[This commit is the output of the "horizontal scrolling" GSoC 2013
project, by Hashini Senaratne. The code has been cleaned up, some
variables have been renamed and moved from the Cursor class to
BufferView::Private. This is the base from which I (jmarc) will polish
the feature for landing on master.
Below is the original commit log of Hashini, updated to reflect the
changes that have been done.]
This feature also applicable for other insets; graphics and labels.
This implementation is capable of scrolling a single row when reaching
its content which is beyond the screen limits, using left and right
arrow keys.
The attribute 'horiz_scroll_offset_' introduced in the
BufferView::Private class plays a main role in horizontal scrolling of
the wide rows that grow beyond the screen limits. This attribute
represents by how much pixels the current row that the text cursor
lies in should be get scrolled.
The main logic that is responsible for drawing the scrolled rows is
within the BufferView class, BufferView::checkCursorScrollOffset.
* The main logic is called via BufferView::draw.
* What this does is set the horiz_scroll_offset_ attribute in in order to
show the position that the text cursor lies in.
* To make sure that BufferView::draw gets involved when Update flag is
FitCursor, necessary changes are made in BufferView::processUpdateFlags.
Basically what the logic that used to set the horiz_scroll_offset_
does is,
* The row which the text cursor lies in is identified by a
CursorSlice that points to the beginning of the row. This is the
'rowSlice' variable used in BufferView::checkCursorScrollOffset. Acessors
are added to obtain this variable. Here row objects were not used to
identify the current row, because it appears that row objects can
disappear when doing a decoration update for example. This means that
comparing row pointers is not a good idea, because they can change
without notice.
* Stop calculations of horiz_scroll_offset_ variable, if metrics have not been
computed yet. Otherwise the calls to TextMetrics::parMetrics, calls
redoParagraph and may change the row heigths. Therefore vertical scrolling
feature may get disturbed. This is avoided.
* Using BufferView::::setCurrentRowSlice resets horiz_scroll_offset_
when changing cursor row. This is done in order to prevent unwanted
scrolling that happens when changing the selected row using up and
down arrow keys.
* Recompute inset positions before checking scoll offset of the row, by
painting the row insets with drawing disabled. This is done because the
position of insets is computed within the drawing procedure.
* Current x position of the text cursor is compared with the
horiz_scroll_offset_ value and the other variables like row.width(),
bv.workWidth(). Compute the new horiz_scroll_offset_ value in order
to show where the text cursor lies in. The basics conditions that we
check before recomputing it are, if the text cursor lies rightward to
the current right screen boundary, if the text cursor lies leftward
to the current left screen boundary, if the text cursor lies within
screen boundaries but the length of the row is less than the left
boundary of the screen (this happens when we delete some content of
the row using delete key or backspace key).
* Change update strategy when scrooll offset has changed. This allows to
redraw the row when no drawing was scheduled. By doing so, it was
possible to redraw a wide row when moving to the leftmost position of the
wide row, from the leftmost position of the row below, using the left
arrow key.
In TextMetrics::drawParagraph it is checked whether the current row is
what is drawing now. If it is so, the value used to the x value of the row
for drawing is adapted according to BufferView::horizScrollOffset.
The method used to pass boundary() was fixed to get row when cursor was in
a nested inset. This matter is considered in Cursor::textRow and it is
modified accordingly.
GuiWorkArea::Private::showCursor() is modified to show the cursor position
in a scrolled row.
This happens when part of the word is selected.
To reproduce:
1. Start a new document
2. Type "af"
3. Select 'a'
4. Observe that the right part of the 'f' is clipped away.
This patch uses QRegion to set a clip region that is everything except
the part that is drawn in another color.
Fixes: #9223.
This is the first part of bug #8553. \textgreek was interpreted as the
polyglossia version even if polyglossia was not used. Now it is not recognized
at all and converted to ERT, but this ensures at least correct LaTeX output.
Actually I wanted to do that in cc87f8100 but forgot to adjust the original
solution completely. Now we do not search for an arbitrary latexkeys instance
which just happens to have the same name as the macro, but we only use the
symbol that was explicitly set for global macros.
Previously tex2lyx did ignore table columns if the column specifier was
unknown. This can lead to data loss, and now it does not ignore these columns,
but assumes that the column specifiers use only once character. This can still
lead to data loss, but less often.
If the user gave the -fixednec argument to tex2lyx then set that encoding as
LaTeX input encoding in the generated LyX file. Otherwise, included .bib files
could be interpreted using a wrong encoding.
It works with gcc >= 4.9.0 and clang (with libc++ or gcc libstdc++ from gcc
>= 4.9.0). The MSVC parg is missing, because I cannot test it, and the
autotools build still link against boost::regex even if it is not needed, but
I don't know how to fix that.
Now we can handle all types of captions defined in layout files or modules.
This does also mean that \captionabove and \captionbelow are only recognized
in KOMA-classes and not always anymore, but this is correct, since these
captions would not work in other document classes anyway.
As discussed on the list. We don't need it anymore, either we have a modern
compiler that supports C++11, or we fall back to boost. I kept and adjusted
the regex #define, since we cannot use std regex completely yet.
