Simplify the logic for language package selection and make it more consistent:
Use polyglossia with non-TeX fonts (system fonts/Unicode fonts) for all
export flavours (XeTeX, LuaTeX, DVI-LuaTeX), if the language package setting
is "auto" and there is no language not supported by Babel and no package
providing Babel.
This solves some Babel-related autotest cases and leads to some new failures
due to the polyglossia language nesting problem.
The lib/unicodesymbols part is based on work by Günter Milde:
Both, \r{A} and \AA (rsp. \r{a} and \aa) are equivalent standard LICR macros
for Aring/aring as well as the deprecated "angstrom sign" character (212B).
However, with \AA for 212B and \r{A} for 00C5, tex2lyx converts \AA to the
deprecated "angstrom sign" which is missing in many fonts including the
Unicode version of Latin Modern.
I added the normalize_c() calls so that tex2lyx prefers the precomposed forms
(these are better editable in LyX) and the deprecated flag.
The conversion is not completely correct yet, but adding the test before the
fix has the advantage that the fix can easily be documented by changing the
test reference.
whose children is also a child of another buffer, then try to close that
one.
The problem is that we do not check properly to make sure that the child
is not a child of some other buffer. Now we do.
whether a given child Buffer is also a child of some other parent.
Then do the releasing or resetting where this method is called.
There should be no change of behavior with this patch.
Improve the test whether cursor has moved in Cursor::posVisLeft. The code for posVisRight had already been fixed for #5764 at 0730c923, but this replaces both tests by the proper == operator.
Fixes bug #9913.
By initializing 'to' to a value, the code made it seem like that
value mattered. But the value is overwritten in getWord().
Further, now if 'to' is used before it is initialized, there might
be a useful compiler warning that could point to a bug.
Gcc STL debugging feature asserts when swapping an object with itself. This happens in some cases with math grids that have only one column.
A quick review of other uses of swap() in the code base did not reveal any other dubious case.
Fixes bug #9902.
At d449e7e6 it has been decided that submenus are going to be displayed even if
all their items are disabled. Here we make an exception for OptSubmenus.
Example of submenu no longer shown: Insert > Insert Regexp
Example of submenu always shown: Edit > Math > Limit Type, Macro Definition
\output_changes is now output at a distance from \tracking_changes.
Since both parameters can be seen as per-user preferences, they can cause
undesirable merge conflicts, in a multi-author setting, were it treated as a
single block by the version control system, as was the case before this patch.
After d5a5fbb8, as indicated in the commit log, it remained to make sure that
the sub-menus of the navigation menu showing the TOCs are generated in a delayed
fashion, to avoid corner cases regarding performance when documents have very
lengthy tocs (e.g. in a document with 1000 sections it takes a few hundreds
milliseconds for the menu to be refreshed). But this idea actually requires
substantial changes to the way menus are computed, so it is not for now.
In the meanwhile, I reintroduce a max size for menus, after which it is cut
off. This differs from the one that I removed at d5a5fbb8 in two ways: 1) if
there are more items than the max size, then we still show something instead of
nothing, 2) we allow ourselves to rely on qt's scrollable menus and therefore
allow bigger menus than before the above commit. The philosophy is that it is
better to show something than nothing, that it's better to show a scrollable
menu than to cut the menu to fit the screen, and that beyond a certain size the
scrollable menu becomes useless anyways.
It is a bad idea to have a QObject variable that oulives the main QApplication object. See for example:
https://www.ics.com/designpatterns/book/globals.html
Here the QTextLayout object was static to the anonymous namespace getTextLayout function, and got destroyed after the freetype renderer had been disposed of by QApplication.
This causes segmentation faults when quitting LyX on some systems.
This patch moves the cache together with other GuiFontMetrics caches. It means that one will have one such QTextLayout per font type, but this will not change much.
Preliminary work for addressing #7790. Thanks to Richard for providing initial
files this is based on.
