ when nothing could be done, as last resort keep the first element
of the row only.
Prevent the return of impossible values for 'enum flags'.
For instance the result of '~Update::Force' at src/BufferView.cpp:3025 without this patch
is '4294967293'.
The string that Qt sees is not the original string, since we add
zero-width nonbreaking spaces around it, plus an override character to
force rendering direction. This means that translating positions
between these two representation can be a bit tricky.
We use a max(foo, 0) here to make sure that our intermediate
computation is postive.
Fixes part of bug #12452.
Setting sufficiently high value allows to use
ColorCode enums for new up to 32000 branch insets.
Previous setting printed messages if using more than 25 new branch insets.
"'ColorCode' ... src/Color.cpp:435:10: runtime error:
load of value 128, which is not a valid value for type"
It is not a good idea to take into account the horizontal scrolling
that may have happened. For example, this leads to display glitches
when a Description label is larger than text width.
This explains why SingleParUpdate strategy did not work with home/end
on a long row.
The spacing of Labeling, Description and friends shall be computed
when breaking the row, not when tokenizing it. Indeed, this is the
right place to determine its correct value.
To this end add a new MARGINSPACE row element type.
This allows to remove TextMetrics::labelEnd, which is not used anymore.
When a string is broken at the margin by the Qt algorithm, the space
at which breaking occurred is automatically skipped in width
computation. However, the ending space of the string is taken into
account and is visible for example at paragraph end.
When the trailing space is followed by a displayed inset, then the
space should be skipped too, which means that the width of the last
row element has to be recomputed. For the sake of performance, the
width of the element without trailing spaces is computed in advance in
FontMetrics::breakString.
This "no space" width will be used when trimming a row element of its trailing
spaces instead of the original one.
Additionally, do not trim trailing spaces when the row is flushed.
Fixes bug #12449.
The search widget triggers a showEvent() in updateTitle() leading to setting the focus to the default push button.
The check in updateTitle() for the need to restore the title avoids superfluous show events and avoids the unwanted focus change.
This ensures that we use a consistent Python interpreter in LyX.
$${python} is replaced by the Python version found.
Users can apply this in preferences and use the same version defined by
LyX.
An inset that resets its font (like Footnote) does not care at all
about enclosing font. Therefore the real starting point is the class
default font. This avoid cases where the footnote contents is forced
to \normalsize.
It turns out that the Greyedout note inset, did inherit font but was
declared as not doing it. This commmit changes the definition by
adding \normalfont\normalsize so that no inheritance happens.
Note that actually \normalfont resets everything but the font size.
This does not matter for footnote (which has its own font size) and
greyedout (which is fixed now), but may matter elsewhere. Also, I do
not know what the situation with HTML is.
We could not figure out the root issue, or at least not to the point
where we felt changing this code considering we could not trigger
any bug from a user perspective.
For now, we just add a comment in the code.
For more information, see our ML discussion here:
https://www.mail-archive.com/search?l=mid&q=20211227113249.53bf5a63%40admin1-desktop
This catches the case where the caption only contains a comment, as in export/export/latex/lyxbugs-resolved/cprotect/9313-comment-in-figure-float-caption. No duplicate work is performed to ensure the same level of performance as before.
The checkbox is hidden when backing store is enforced (wayland,
macOS). In practice, only X11 and Windows users will see it ; I only
have evidence of X11 people needing it, I can hide it for Windows
users too if necessary.
Fixes bug #12119.
LyX relies on a a backing store to draw when running under macOS or
Wayland, because Qt arbitrarily overwrites parts of the workarea
before we paint (and we paint only the parts that need to be painted).
However it seems that this is also necessary on X11 when the WM theme
is translucid. Since there is no way that I know of to detect this
situation, this patch adds a LyXRC setting to manually select this
drawing strategy.
Note that using a backing store is not always a good solution, since
this disables subpixel aliasing.