This fixes the tex2lyx test test-refstyle-theorems.tex. It seems that the
intent of the fix was to remove a \protect in front of a \caption,
\captionabove or \captionbelow, but the implementation did not really do that.
Furthermore, it is not clear in which cases a \protect in front of a caption
needs to removed, and in which cases it needs to be kept: After looking at the
LyX sources I could not see that caprions are always output with \protect.
All the code that is run before row metrics have been computed should use int arithmetic. After metrics have been computed, we still need doubles because fully justified rows use double for Row::Element::extra.
Rename Row::x to Row::left_margin and change its type to int.
Rename Row::Element::width() to full_width(). In some places of the code, use dim.wid (the int version without the extra separator) because metrics have not been computed.
Let Row::Element::x2pos take a int& argument instead of double&
Let Row::Element::breakAt take a int argument instead of double
This patch fixes a series of warnings like:
{{{
In file included from ../../master/src/mathed/InsetMathBoldSymbol.cpp:13:
In file included from ../../master/src/mathed/InsetMathBoldSymbol.h:15:
../../master/src/mathed/InsetMathNest.h:37:7: warning: 'lyx::InsetMathNest::metrics' hides overloaded virtual function [-Woverloaded-virtual]
void metrics(MetricsInfo const & mi) const;
^
../../master/src/insets/Inset.h:186:15: note: hidden overloaded virtual function 'lyx::Inset::metrics' declared here: different number of parameters
(2 vs 1)
virtual void metrics(MetricsInfo & mi, Dimension & dim) const = 0;
^
}}}
For a description of the problem, see for example:
http://stackoverflow.com/questions/18515183/c-overloaded-virtual-function-warning-by-clang
3 different strategies have been used:
* in frontend, some functions have been renamed.
* in InsetMath.h, Inset::write has been explicitly imported too
* in InsetMathNest.h, since a comment said that hiding Inset::metrics is intended, a special trick has bee used to silence the warning.
This is needed since all formats are stored in a global list which is shared
between threads, but never modfified except from the main thread.
The only missing bit is extension_list_, which is not so easy to do.
If we compile in C++11 mode, do not use the boost replacements for bind,
functional and shared_ptr. regex is excluded, since it misses match_partial, and
gcc does not provide a usable one in versions less than 4.9.0.
I also removed the #define for match_partial, since this is dangerous. Now you
get a compile error instead of subtle runtime differences.
The old detection did only work if CFLAGS contained -std=c++11, since ciso646
was only included for __cplusplus > 199711.
Thanks to Koernel for the cmake part.
Now all const methods may be called without additional locking.
This is assumed by the threaded LaTeX export, which always useses a globally
unique instance for each encoding.
Read-only access to these classes is now threadsafe, with one exception:
The encoding neds to be already initialized (i.e. init() must not be called).
This makes bug 9336 unreproducable on my machine, although it is not completely
fixed yet.
The interface is now 100% unit tested, and the typedefs depend on the new
STD_STRING_USES_COW configuration variable. The only missing bit is to detect
clang and disable STD_STRING_USES_COW for clang.
As discused on the list. This is not used yet, but it is intended to provide
thread-safe read-access without the need for synchronization if the used STL
implementation does not provide it for std::basic_string. This is the case for
all implementations using copy-on-write.
docstring is already defined in strfwd.h (which is included from docstring.h).
There are only two possible cases:
Either the typedef in docstring.h defines an identical type (then it is not
needed), or it defines a different type (then it generates a compilation error)
=> it is not needed.
The reason being that the TEXINPUTS path list was not quoted on Windows.
This was no problem with spaces but some special characters are
interpreted by the shell and can cause problems. In this particular
case, the '&' character was being interpreted as a command separator.
Thanks to Enrico, who noticed that the previous fix did not take into account
the case of nonempty length argument + the next line beginning with [.
Now the parsing is exactly the inverse of InsetMathGrid::eolString().
These are now in version.cpp. The build machinery should therefore make sure
that version.cpp is recompiled at every compilation.
These variables are now referred to by the other places that made use of __DATE__ and __TIME__.
This allows to address two main issues
* \thanks does only accept one paragraph, while \footnote allows several (ticket #2666)
* footnotes in titling environments were not numbered on screen.
Moreover, the code reduces hardcoding of features, which is always a good thing.
There are several pieces in this commit:
* new numbering type \fnsymbol for counters
* the Foot inset changes its layoutName() to Foot:InTitle when inside a paragraph with InTitle property. This is set when running updateBuffer.
* Foot:intitle uses the \thanks command, does not allow multiple paragraphs and marks its contents as moving argument.
* The InsetLayouts for Foot now have properLaTeXName/Type, so that InsetFoot::latex can be removed; further code simplification is probably possible.
Fixes: #2666
There was an unsymmetry between reading and writing: InsetMathGrid::eolString()
adds curly braces if the first cell of the next line starts with [ to prevent
misparsing as optional argument of \\. These braces were not removed on reading.
Some applications do not accept forward slashes, so call external viewers
and editors with backward slashes which is the usual convention under windows.
