Useful to cache copies of objects, including shared_ptrs. No risks of dangling
pointer, and avoid naked pointers in the source.
Fix memory leak when compiling with Qt5.
For biblatex, we need the file name as it was entered in the inset
(abs path, rel path, only file name) in order to resolve it properly
for the export (as in the BibTeX inset intself).
Therefore, we now store a pair<docstring, FileName>.
A FileNamePairList has been added to FileNameList for this purpose.
Fixes: #8831
This introduces
* a new inset member isPartOfTextSequence() that returns
whether the inset produces something visible at the current position
in the text flow
* an isOpenPunctuation() test that returns whether a character is in the
'Punctuation, Open' unicode class. This is used instead of just checking
for two (Western, ASCII) opening brackets
It also fixes the isChar() and isLetter() value of InsetSpecialChar,
since some types have not been assigned correctly.
The QTextLayout handling is terribly slow on Qt 4.8.7, but some
caching has been added in Qt5 that makes it much faster. For some
reason, it is not that slow with Qt 4.8.1.
Caches are introduced for the three following methods
* width(doctring), controlled by CACHE_METRICS_WIDTH. This cache already
existed, but the code has been cleaned up
* getTextLayout, controlled by CACHE_METRICS_QTEXTLAYOUT (disabled by
default on Qt5, which does its own caching). This is used for pos2x
and x2pos and now for drawing of text too. The previous code used a
trivial caching scheme of the last used QTextLayout, but now they
are properly kept in a QCache. Moreover, the cacheEnabled() property
is enabled for these QTextLayout object (not sure what this does).
* breakAt, controlled by CACHE_METRICS_BREAKAT. This is the only user
of QTextLayout which did not have some kind of caching already.
For some weird reasons related to Argument-dependent look-up, the
qHash(docstring) function has to be defined in std namespace, since
lyx::docstring is actually std::basic_string<wchar_t>.
[NOTE: this version has profiling hooks, enabled by commenting out the line
#define DISABLE_PMPROF
that should eventually be removed.]
In fact having an extra parameter "bool cond" is no longer useful because it can
now always be emulated with a ternary operator:
Changers dummy = cond ? do_change() : Changer();
A static local variable is guaranteed to be initialized only once, and in time.
Lambda expressions can be used to perform complex initialization of those static
variables on the spot.
(starting from: gcc >= 4.8, msvc >= 2015)
The function is no longer used in LyX's sources (as of the previous
comit, 9b64d7bd) and is thus removed with this commit. Perhaps the
advantage this function had over other path functions we have has
disappeared over time (see e.g. 1a7b7f65).
The only exceptions are:
- The purpose of the header is to drag in the used symbol, e.g. unique_ptr.h
- The used symbol is inside a class or a namespace other than lyx
The reason for this is that global 'using' statements effectively forbid to
use the used symbols in any other namespace in the whole program, since simply
adding or removing an #include of the corresponding header subtly changes the
name lookup. The namespace lyx is sort of global, so it should not have these
statements either.
RefChanger temporarily assigns a value to a non-const reference of any
kind. RefChanger provides a flexible and uniform generalisation of the various
scope guards previously derived from the old Changer class in MetricsInfo.h.
As before, a temporary assignment lasts as long as the Changer object lives. But
the new Changer is movable. In particular, contorsions are no longer needed to
change a private field. Special code can be moved into the appropriate classes,
and it is no longer necessary to create a new class for each specific use.
Syntax change:
FontSetChanger dummy(mi.base, value);
-> Changer dummy = mi.base.changeFontSet(value);
New function for generating arbitrary Changers:
Changer dummy = make_change(ref, val, condition);
Bugfix:
* Fix the display of \displaystyle{\substack{\frac{xyz}{}}} (missing style
change).
Boost.Signals is deprecated. This fixes bug #9943.
The only thing left to do is to rewrite (or get rid of) the boost -mt test
in config/lyxinclude.m4 not to use signals anymore.
It is no longer needed to create fake copy constructors and assignment and to
deal with deletion by hand, thanks to unique_ptr, the inference of move
constructor and assignment operator, and the compatibility of standard
containers with movable objects.
Two better ways of making a class non-copyable in C++11:
* Store the p. impl. in a unique_ptr (for the cases of classes with p. impl.),
or:
* Define publicly the copy constructor and assignment as deleted
Lots of other classes could be cleaned up in this way.
As discussed on the list. If no C++11 compiler is found configuration stops
with an error. There are now unneeded parts of boost, the will be removed in
a second commit.
It turns out that it did not take off since introduced in 2011. It is better to remove it and the associated boost headers (extract.sh was run against boost 1.60 to do the update).
Since we will move away from several boost classes when transitioning to C++11, it is good to start by removing lesser used ones.
It is easier to use instead getVectorFromString for the use we have of this tokenizer. The two places are environment.cpp (path stuff) and qt_helpers (file fileters). The new code is much shorter.