At this point there is no UI for the variable.
Fixes bug #12119
The default in the .ui is not considered in
GuiSearchWidget::restoreSession(), which sets a hard-coded default
if the stored setting is not found in the session file.
I'll start an ML discussion for whether the code in restoreSession()
should use the ui's default if the setting is not found in the
session.
For some reason, Qt can break strings before the first character
(although we try to prevent that). The code was not prepared to that,
now it is.
Remove a forgotten debug statement.
When a selection extends on more than a row, the space between the two
rows should always be painted in full. Otherwise, with layouts like
Chapter, the "Chapter" label may seem selected in part.
Increase the maximal size of the breakString cache (to compute where
to break lines) from 512kB to 10MB. This has a big impact of cache
hits on large file like the example in #12297, which is now 99%. On
this example the time taken by breakString decreases from 33.5us to
2.4us.
The string width cache has been increased fro 512kB to 1MB, but this
does not make such a big difference.
Additionally, comments and variable names have been improved.
Related to bug #12297.
Add new row flags Flush and FlushBefore to let insets indicate whether
they cause flushing of current row (eg. newline) or of previous row
(e.g. display insets).
Replace FontMetrics::breakAt, which returned the next break point,
with FontMetrics::breakString, which returns a vector of break points.
To this end, an additional parameter gives the available width for
next rows.
Rename various variables and methods accordingly. Factor the code in
breakString_helper to be more manageable.
Adapt Row::Element::splitAt to return a bool on sucess and provide
remaining row elements in a vector. The width noted above has been
added as parameters.
Rename the helper function splitFrom to moveElements and rewrite the
code to be more efficient.
Remove type of row element INVALID, which is not needed anymore.
The code in TextMetrics::breakParagraph is now much simpler.
In Row::finalize, remove the code that computed inconditionnally the
current element size, and make sure that this width will be computed
in all code paths of Row::Element::splitAt.
In TextMetrics::breakParagraph, get rid of the fragile `pos' local
variable, which was not correctly updated. Rely on the endpos of the
last element in row instead.
Rewrite cleanupRow to rely on the endpos of last the row element to
set row endpos, instead of a `pos' parameter.
Instead of having breakParagraph decide when breaking a row is
necessary, let Row::shortenIfNeeded set the row_flag of the last
element to request a row break. This was already done in splitAt.
This is in preparation of splitAt splitting in more than two elements.
Move to Row::Element::rtrim the code in Row::shortenIfNeeded that
removes trailing spaces from last element in row, so that it can be
called when actually breaking a row.
Fixes bug found by Kornel.
In this case, the extra element returned should empty but valid. The
row flag BreakAfter is set to indicate that we have a break there
(this principle will be used more generally in a forthcoming commit).
To detect that we cut at the trailing space, it is necessary to rely
on the difference between QTextLine::horizontalAdvance() and
QTextLine::naturalTextWidth() when the flag
QTextOption::IncludeTrailingSpaces is used: the trailing space is
taken into account in the later, but not in the former.
Somme comments have been added to make code intent clearer.
At least with Qt 4.8.7 on Ubuntu 16.04, QTextLine::lineWidth() can
return a bogus value, at least with Courier font. One hypothesis is
that the invisible characters that we use in breakAt_helper are given
a non-null width.
Work around it, although the exact bug has not been pinpointed.
Change semantics of Row::shortenIfNeeded: instead of breaking the row
and returning a boolean, it returns the list of row elements that have
been removed (or broken) from the row. The logic of the method remains
the same.
Use shortenIfNeeded in breakParagraph. This was the last missing block.
Remove Row::breakAt and the old breakRow. Only bugs remain now :)
To this end, add the helper function needsRowBreak which computes the
effect of two consecutive row flags. This function implements the
priorities described in RowFlags.h.
This function is called with the relevant flags, or NoBreak* when at
boundaries and updates need_new_row.