The many occurances of the mutable keyword in LyX sources make it quite
difficult to understand what is really const and what not, and to debug
multithreading memory problems. In this particular case it is not needed
at all to have a mutable member, so make it non-mutable.
The advantage is that the buffer will no longer be
marked as dirty when converting to the same type.
A quick break here is done instead of disabling the
LFUN because disabling the LFUN would lead to a greyed out
entry, which might confuse users. In the future, we might
want to have a radio button for switching between notes.
The assignment name = sub.str(1) reads from the first argument given to
regex_match(), but previously this was a temporary object which was already
out of scope. This did probably not matter much in practice, but invoked
undefined behaviour, and as we all know this is allowed ton format your hard
disk or kill to your cat, so better fix this.
One declaration of doExport() was not used, and the other one was only used by
Buffer, so make it private since returning the exported file name is a bit
strange if it can also appear as input argument.
This makes the script usable on windows and speeds it up by an order of
magnitude, since no new process needs to be forked for each layout file.
It also does not conevrt .old files again.
This is different from bug #8999, since in this case a new macro instance is
created. You still get a TeX capacity exceeded error if you try to typeset the
exported document, but this is the same as for bug #8999 and better than a
crash.
This is a patch I originally sent to lyx-devel in 2012 with subject
'Load footmisc.sty instead of using copied code from obsolete stblftnt.sty'.
It now takes all comments into account: It works also if the user loads the
package herself, it can be disabled by providing the footmisc feature in a
layout, and it does not use the ugly \AtBeginDocument{}.
The symptoms are the same as the original #9224, but the reason is different. In the new code, the width of justified rows includes the Row::separator value. Therefore the CRC computed in ParagrahMetrics::computeRowSignature is not updated in some cases.
The fix is to add Row::separator as one of the elements of the row crc.
Fix a crash reported in #7727. This happened because cur.pos() was reset before cur.pit(). In this case, cur.lastpos() will usually be wrong.
Fix bad behaviour when selecting at top level with several paragraphs.
Update documentation.
BufferParams::setDocumentClass() deletes the old document class of the buffer
and creates a copy of the new one. Therefore, the stored layout pointers in
the paragraphs (actually only one paragraph exists) are dangling afterwards.
Resetting them before setting the document class helps here.
In the long term we need to get rid of BufferParams::setDocumentClass()
completely, this is very ugly.
Invoking a computer algebra system program for computing a selected
subformula has never worked (checked with all LyX versions back to 1.3)
and, moreover, in the presence of a selection things go awry.
This commit allows this computation by replacing the selected
subformula with the result of the computation.
A similar fix was reverted (453ce611) because of crashes.
The crashes occurred simply because of a failed check that
we have a buffer view before using it. That is now done in
this commit.
The below commit description is copied from the original
commit (fb05011a):
Empty selections can cause confusing behavior for a few reasons:
(1) some functions behave differently depending on whether there is a
selection. If I press delete, nothing happens (where I expect the
character or inset before the cusor to be deleted). If I toggle bold or
emphasize nothing happens (where if there is no selection the entire
word is toggled). There are other LyX functions that depend on whether
there is a selection or not. Further, I wonder if any part of LyX's code
assumes that if there is a selection it is non-empty.
(2) menu options are incorrectly set. For example, the scissors icon.
For remaining empty selection issues, see #9222.
For more information, see:
https://www.mail-archive.com/lyx-devel@lists.lyx.org/msg184758.html
Our TempFile class uses QTemporaryFile internally, and, on Windows,
this keeps the file locked so that we cannot rename it. So we have
to release the lock.
There are 3 possible actions (in order)
* select current cell
* select all calls of inset
* select the inset from outside (in the containing inset)
This fixes completely #7727.
The shape of the parbreak separator is slightly changed in order to
better distinguish it from the forced newline. This allows using the
same color of the plain version without risk of confusion.
The problem was that some parts of the code assume that the right margin is part of the width of the row. This is not the case, only the left margin (Row::x) is counted.
The code has been fixed and documented to reflect this.
Add display_pixel_ratio to buffer params to use it for displays with high resolution.
It holds the highest ratio between physical pixels and device-independent pixels of the LyX application.
Preview snippets will be generated using this value to get high resolution preview.
Add pixel_ratio to graphics params to use it for displays with high resolution.
It holds the ratio between physical pixels and device-independent pixels of the graphics.
Introduce the concept of pixel ratio: the ratio of physical and device independent pixels.
This is useful for rendering of content on Retina-displays of Mac hardware with high resolution.
Qt has real support for this starting with Qt5 - therefore it has to be compiled conditionally.
This change uses some work of Marcelo Galvão Póvoa, thank you.
What a selection spans several cells in mathed, and a deletion occurs though LFUN_CHAR_DELETE_FORWARD for example, only the case of a proper grid was handled. When no such grid exists, all cells between the first and the last index are cleared now.
If there is only one row, a new line should not be appended. If
there is more than one row, the new line should not be appended for
the last row.