This allow to remove boost/tokenizer.hpp and friends from our boost tree.
Compiling different parts of the sources with different WINVER may lead to
subtle and hard to detect problems. Better use the same value everywhere.
The existing error message suggests that this was wanted anyway, and it
fixes a compiler warning when cross-compiling for mingw on linux. Our code
does not require a specific value, only a minimum value of 0x5000, which
means the resulting executable will require at least Windows 2000.
The included iconv should not be used on Linux or OS X, but (depending on
local configuration) it might be needed for crosscompiling a mingw target
from Linux. Now the user can choose whether to use the included iconv or not.
cmake does already support that.
eilseq.m4 was taken from the original libiconv 1.14 package.
On mingw-w64, long long (64bit wide) is larger than long (32bit wide).
Therefore we need some more specializations for string, docstring,
otextstream and << overloaded ostream functions. The configuration code
is by me, the source code changes by Shankar Giri Venkita Giri (bug 10053).
This is needed for warning-free compilation with mingw-w64, and does not hurt
for other build configurations. Patch by Shankar Giri Venkita Giri (bug 10053).
This fixes a failing unit test with 32bit gcc 4.9.3 and -O2 optimization:
It computed 9953 instead of 9954 for Length::inPixels() of value 2342.
The reason for this is probably different rounding behaviour caused by storing
the unrounded value in a processor register (uses 80bit accuracy) vs. writing
it back to memory (uses 64bit accuracy). The unrounded value is very close to
9953.5 (which is not representable as an exact IEEE floating point value).
Apart from that, having a proper function for rounding makes the code more
readable, and has the nice side effect to make Length::inPB() work for
negative lengths as well.
We open the input file now twice: The first time in latin1 encoding to read
the document encoding from the preamble. This does always work, since
traditional TeX does not allow non-ASCII contents without an encoding changing
command (except for comments, but we do not need them, and using latin1 rather
than utf8 ensures that they do not produce an iconv exception, but are simply
recored with wrong characters), and we do detect the utf8 based TeX engines
XeTeX and LuaTeX as well. The second time we open the file directly with the
document encoding.
This fixes a few tex2lyx tests on OS X, since changing the encoding of an
open file steam does not work with clang on OS X. Files using more than one
encoding are still broken, but all single-encoding files are fixed now.
Changing the codecvt_facet of a file stream after the file has been opened
does not work with clang on OS X. Therefore we avoid it if possible (i. e. the
new encoding is the same as the old one).
External processes cannot access files which are open in LyX. Therefore the
temp files created by the external inset need to be closed right after
creation. The symptom was that the date inset did not produce any outout on
windows (bug 9925). This change reverts a small part of f09a9fe2.
Although the date inset is unimportant and will probably be removed, this
change is important for all external insets that make use of temp files.
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 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.
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.
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.
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().
These were found by cppcheck:
Member variable 'x' is not initialized in the constructor.
The crash #9788 would not have happened if this had been done earlier.
Use the function support:truncateWithEllipsis() to shorten a docstring with
... at the end. Actually we use U+2026 HORIZONTAL ELLIPSIS instead of "..." when
automatically shortening strings. This is to be consistent with Qt's own
truncation and is much nicer on the screen.
This includes the bugs #9575 and #9572 regarding broken text elision in the
outliner.
Known issues (non-regressions):
* TocBackend::updateItem() should be rewritten to update all TOCs. (#8386)
* "..." should be replaced with … everywhere else on the interface (including
translation strings).
* We should prefer to rely on QFontMetrics::elidedText() to truncate strings
with an ellipsis whenever possible, or an equivalent for the buffer view
dependent on the font metrics. See the warning in src/support/lstrings.h.
These were all flagged by "(style) The scope of the variable 'x' can be reduced."
Narowing the scope improves readability, and if it is in a loop then the
compiler will be clever enough to produce efficient code, we do not need
manual optimization for POD types.
These were all found by cppcheck. Even in constructors that are there "only
because of std containers" the class should be initialized correctly. You can
never know whether such an object does not get used, and then a nice crash
caused by dereferencing a NULL-pointer is better than undefined behaviour.
The 'using namespace std' at the top of the file makes it quite difficult to
understand which abs is used: double std::abs<double>(double) or
int ::abs(int)? Now it is explicit, and the code does not change in subtle
ways if somebody removes the using statement.
This replaces tests for __cplusplus >= 201103L, which are wrong with gcc 4.6 and earlier. Indeed these versions of gcc define __cplusplus = 1.
Reference:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=1773
Newer boost versions use complicated type traits for boost::next and
boost::prior, which do not work with the RandomAccessList iterators.
The long term solution is to use std::next and std::prev, for now supply
simple replacements for compilers that do not support C++11 yet.
This fixes hopefully the compilation of check_trivstring on cygwin and
solaris. Previously I did not use the explicit instantiation, because c_str()
is used by other explicitly instantiated methiods, but for some reason this
does not seem to suffice.