Some common code is factored in a new cleanupRow() helper.
Remove the code that computed the width every 30 characters (yay!).
Make sure that finalizeLast() is called after inserting a row element in
a row in breakParagraph.
Still many features missing:
- handle insets that break rows (display math, newline, ...)
- handle rows that are too long by replacing the single call to
breakAt with a call to a reworked Row::shortenIfNeeded.
- some easy things at the end of breakRow (bidi text, etc.).
Move the enum definition RowFlags in its own include file, to avoid
loading Inset.h. Document it more thoroughly.
Rename RowAfter to AlwaysBreakAfter.
Add CanBreakInside (rows that can be themselves broken). This allow to
differentiate elements before bodyPos() and allows to remove a
parameter to shortenIfNeeded().
Make the Inset::rowFlags() method return int instead of RowFlags, as
should be done for all the bitwise flags. Remove the hand-made bitwise
operators.
Set R::E::row_flags when creating elements.
* INSET elements use the inset's rowFLags();
* virtual element forbid breaking before them, and inherit the *After
flags from the previous element of the row;
* STRING elements usr CanBreakInside, except before bodyPos.
More stuff may be added later.
This contains large parts of breakRow, but creates a unique row for the paragraph.
The parts taken or not in redoParagraph are annotated.
The new method is not used yet.
Move declaration of RowList to Row.h
Move initialization of POD members of Row and Row::Element to declaration.
Make method isVirtual() depend on type.
Add new row element type INVALID and method isValid()
Make methods R::E::left/right_pos inline.
Add method R::E::splitAt() that returns an element containing the
remaining stuff, or an invalid element if nothing was split. breakAt
is now a simple wrapper around this function.
Add method R::push_back().
Since we intend to break the row element in two, it is not good to
truncate the string too early.
Moreover, the row element width is now set at this point, even if no
breaking occurs.
When inserting € in a math cell, it is put in a text inset and the
cursor leaves the inset. However, inserting ¤ then leads to
\text{€}\text{¤}.
Therefore, try to see if there is a previous \text inset that can be
recycled and insert the new inset there in this case, leading to
\text{€¤}.
Fixes bug #11979.
This allows to simplify the code in Lexer and to remove the dependency
on Formats class.
As a consequence, a pair of dummy definitions of isZippedFile can be removed.
Some new parskip possibilities had been added, but the check for
custom length index had not bee updated.
This code is very fragile.
Related to bug #10968.
The code that determine whether an InsetArgument is passThru is
complex and lives in updateBuffer.
This patch factors out the code in a new init method and calls it also
in doInsetInsert when inserting a InsetArgument.
Fixes bug #12143.
Rather than that, keep it with a warning that it is not available.
Fallback procedure (which maintains security measures) is done in
the conversion step.
This prevents document properties being silently changed on sharing.
The QWinMime class has been removed in Qt6 but the functionality
is still present. However, one has to allow inclusion of private
headers and register the mime handling to the QWindowsApplication
native interface.
Those checks might not be needed, but it's not self obvious from
the surrounding code. Because we already experienced crash from
similar change (cf 1c1c83eced), let's be prudent here.
If you know that these pointers can't be null from broader context
feel free to remove the guards.
Introduced by 24926b2e23, fix 104fdcc9be not backported
but now fixed by 1c1c83eced in 2.3.
https://www.mail-archive.com/lyx-devel@lists.lyx.org/msg216414.html
Clang 12 (at least) misses the fact that tag is always initialized,
because the if/else sequence does cover all cases.
Initialize the variable although it is not required. It does not hurt
at least.
The code is written in such a way that the elements are searched
several times. This can be expensive when there are a lots of insets
in the document.
Concerning the sanity checks, they are now conditionned on the
presence of assertion.
Related to bug #12297.
Add parameter 'force' to scrollToCursor(...) to avoid the case where the
cursor is not set to top because it is already visible on screen.