For example, if in LyX there was "as you can see in $x$ there..." a
copy/paste would lead to a newline after "x".
lyxfind.cpp(findNextChange, findPreviousChange, findChange, selectChange): factor the change-selection part out of the change-finding part
Text.cpp (acceptOrRejectChanges): call only selectChange
mhchem loads amsmath and needs therefore be loaded after it but before packages that redefine commands of amsmath. We already take care of esint but not of wasysym
If we are testing tex2lyx the first time, then the userdir
is empty. We call now configure.py to initalize in case it is needed.
The algorithm to detect if reconfigure is needed mimics the lyx behaviour.
The check for a latex format is very fragile. Both libmagic and our
guessing from contents cannot distinguish the pstex and pdftex formats
used by the xfig external inset. Moreover, it may also happen that
lyx files are mistaken as latex ones. Thus, when the guessed format
is latex, the only solution is to give precedence to the format
determined by the file extension.
This commit replaces almost all occurrences of Q_WS_WIN to comply
with Qt5. The remaining occurrences should *not* be replaced,
because the guarded code won't compile on Qt5.
The command line argument -geometry WIDTHxHEIGHT±XOFF±YOFF
specifies a preferred size and location for the main window.
Currently, this is semi-broken on Windows. Indeed, only
specifying WIDTH and HEIGHT places the main window such that
the left and top borders are invisible such that the window cannot
be moved. Moreover, the XOFF and YOFF parts (when present) are
used to specify the distance of the window from the left and top
or right and bottom edges of the screen, when using '+' or '-',
respectively. However, -geometry 800x600-20-20, instead of placing
the window such that its bottom and right edges are at a distance
of 20 pixels from the corresponding screen edges, places the
window such that its left and top borders are out of the screen.
This is corrected by this commit, which also addresses the fact
that Qt5 does not define Q_WS_WIN anymore.
Investigation of bug #9236 showed that crash to be due to a Paragraph's
holding a dangling pointer to an old and deleted Layout after the
DocumentClass was reset. Since the backtraces look almost identical, it
seems likely that we have the same problem here.
Since this crash seems almost always to involve tables, I looked at the
code in switchBetweenClasses() and found that the Paragraphs that belong
to "hidden" table cells are not seen by the initial recursion using a
ParIterator: It skips right over them. This was confirmed by test code
suggested by Enrico, with results reported in Trac.
The present patch attempts to deal with this problem in the second
recursion, over Insets. When we see an InsetTabular, we call a new
routine that recurses through the cells, looking for hidden ones. If it
finds a hidden one, it then resets the Layout for the cell's Paragraphs
(there should be only one, but we do not make any assumptions) to the
PlainLayout that belongs to the new DocumentClass. This is good enough,
since such cells never have content.
There is extensive discussion of the patch here:
https://www.mail-archive.com/lyx-devel@lists.lyx.org/msg185095.html
Additional testing by Enrico and me confirmed the existence of the
dangling pointer.
This makes virtually impossible copying a separator inset whithout
also copying the end of paragraph. These insets are not supposed to
be directly inserted by users. For example, the parbreak version
represents a LaTeX paragraph break, not a LyX one. So, if it is
possible to copy and paste it by alone, an unsespecting user may be
surprised to see a paragraph break in the output but not on the LyX
screen (because of the lack of indentation, for example).
In this way, it also becomes a LyX par break from a user point of
view, not any more useful than simply introducing a par break by
hitting <return> (except in those cases where it makes a difference,
in which case they are automatically inserted by LyX).
Empty selections can cause confusing behavior for a few reasons:
(1) some functions behave differently depending on whether there is a
selection. If I press delete, nothing happens (where I expect the
character or inset before the cusor to be deleted). If I toggle bold or
emphasize nothing happens (where if there is no selection the entire
word is toggled). There are other LyX functions that depend on whether
there is a selection or not. Further, I wonder if any part of LyX's code
assumes that if there is a selection it is non-empty.
(2) menu options are incorrectly set. For example, the scissors icon.
For remaining empty selection issues, see #9222.
For more information, see:
https://www.mail-archive.com/lyx-devel@lists.lyx.org/msg184758.html
This branch implements string-wise metrics computation. The goal is to
have both good metrics computation (and font with proper kerning and
ligatures) and better performance than what we have with
force_paint_single_char. Moreover there has been some code
factorization in TextMetrics, where the same row-breaking algorithm
was basically implemented 3 times.
Globally, the new code is a bit shorter than the existing one, and it
is much cleaner. There is still a lot of potential for code removal,
especially in the RowPainter, which should be rewritten to use the new
Row information.
The bugs fixed and caused by this branch are tracked at ticket #9003:
http://www.lyx.org/trac/ticket/9003
What is done:
* Make TextMetrics methods operate on Row objects: breakRow and
setRowHeight instead of rowBreakPoint and rowHeight.
* Change breakRow operation to operate at strings level to compute
metrics The list of elements is stored in the row object in visual
ordering, not logical. This will eventually allow to get rid of the
Bidi class.
* rename getColumnNearX to getPosNearX (and change code accordingly).
It does not make sense to return a position relative to the start of
row, since nobody needs this.
* Re-implement cursorX and getPosNearX using row elements.