Adding to TextClass:
OutlinerName <string> <string>
(the second string is translated)
e.g.:
OutlinerName thm "Definitions & Theorems"
Adding to Layout:
AddToToc <string> (default "", means no)
IsTocCaption <bool> (default 0)
e.g.:
AddToToc thm
IsTocCaption 1
Adding to InsetLayout:
AddToToc <string> (default "", means no)
IsTocCaption <bool> (default 0)
e.g.:
AddToToc literate
Adding to inset arguments:
IsTocCaption <bool> (default 0)
The active/inactive status of hfills was not checked in numberOfHFills.
The code is reorganized a bit, but the important part is the change in the above function.
Part of bug #9870.
This crude caching mecanism is useful in the particular case of a screen with many misspelling dotted lines. In this case, it is necessary to build a QTextLayout in order to know where to put the start/end of the spell line. Since rows typically contains text snippets longer than a word, we may be in a situation where the same QTextLayout is constructed repeatedly.
This commit is useful in this particular use case, and should not be costly in other cases. A better fix would be to remember the QTextLayout associated to each row element. This is a bit more work, so this fix should be sufficient for now.
Additionally, do not paint misspelled marks when painting is disabled.
Fixes bug #9890.
Although af8ba5e0fixed#9321, it broke a valid use case that was
reported by a user at #9785. Hopefully we can come up with a better
fix for #9321.
This reverts commit af8ba5e069.
exist. This will allow for a proper fix for bug #8796, though that may or may
not get fixed before 2.2.0.
Also, change the InStyle tag to ModifyStyle, per a suggestion of Jurgen's.
(http://mid.gmane.org/565E17DD.7090008@clear.net.nz)
Fix the pilcrow and the cross in the outliner for Windows and old Ubuntus.
The thin space after the cross is based on aesthetics (in Ubuntu at least) after
trying both with and without a normal space. A thin space is already used for
display in BiblioInfo.cpp, so I presume it is safe.
Ideally, I prefer ❌ (CROSS MARK), whose description fits the purpose, over ✖
(HEAVY MULTIPLICATION X), which looks too bold in the UI. Whereas ✕
(MULTIPLICATION X, successfully tested in Windows by Andrew) looks too dim. But
❌ (CROSS MARK) is next to ❎ (NEGATIVE SQUARED CROSS MARK) so it might be as
problematic as the latter.
Do not output the cross on the author name in the toc of tracked changes.
Previews are now generated when previews are turned on in
preferences. This change ensures that when users activate previews
for the first time, they are not confused by no previews showing up
(a restart of LyX or a triggering of each individual preview would
be required).
There was a previously attempted fix for #9507 at 390ae054 which was
reverted at 358745d0 for performance reasons: it updated previews
after every preference change and updating previews is costly (even
if the cache signals there are no changes needed).
This implementation is consistent with what we do for updating the
system fonts in preferences.
boost::regex supports escape sequences starting with a backslash in format
strings of regex_replace, but std::regex does not. Therefore format strings
involving literal backslashes have to be written differently for both flavours.
The special MSVC handling in regex.h is removed, since it is not needed
anymore, and using grep syntax would definitely be wrong.
The computation of the legth of expanded hfills did not take into account their unexpanded size. This is done now by increasing the dimension (+=) instead of merely setting it.
Moreover, since the insets sizes are integer number, rounding effects have to be taken in account. To this end, the extra number of pixels is added to the last hfill in the row.
This fixes part of bug #9860.
Note not everything is fixed by this patch: the logic of ParagraphMetrics::hfillExpansion seems bogus to me. I do not see why consecutive hfills at the beginning of a row should not be all expanded. Since I do not know what are the peculiarities of hfill handling in LaTeX, I did not change it (yet).
I did not either try to investigate the label hfill part, because I do not even know what is so special about it. I think there is a lot of old logic that nobody ever tried to question.