Change screen offset in this method so that the paragraph is really at
the top of the screen. This part may cause unforeseen issues and needs care.
gotoInset: use the new force flag and do not trigger a redraw.
Instead, return a boolean telling whether redraw is needed.
In the code that use it, set an update flag instead of the extra redraw.
In the handling of paragraph-goto, also set the update flag instead of
triggering a repaint.
Remove Bufferview::scrollToCursor(), which was equivalent to showCursor().
Fixes bug #10425.
Lyx crashes on export to pdf if used with sanitizer set to 'unspecified'.
Crash found by Scott.
Given that if we export without GUI, there is some weirdness here though.
1.) Why does lyx not crash if not using '-fsanitize' compile-option
2.) Why is export to pdf dependent on the screen-resolution
This fixes two performance issues and improves the performance of
TextMetrics::redoParagraph by 15% in a workload that uses the cache a
lot. The difference will be much less when the cache is not used much.
1/ repetion of the hash code computation
The code
if (cache.contains(key))
result = cache[key]:
is not efficient, since qHash(key) has to be computed twice.
To fix this a new Cache::object_str() method is added, which allows
if (auto * obj = cache.object(key))
result = *obj;
2/ code of has code computation
Instead of using a verbose string that is complicated to build as
key, new key structs BreakAtKey and TextLayoutKey are introduced,
along with the relevant qHash() implementation.
IGNORE is a typical placeholder for tags that should not be output, along with NONE. At some point, we should check if both are required, or if NONE is enough…
Initially Qt5 modifier handling was broken. Therefore a workaround was introduced.
This workaround broke the LyX modifier handling with swap of Command-Control-key disabled.
The change disables the hack to get the correct behavior in LyX.
We now use a new variable, "wrap", to track if a wrap should be
done, which is true either if "auto_wrap" is true or if the user
chooses to wrap in the dialog.
This preserves the meaning of the "auto_wrap" variable and also
removes the confusion of why the previous code of
if (!auto_wrap) {
...
}
if (auto_wrap) {
did not use an "else" instead of the second "if".
No change in functionality.
Does not change much, but Coverity complains about it. Let's see if
this allows Inset's child classes to grow their own move constructor.
Also some whitespace.
This requires the introduction of the booleans selected_left and
selected_right in PainterInfo. These tell whether the selection
continues at the left/right of the inset.
This information allows to
1/ paint equation number in the right color: either current text color
or selection text color.
2/ before that, paint a small background rectangle of the correct color.
This allows to avoid painting a large rectangle of an arbitrary color
that was the cause of the bug.
Fixes bug #12319.
When the cursor cannot move on cursor up/down, at least the selection
should be cleared (when not selecting).
To detect this, the method Cursor::upDownInText has been modified to
return true when cursor is at top/bottom of inset, but there is some
room above/below.
Moreover, introduce the functions LFUN_FINISHED_UP/DOWN, which is
dispatched at upper cursor level as long as no local movement is
possible. This allows to handle differently the original char moving
action and its consequences.
Fixes part of bug #12310.
Example: when a selection is set, a `Left' cursor movement would not
reset selection when the cursor was at the beginning of buffer.
To fix this, it is necessary, when cursor is in top-level text, to
avoid the mechnanism (undispatched cursor) that sends the action to the
upper level (necessary when the cursor leaves an inset).
The change is mechanical and is done for : char-backward,
char-forward, char-left, char-right, word-left, word-right, word-left,
word-right. It might be possible to factor this code a bit, but there
is no evident solution.
char-up/down are *not* handled at this point.
Fixes part of bug #12310.
Horizontally, the old code would count an extra pixel on the right.
The vertical test is not changed, and should be eventually audited.
Fixes bug #10468.
I wanted to remove the test Qt >= 5.15, but now I see that it is
not possible because of QProcess::splitCommand.
Neverthless, the best would be to change parsecmd.