* Get rid of lyxrc.force_paint_single_char. This was a workaround that
is not necessary anymore.
* Implement proper string metrics computation (with cache). Remove
useless workarounds which disable kerning and ligatures.
* Draw also RtL text string-wise. This speeds-up drawing.
* Do not cut strings at selection boundary in RowPainter. This avoids
ligature/kerning breaking in latin text, and bad rendering problems
in Arabic.
* Remove homebrew Arabic and Hebrew support from Encoding.cpp. We now
rely on Qt to do handle complex scripts.
* Get rid of LyXRC::rtl_support, which does not have a real use case.
* Fix display of [] and {} delimiters in Arabic scripts.
Actually the workaround that is used to show parenthesis in the right direction
is not needed any more, since this is part of the unicode bidi writing algorithm.
This fixes at the same time the use of [] or as delimiters in arabic, which was wrong on screen.
Note that there is a problem with hebrew, but this will require a fileformat change.
The use of RLO/LRO overrides to force text orientation was really hackish and the way it was done caused dropped letters in Mac OS X (for some unknown reasons).
This new approach is much cleaner, except that it relies on features not advertised in documentation
but present at least from Qt 4.5 to Qt 5.3:
* TextFlag enum values TextForceLeftToRight and TextForceRightToLeft, which are strong versions
of QPainter::setLayoutDirection; they are passed as a parameter of QPainter::drawText.
* QTextLayout::setFlags method, which is required to pass the above flags to QTextLayout.
The unicode override method is still used to draw strings Mac OS X because, for some reason, the direction was not really enforced in this case.
This can only be done where splitting of string is identical in row breaking and display. It will be possible to reintroduce this when row painting uses the tokenized row information.
The option --enable-qt5 allows configuring for Qt5. The default is Qt4.
Nothing special is done with respect to Qt4, apart from pulling in the
correct libraries. Indeed, other than the core and gui libraries, now
also the concurrent and widgets libraries are needed.
The conversion from floating point to string performed by
boost:lexical_cast does not allow specifying a precision and,
for example, values such as 0.9 are resturned as 0.899999976.
The standard C++ way for performing the conversion is using
std::ostringstream which is exempt from this problem, even if
less efficient. For the sake of accuracy, boost::lexical_cast
is ditched in favor of the ostrinsgstream implementation.
In C++11 another option would be using std::to_string, but I
think it is not as efficient as the boost way and not worth
implementing through #ifdef's.
Incidentally, this patch would have also fixed#9190 and all
similar cases involving the use of convert<string>(float|double).
When a font is scaled by a certain percentage in the document settings,
LyX was outputting a ridiculous parameter value. For example, if the
font is scaled 90%, the corresponding parameter was "scaled=0.899999976".
The patch avoids this and, in the previous case, one gets "scaled=0.9".
This is not only cosmetic, because in roundtrip conversions the parameter
would be continuosly changing.
This commit and b60b505f should be backported to the 2.1.x branch, where
reimporting with tex2lyx an exported document produces wrong results
(also in version 2.1.0).
This variable was introduced to guard against any bad consequence of the then-new right-to-left
languages support. Let's be bold and get rid of it altogether!
Now right to left support is always enabled.
This commits (tries to) reintroduce properly the code that was reverted at the beginning of this branch. This had to be done because these patches interefered with the big refactoring of TextMetrics.cpp.
This commit reintroduces the changes to TextMetrics.cpp contained in c668ebf6, c85dbfea9 and 061509bf.
This is handled by Qt now.
Note that a small optimization (do not draw text that is to the left
of WorkArea) is removed because it cannot be guaranteed to be exact
anymore. It was probably not very useful anyway, and would become
useless once the RowPainter is rewritten to use Row information.
Update 00README_STR_METRICS_BRANCH.
The display of partially-selected word is now done in a new Painter::text method
which displays the string twice with different clip settings. This allows to
catter for the case where Color_selectiontext is not black.
Morover, the code that uses unicode override characters to force the
direction of a string is moved to lstrings.h.
Fixes: #9116
Moreover, breaks row at insets when there is no suitable separator.
Also make the code of Row::shorten_if_needed somewhat simpler by using
iterators and factoring the code.
Fixes: #9120
The old implementation of Row::Element::pos2x and x2pos did not work
correctly with Arabic text, because characters can have shapes that
depend on context.
This new implementation leverages QTextLayout in a simplified way,
since only one word is added to the layout.
This allows to make Row::Element::x2pos more readable.
Fixes: #9115.
Do not cut strings at separators in RowPainter when text is not
justified. This speeds-up painting by reducing the number of strings
to draw.
Do also a modest cleanup of paintChar (remove dubious optimization).
Instead of relying on character range (Hebrew or Arabic) or character
direction, use RLO unicode character (Right-to-Left override) to force
painting in the direction indicated by the current font. This should
be as close as we can to the old LyX behavior (and requires less
code).
If this code works as intended, it will be possible to remove a lot of
code from Encodings.cpp.
We rely on Qt built-in unicode support for handling Arabic and Hebrew
compose characters. This allows to avoid to use our homegrown
machinery.