This is done by implementing contentAlignment() for this inset.
In order to display properly 'stretch' alignment, the code for TextMetrics::getAlign is rewritten to include some code that was in computeRowMetrics.
Back references in the format string of regex_replace use the syntax $n both
in std::regex and in boost::regex for the default format. Boost seems to
support the syntax \n in addition, but it is not documented at
http://www.boost.org/doc/libs/master/libs/regex/doc/html/boost_regex/format.html.
Therefore it is a good idea to use $n also for boost.
The changed code is not used, but I tried to use a similar approach for
boost::regex, and found some problems:
- regex_replace and regex_search are implemented in the replacement, so they
must not be used directly
- an smatch object must be given by reference (as in the called methods),
otherwise an exception would be thrown at runtime
- the commented out regex_replace version is actually needed
This code is supposed to be deleted, but nevertheless I wanted to record here
how it had to be modified if it was actually needed.
It was possible for errors that occured in the first run to be shown
in the error list after the second run. Now, the errors are cleared
before the second run.
Although I do not have a reproducible example at hand, I imagine
this situation would occur if a rerun is required and there is still
an error after the rerun.
Related to #9765.
Before, the exit code for the first LaTeX run was used to set the
flag, which caused an error to be reported when in fact there was no
error on the second run.
This fix ammends 1dbf0e5a.
When importing TeX code like
%comment
TEXT
the LyX document currently becomes
[ERT %comment] TEXT
so that TEXT is now part of the comment.
Now output_comment adds a trailing newline if the token after the
comment is not a newline. Note that the newline that marks the end of
the comment has already been parsed at this point.
tex2lyx tests have been checked manually and updated.
Fixes ticket #9551.
A plausible scenario is that change tracking is used together with a versioning
system. In this case, parallel modifications might remove an \author line on one
side, and add another change of this author on the other side. This scenario
causes a bad merge after which the added change has no associated author. In
this case, LyX used to display a list of errors on opening and deliberately
removed the corresponding change tracking information.
* If ever a tracked change refers to an author that does not exist, then add a
dummy author. This dummy author is not saved to the file afterwards.
* Have a very clear error message on opening such a corrupt file.
This is better because it implements a LRU cache. Indeed, while editing in particular, width of many different strings has to be computed. This is different from the previous situation where only width of single characters was computed and cached.
This popped up in cmake monolithic build once. It looks like BufferView is
included indirectly by some of the other headers (otherwise we would have seen
compile errors for other build configurations as well), bu I'll keep the
forward declaration since we don't want to depend on this indirect header
inclusion.
N_() is a preprocessor macro to mark translatable static strings. It is not a
good idea to also name a class member variable N_: It did only work in the
other build configurations because gettext.h was not included.
Updating all previews (even if only one has changed) is more costly
than I thought. Thanks to Guillaume for tracking down this
performance issue.
This reversion is related to the reversions at 358745d0 and
a7a14395. See also #7242 and #9855.
This reverts commit 29948eec26.
Updating all previews (even if only one has changed) is more costly
than I thought. Thanks to Guillaume for tracking down this
performance issue.
This reversion is related to the reversion at 358745d0.
See also #7242 and #9855.
This reverts commit 66f527e417.
Prevent encoding changes whenever the TeX engine is XeTeX or LuaTeX,
as XeTeX/LuaTeX use only one encoding per document:
* with useNonTeXFonts: "utf8plain",
* with XeTeX and TeX fonts: "ascii" (inputenc fails),
* with LuaTeX and TeX fonts: only one encoding accepted by luainputenc.
+1 no needless encoding switches
+1 runparams.encoding matches the correct encoding at any time
+1 less complicated code.
-1 there may still be problems with CJK (possibly impossible to
solve for Xe/LuaTeX with TeX fonts).
For LuaTeX & TeX fonts, the complete document uses the encoding
of the global document language.
See also #9740.
The referenced commit lead to performance issues and possibly
crashes in some cases when saving preferences. Thanks to Guillaume
for discovering the issues. See #9828 for more discussion.
I hope to come up with a better fix for #9507.
This reverts commit 390ae05444.
LyX format, create a backup of the original file. We put it in the backup
directory, if one exists, otherwise in the directory the original file is
in. This is the same strategy as for normal backups. Basically, the only
diferences are: (i) what name we use and (ii) we do not over-write any
backups that may already exist.
Actually, the changed tests were used to prevent overwriting the encoding
changed in Buffer::writeLaTeX with a language-default encoding.
This is still required for XeTeX with TeX-fonts unless a proper solution is found.
Documents with more than one encoding and TeX-fonts fail with LuaTeX,
as "luainputenc" can only handle one encoding.
With inputenc == "auto" or "default", the encoding changes with
the language and must be reset after an eventual language switch in insets
or environments (see #6216).
However, whether we need to do this does not depend on 8-bit TeX vs. LuaTeX
but on the possible use of more than one encoding for the document.
With "nonTeXFonts", the encoding is utf8,
LuaTeX with TeX fonts requires encoding handling similar to 8-bit TeX.
(Additionally, the value of "params.inputenc" could be tested: if it is
not "auto" or "default", we have just one common encoding and could skip
the reset as well.) Not sure how much time this saves, though.
This is one part of bug 9744: If you toggle between TeX fonts and non-TeX
fonts, the settings of the other choice are no longer thrown away, but stored
and re-activated if you switch back. Most parts of the patch are purely
mechanical (duplicating some BufferParams members), the only non-mechanical
change is in the GUI logic.
and record an undo call. This deals with the problem of marking
the Buffer dirty, as well.
Move LFUN_BUFFER_TOGGLE_COMPRESSION and LFUN_BUFFER_TOGGLE_OUTPUT_SYNC
to Buffer.cpp, and add an undo call.
In computeRowMetrics, the right margin of the row shall be taken in account when the text is not left-aligned. Rewrite the code in terms of the available width `w' to avoid this pitfall.
The pointer macroInset points to a vector element. When another element is inserted in this vector, some reallocation occur and the pointer points to a deleted element.
This does not crash LyX by default, but it is bad enough to make valgrind cry.
See ticket #9804.
There was a problem with end of paragraph markers. By design, wide insets choose their size without taking in account the marker. This should not lead to unneeded horizontal scrolling.
Part of ticket #9807.
PDF properties (Author, Title...) need to be quoted properly when writing (normal operation and tex2lyx). Also, reading them requires to use Lexer::getString() directly, because >> uses next(), which does not handle escapes.
Fixes bug #9830.
hyperref expects LICR macros for non-ASCII chars in the PDF Header Information.
As hyperref provides good coverage for \inputencoding{utf8}, we try
this if the current input encoding does not support a character.
With XeTeX, we do not load inputenc and cannot use \inputencoding.
However, utf-8 works out of the box so we can write the content in UTF8.
The \jobname macro is redefined for the preview snippets such that
it contains the name of the document without extension. This is
harmless because the moment it is redefined, TeX has already chosen
the name of the output and log files. On the other hand, any use
of \jobname in preview snippets is now consistent with what one
obtains after exporting and compiling by hand.
Other than BIBINPUTS, also BSTINPUTS and TEXFONTS are exported.
They do not replicate the setting for TEXINPUTS but are set such
that the current dir (i.e., the temp dir) and the document dir
are also searched for bibtex and fonts related files.
The code in InsetTabular disables the toolbar icons and menu entries related to vertical lines.
The code in GuiTabular disables vertical lines in the Settings dialog.
Note that there was a bug in
GuiSetBorders::set(Left|Right|Top|Bottom)Enabled, where the borders
did not get drawn correctly.