This should provide a nice speedup at a low cost and
will eventually allow us to get rid of:
* most of our Arabic/Hebrew machinery in Encodings.cpp,
* Paragraph::transformChar,
* and probably more.
All these problems are related to what happens at the extreme points of rows
* since VIRTUAL elements have a width but no contents, they have to
be treated specially at some places. It would have been better to
avoid testing for them explicitly, but I did not find a way.
* Improve and cleanup the code in breakRow and fix in passing a crash
when clicking on the right of an incomplete MARGIN_MANUAL
paragraph.
* improve the computation of row width in TextMetrics::computeRowMetrics.
* handle properly the case where a position if not found on the row
in both cursorX and getPosNearX (actually, this happens when
selecting).
* Some code cleanup and comments.
The fact that the bug was still present in the features/str-metrics
branch comes from a goof in the initial implementation of 'virtual'
row elements (completion and end-of-par markers). Now that this is
corrected, everything works as it should.
The fact that the bug is present in master is due to some other reason
that is not useful to investigate now.
build_script() was already threadsafe, since it used a TempFile, and the
counter was basically not needed, but the new solution makes this obvious
and has the additional advantage that TempFile constructs the real output
file, not a dummy without extension which is not needed.
It was broken in two ways: It was not threadsafe, and it did never detect
any recursion, since the counter was decremented for each non-recursive call
and never incremented again.
This is one of the more important threadsafety issues because of export in
thread and simultanous view source. The solution is ugly, but a better one
(see FIXME) would require major rework. These static variables should not
have been used in the first place IMHO.
Using a static variable here was premature optimization: fileNames() is only
called from GuiRef (directly or indirectly), and since this is a dialog the
copying of a FileNameList is not noticeable at all.
In this case I use a mutex, so the zip status of files is shared between
threads. This is possible because a deadlock can't happen, and it should give
better performance.
Don't create an intermediate copy (found by Jean-Marc).
I doubt that this has anything to do with the mystery crash, but it works, and
following the standard patterns is better anyway.
Without this, you get crashes in a few second when you set the autosave
interval to one second and edit quickly (typing new words etc). The reason
is that the cloned buffer wants to insert words into the word list and
remove them again, but it lives in a different thread.
The compiler-generated copy-constructor and assigment operators would be wrong
for IconvProcessor::Impl, since cd would be copied, and iconv_close() could
thus be called twice on the same descriptor. The old code did work, but now
IconvProcessor::Impl cannot be copied by accident in the future.
The IconvProcessor assignment operator did not delete pimpl_ and used a
non-standard signature. If you want to know why the standard signature is
important, read "Effective C++" by Scott Meyers.
The statement
if (pos < from + lyxrc.completion_minlength)
triggers a signed vs. unsigned warning. I don't know why this happens, it
could be a MSVC bug, or related to LLP64 (windows) vs. LP64 (unix)
programming model, or the C++ standard might be ambigous in the section
defining the "usual arithmetic conversions". However, using a temporary
variable is safe and works on all compilers.
* fix handling of boundary situations in Row::Elements::x2pos;
* fix handling of boundary situations in TextMetrics::getColumnNearX;
* make sure to always use Font::isVisibleRightToLeft instead of Font::isRightToLeft;
* Improve debug messages.
The horizontal position of the inset was not taken in account.
The rounding is not always the same as with the old code, but this
is not really important.
Additional changes:
* improve debug output of rows
* remove Bidi& argument of the RowPainter constructor, since it is always
an empty Bifi that is passed. This means that the Bidi class is not
used at all any more in TextMetrics.cpp. The only remaining user is
RowPainter.
Break words longer than the screen width. The code is more complicated
than I would like, but I have no better idea right now.
Implement properly the notion of a row broken by a display inset. This is useful in different places.
Also fix a bug with last line of a paragraph spotted by Kornel.
Use proper font everywhere for end-of-par marker
Fix getColumnNearX for RTL text and for centered/right-justified paragraphs.
Let computeRowMetrics update the row width.
* convert cursorX to new scheme; old computation is still there for the sake of comparison.
* use Row to compute separator width in computeRowMetrics; again, the old code is retained for now.
* Get rid of rowWidth()
* Row now contains a vector of Elements
* replace Row::dump by a proper << operator
* the width is updated as elements are added
* breakRow is reimplmented to use this infrastructure
Collect properly all strings to compute string metrics
Some changes with respect to existing code
- end of paragraph font is now the one of the text
- words longer than a line are not broken anymore. I think this is not useful if we have horizontal scrollbar
Other than that, the code is still compatible with rowWidth and friends.
The magic library can detect the charset used by a file. While this
detection is not full proof, actually the library seems to be infallible
as regards the binary nature of a file. So, use libmagic for the detection
and fallback to the previous method if the library is not installed or
its database cannot be loaded.
On startup, the default locale is "C", meaning that all system
functions assume an ascii codeset. The environment's locale
settings should be selected by calling setlocale(LC_ALL,"").