Fixes bug #9816
Rewrite the logic completely:
* fix cases where the offset was reset unnecessarily
* fix cases where the row was scrolled too much: as soon as a side of the row is completely visible, there is no need to scroll more.
* fix cases where offset would never reset
This partially reverts commit 0f1fdaaa9, but only redraws the current
row when the value returns by getPos is obviously false. It would also
be possible to check whether the current cursor has all its insets in
cache instead.
A better fix should be found, but this is unfortunately not 2.2 stuff.
Fixes bugs #9796 and #9812.
The execution took some time and the messages were delayed
until all test files were processed. Now as each testfile
is processed, the appropriate messages are printed.
There is also a new target: cleanupdatetex2lyxtests
It is the same as updatetex2lyxtests, but removes also
some extra created files from the source directory.
Cmake versions prior to 2.8.11 don't know this command, but
our minimal requirenment is 2.6.4
Thank to Vincent, it is replaced by using 'if(file1 IS_NEWER_THAN file2)' comparision.
Sometimes when compilation fails a PDF file is created but is empty
and the viewer gives an error when trying to view it. One such
example is compiling the current lib/examples/PDF-comment.lyx file
with TeX fonts and the following version of LuaTeX:
beta-0.80.0 (TeX Live 2015) (rev 5238)
The wordSpacing property of a QFont object applies only once when there are multiple spaces between words. Therefore Row::Element::countSeparators shall not count spaces, but groups of spaces.
Fixes bug #9808.
This is a tiny simplification that makes understanding the code more easy and
will help in fixing the remaining regressions. The logic remains the same, no
export test result is changed.
Starting with Qt 5 the modifiers aren't reported correctly for disabled swap.
Until this is fixed it is corrected by reverting the modifier mask to check for.
The horizontal alignment changes caused a regression for documents in older
formats. This could have been fixed by adding the following lines to the
format entry for 489 in development/FORMAT:
Previously, the horizontal position was ignored except for the
following parameter combinations:
- fixed width and type Boxed and without inner box
- fixed width and any type and with inner box and with makebox
Now, it is also used for the parameter combination below:
- makebox is not used and not (type Boxed and without inner box)
and a corresponding conversion to convert_BoxFeatures() in
lib/lyx2lyx/lyx_2_2.py. However, it was decided to revert the box alignment
changes instead: The box alignment can contradict the paragraph alignment. A
better way to avoid the additional space that can be created by paragraph
alignment (which was the motivation for implementing box alignment) is the
same as for table cells (see noTrivlistCentering() in src/Paragraph.cpp).
This would be a file format change and it is too late now for 2.2.0.
See http://www.mail-archive.com/lyx-devel@lists.lyx.org/msg188147.html for the
complete discussion (subject "Regression in lyx2lyx box alignment").
Fixes output for 3 of the 4 test lyx-files.
Includes "FIXME"s at places where further action is required to get the XeTeX
export right but I don't know how.
Remove member disp_ that is defined both in Cursor and CursorData classes
Pass parameter of convert<T>(docstring const) templates as const reference for performance.
b1c68dccf8 and 46aed6d2b9 fixed some language nesting issues, but introduced
a regression for the case that there is a standard paragraph in a foreign
language, followed by a list (e.g. itemize) in the same language, followed
by the end of the document, as e.g. in lib/doc/de/Additional.lyx. The reason
for this was that not all language ending commands did reset
state->open_polyglossia_lang_ correctly.
I am sure that one can still construct broken corner cases, and I am also sure
that this was already possible before b1c68dccf8 and 46aed6d2b9. However,
this fix seems to fix the most important issues, and to get nesting completely
correct we would probably need some stack-like structure, for languages and
encodings, also for the CJK part (which is not touched at all by this commit).
The correct mappings break the tex2lyx roundtrip tests because both,
the tests and tex2lyx use the wrong mapping (\b for "combining minus below").
Fix tex2lyx and the test files so that round-trip tests pass again.
The culprit was that the computation of textwidth did not take wordspacing in account.