This is done by Qt during the QCoreApplication initialization
but this inizialization is never performed for batch processing
and, as a result, LyX is not able to process files whose names
contain non-ascii characters. This is not an issue on Windows,
where the file names are always stored as UTF-16, so the call is
only performed for unix-like platforms (this also includes cygwin,
due to its own filenames management that allows using characters
which are forbidden to native programs).
FileName::tempName() created a new temp file name by using qt to create a
temporary file with a unique name, and then deleting that file and returning
the name. This was unsafe, since other processes or even other threads of the
running LyX could create files with the same name between deletion and then
using the temp name.
This is fixed by using the TempFile class instead. As a side effect, a few
cases where the temp files were not deleted after usage were fixed as well.
The only place that is still unsafe is createTmpDir().
The compiler generated copy constructor and assignment operator are wrong.
This could easily be fixed by implementing them manually, but a) they are
not needed, and b) the semantics would be unclear (should the copy point
to a new temp file or not?), so it is better to forbid them.
- The TempFile class guarantees to generate a file name, we are not limited to
100 tries of a predictable scheme anymore, which could break if LyX
frequently crashes.
- The temp file name generation has no race condition against another LyX
instance in the same directory anymore.
- Symlinks survive saving again (regression of 10364082c8).
Thanks to maciejr we know now what the remaining problem was with bug 7954:
My unicode symbol fallback works fine, the problem was that a font named
"Symbol" is available on OS X, but it does not use the font-specific encoding
we expect: Almost all glyphs are at their unicode code point.
Therefore the bug is fixed by re-enabling the unicode workaround and blocking
the Symbol font on OS X.
Maximizing the document settings window when on the modules pane,
the horizontal space is now given to the module names and so no
scrollbar is needed. Before, even when maximized, a scrollbar was
sometimes needed because the horizontal space was given to the
buttons in the middle, which did not provide an extra benefit.
The default sizeType was "Expanding" and is now changed to "Minimum"
for the horizontal spacer above the middle buttons.
Now the cursor in LyX jumps to the right spot instead of simply the
beginning of the paragraph. This is most useful for branch insets,
for example, which may contain long paragraphs.
If the reverse position corresponds to an inset, its paragraph id
does not follow the main text numbering. Typically, an inset has
only a few paragraph, so that we would jump near the beginning of
the document. In this way we at least jump at the beginning of the
inset.
If LyX does not know about a given file format, it may easily
happen that the format is recognized as "latex" and this causes
bug #9146. This patch limits the check for a latex format to
non-binary files. The strategy for deciding that a file has
binary content is the same as that adopted by the "less" program.
The problem is the use of cursor movement methods to update cursor.
Cursor::forwardPos() steps into insets, which is not always what we
want. The problem here is that there is a math inset just after the
accepted change, and that the cursor steps into it for some reason.
This code is a nightmare anyway.
Fixes: bug #9145
This regression was introduced by me at 8b66f9ce. I did not take
into account that a call to a python script containing $$s is embedded
within a separate python script. Thus, when commandPrep() is called it
only sees the call to the outer python script, and does not see the
$$s contained in that python script. It therefore did not substitute
for it. This fix simply calls commandPrep() directly before writing
the embedded command.
The Adobe EPS file format specification (TN-5002, currently available at
http://partners.adobe.com/public/developer/en/ps/5002.EPSF_Spec.pdf) specifies
a binary version of EPS files with integrated preview, sometimes also called
DOS EPS binary files. LyX 2.0 did recognize these files unreliably by
extension, but since f4eae12d they are misdetected as latex (bug #9146).
This change adds proper detection for these files using the officially
documented binary header.
This is needed so that the new format number is actually used. Since the
conversion is a noop I did not update the layout files (more updates will
come).
Do not assert if an inset separator is the only item of a List
environment. Although it is a weird thing to do, both GUI and
latex output can deal with it.
This fixes a crash in examples/fa/splash.lyx when selecting text
representing menu entries. This happens because menu names are in LTR
English, while the inset itself is in RTL.
The problem is that the current code relies on the fact that
1. getColumnNearX and checkInsetHit share the same idea about cursor
position.
2. pos and pos + 1 are in general consecutive on screen.
It seems that 1. is wrong here (for reasons I did not try to
understand); the second assumption is definitely false with
bi-directional text. This makes editXY very fragile.
The new code should be more robust in this respect. The logic is:
* if checkInsetHit finds an inset, use its position,
* otherwise, ask getColumnNearX for the cursor position.
Fixes: #9142
When deciding whether a paragraph should be indented or not, LyX
only takes into account default layouts. This is wrong, because
an environment could be nested into another one and thus a following
paragraph would not be "default". With this patch all paragraphs
after an environment are correctly indented, independently of
whether their layouts are "default" or not.
The latex output (which was modeled following the previous wrong
assumption) is also correspondingly adapted.
This is mainly needed to reduce the amount of ERT if you convert AMS example
documents with tex2lyx. No GUI support is needed, since \notag is equivalent
to \nonumber.
This is a follow-up of bug #8967. The implementation is self-explaining, the
only part which needs a comment is lyx2lyx: Since a 100% correct solution is
not possible, it has been decided not to switch amsmath off in the forward
conversion if no other ams command than \smash[t] and \smash[b] is used, but
to consider it a bug that older versions do not load amsmath automatically for
these commands. In the backward direction it is easy to keep the document
compilable, so just do that.
The firs tinvolves a thinko in BibTeXInfo::expandFormat. We were previously
counting passes through this routine, which means: one for every character,
more or less. So long strings would hit the "recursion limit". But what
we are worried about is an infinite loop caused by misues of macros, so that
is what we need to count.
This prevents the error we were previously getting, but it reveals a huge
slowdown when one tries to open a citation inset with a large nubmer of keys.
So we also limit the number of keys we try to process, and the length of the
string we try to display, when we are generating citation information.
I'm convinced that there is a deeper problem in how citation information is
generated (see the bug tracker for more info), but that will require major
surgery and a file format change
If a new paragraph is created just before a nested environment,
the indentation of the nested environment is not computed
correctly because the parindent of the previous layout would
also be erroneously taken into account. This would cause the
nested environment to move back and forth when something is
added to the new paragraph.
Instead of simply taking into account the layout of the previous
paragraph, it is better considering the layout of the environment
in which the previous paragraph may be nested. This makes the test
simpler and, at the same time, more robust.
LyX fails to indent on screen a standard paragraph when it is
nested into an environment. The fix is a one-liner but the diff
is larger because it also fixes a previous wrong indentantion
in the source ;)
This is necessary, for example, if a standard paragraph is nested
in an environment and the environment does not end immediately after.
To be strictly correct, the layout of the following paragraph should
be compared to the layout of the nesting environment, otherwise, if
they are not the same, an empty line is nevertheless output. However,
this is harmless because an "\end{layout}" tag immediately follows.
I am not sure I fully understand the pending_newline/unskip_newline
logic (which seems mainly related to rtl writing), so I prefer to
leave it alone, in the sense that now things go again as in 2.1
until the point where those booleans are used for producing output.
If it turns out that a spurious (and unwanted) empty line comes
from the previous code, it can be easily corrected later.
I am also reintroducing the check about a separator inset at the end
of the paragraph, because that is necessary for the plain version.
This also changes the type of an int to an ssize_t.
nRead is initialized as an ssize_t because it could
be negative. It is cast to a size_t for comparison
to the size of a vector, but only after we check
that nRead is not negative.
This method is a proxy for LyXRC::preview that forces to forward
declare some wrapper around an enum...
Instead, two simple static methods previewMath() and previewText() are
introduced, that make the code much easier to follow.
The format entries should be sorted according to the culture selected for the
UI. This was not the case previously, resulting in unexpected sorting of small
and capital letters. This is now fixed by using the standard C function
strcoll(). Qt does only offer similar functionality in Qt5, and this is not
mature enough yet to depend on it.
Unfortunately we have a report that strcoll() does not work on MSVC, however
this partial fix is better than nothing. The MSVC issue might also be a
configuration problem, since MS claims that strcoll() is supported. This
still needs to be checked.
addPath() always adds a slash at the end, os got a double one before.
Qt and the OS are clever enough to understand that, but a single slash
looks more nice.
The real problem is the encoding of latex_language: It is hardcoded to latin1,
but InsetListig uses the currently active encoding. Therefore, we cannot tell
whether any given character wil be encodable or not, and we should not prevent
non-ACII characters.
In the future, we need to make the encoding of latex_language dynamic, so that
it always represents the currently active encoding. Then, we could do the
correct check both for listings and ERT. For now, I simply disabled the
encoding check for listings, which also means that bug 9012 might occur in
other cases for listings, but this is less important than bug 9102.
If a separator inset is used after an environment with NextNoIndent
set to false, a blank line was already output. So, avoid outputting
another blank line or an odd looking line with only a '%' character
(as the previous blank line provides already a visual separation).
Multirow cells now have the same alignment rendering in LyX
as in the output. The alignment of a multirow can change as
long as the column is not of fixed width. If the column is
of fixed width, the multirow is left-aligned.
When setting a multirow, the alignment is copied from
the last cell in the selection.
LyX assumes that a standard paragraph following an aligned one or
a layout with NextNoIndent==false has to be indented on screen.
This means that in the latex output a blank line has to follow.
In this case there should be no problem as regards extra vertical
space and it simpler and more elegant to simply uncheck "Indent
Paragraph" in the Paragraph settings pane rather than changing
the current logic and allowing to insert a parbreak separator.
If a layout has NextNoIndent set to true, the following paragraph
is not indented on screen. LyX checks the previous layout for that
style parameter to decide whether to indent or not. Of course,
what matters is the latex output and the on-screen representation
should match this output. Now, when a layout has NextNoIndent==true,
the latex output is correctly not indented, while the on-screen
representation may fail to match this output. This can occur when,
for example, a standard paragraph is nested in the previous layout,
because LyX would check the property of the nested layout instead
of the container layout. Thus, LyX should check the property of a
previous layout at the same depth for correctly deciding whether
a paragraph has to be indented or not.
See also http://www.lyx.org/trac/ticket/9055#comment:12 for an
example document where the previous scenario actually occurs.