Also fictor the code so that the pixmap path can use the special RTL handling.
It is not clear however that the handling of left and right bearing works correctly.
We already have a CoordCache of insets dimensions. It is not necessary
to store the same information in two places.
Give a name to CoordCache tables types to improve code readability.
Remove ParagraphMetrics::singleWidth, which is not used anymore.
Please use the new function otexstream.append(str, texrow) to append an
odocstringstream with texrow information to the output when outputing to a
string buffer (e.g. case of subcaptions).
This deep copy used to mess with the unique identifier: what TexRow saw was
different from the original uid. There may also be performance improvements.
(Using Georg's suggestion)
These features are active in DEVEL_VERSION when Debug is set to LATEX.
1. The TexRow information is prepended to the source panel.
2. Clicking on any line in the source triggers reverse search. (This would be an
interesting feature to implement on the user side, but we need a proper LFUN.)
WriteStream is now built from an otexstream instead of an odocstream, and
therefore counts lines in a TexRow. Calls to TexRow are added in relevant places
in math insets.
This finishes adding line tracking for math in the source panel and for forward
search.
This is preliminary work for extending the cursor<->row tracking to math.
TexRow used to associate, to each row, a location id/pos where id determines a
paragraph and pos the position in the paragraph.
TexRow now associates to each row a list of entries, text or math. A math is a
pair uid/idx where uid will determine a math inset and idx is the number of the
cell.
The analogy id/pos<->inset/idx works better than the analogy id/pos<->idx/pos,
because what matters for the TexRow algorithm(TM) is the behaviour in terms of
line breaks.
This only improves the source view and the forward search, not the error report
and the reverse search (though this could be easily added now).
The id is just the memory address.
The status bar now spits out the math inset uid information when in a math cell
in DEVEL_VERSION, like it already does when in a paragraph.
This is preliminary work for extending the cursor<->row tracking to math.
otexstream used to count lines to build a TexRow, and some other things. The new
class otexrowstream has the line counting feature of the previous otexstream
without other stuff. otexstream is now a subclass of otexrowstream that has the
same features as before.
This is preliminary work for extending the cursor<->row tracking to math.
The code that sets open_polyglossia_lang_ is not only executed for polyglossia,
but also for babel, so we have to use the correct language end command.
The context menu did no longer work for some insets, since it requires the
cursor to be in front, and editXY() is also used to determine the inset for
the context menu. Now the cursor is corrected if needed.
This was made visible by aab1b145a5, since xhtml export for
lib/doc/Additional.lyx caused an exception. However, the cause for this was
already present earlier: All attempts to output a std::string to an
odocstream resulted in trying to change the encoding of the stream instead,
since there is no operator<<(odocstream &, std::string) defined, and an
implicit conversion to SetEnc happened instead.
This is fixed by making the SetEnc constructor explicit and adjusting all
code parts that did not compile anymore after that. The parts of the code
that did use the wrong output operator were the std::string version of
htmlize() from output_xhtml.cpp and all changed parts in the other .cpp files.
I also removed the std::string versions of html::htmlize() and
html::cleanAttr(), since it was difficult to see which encodings were used
with these. Now we are always explcit when using html::cleanAttr() and
html::htmlize().
Previously, if one clicked onto a large non-editable inset like the new LyX
logo inset, the cursor was always positioned in front of the inset, even if
the click was almost at the back edge. Now the cursor is positioned at the
correct edge. I tested this also with RTL contents, where from means right
and back means left, but the inset anchor position anchor point is still
at the left, and the right edge is dim.wid pixels to the right of it.
The rule-of-three says that if any of virtual destructor, copy constructor
or assignment operator needs to be manually implemented, then all three
should be implemented. Otherwise you can get subtle bugs which can be
difficult to find. In the changed classes, changing a copy-construction to
an assignment would have had surprising effects. Now they all behave
consistently.
