Instead of annoying the user with an automatically created note in the output
document which she needs to delete manually, determine the document language
automatically for documents that use CJK. This is done using a heuristic which
roughly counts the number of characters in each language and sets the one that
is used most often. This is not perfect, but it works for the two major use
cases: A document with only some CJK parts (in this case the babel language is
used), and a document which is mainly written in one CJK language. It is only
a minor problem if the heuristic is wrong, since the TeX export is still
correct, and there is no spell checking support for CJK anyway.
Now all regression tests do pass except for some relative path issues
depending on the location of the build directory.
The previous output was read correctly by LyX as well, but the new version is
more similar to the files written by LyX. Also update TODO with recent file
format changes.
Commit 7cfac95 got rid of empty lines that were created by removing \usepackage
statements. However, it added an additional newline in case the \usepackage
was not at the end of the line. This is now fixed.
A dummy getGuiMessages function was missing. Instead of adding it everywhere, a new file is created that contains all the dummy functions needed by the tests.
The old fix was incomplete (\verb~\~ was translated to \verb~~ in roundtrip).
The real cause for this bug (and also the mistranslation of \href{...}{\}})
was the misbehaviour of Token::character() (see comment in Parser.h): This
method even returns a character if the category is catEscape, and this is not
wanted in most (all?) cases.
If this option is given, included files will be copied to the output directory.
Also -roundtrip is now allowed with given output file.
-copyfiles is useful if you want to ensure that no file (not even an included
one) is overwritten by a subsequent export from LyX. Both changes are needed
for unit tests that do not write to the source directory.
This adds an optional 'set' argument to the language lfun and reintroduces toggling.
Additions by me reintroduce the possibility to reset to the document language via 'language reset' or just 'language'
This might encourage users to post this information when reporting bugs
and sometimes it's useful to be able to copy the paths of the Library
and User directories.
The input and output file names of tex2lyx may be relative. In this case,
getMasterFilePath() and getParentFilePath() return relative paths as well.
Now the file name translation logic for all kinds of included files can
cope with that.
If the WA is the last one showing a buffer, then the buffer may either be
closed or kept hidden, or the user is asked. The behaviour is controlled
by a new preference option.
For discussion, see http://comments.gmane.org/gmane.editors.lyx.devel/142638
With non-TeX fonts, you can select a 'Non-TeX Font Default' math font, which simply loads unicode-math without actually selecting a math font, this then uses the default math otf font, currently Latin Modern. Other fonts still need to be set manually in the preamble, via \setmathfont.
The implementation suppresses unneeded package requests from unicodesymbols, but the output still uses macros instead of full unicode (both is possible with unicode-math).
The whole thing is a proof of concept, and it needs to be tested. I have tested it with the math manual, which compiles and seems to display correctly if I remove some hardcoded package loadings. OTOH I have not much experience with math.
This addresses #7449 partly.
In 0aead96461 it was introduced that the toolbar popup menu buttons are not
initialized until they are being used (i.e. pressed at for the first
time). As a consequence, these buttons are always enabled on start. Now
that we have the sticky button for view and update, I'm now annoyed by
this every time I start LyX.
This addresses #6543 by adding an option to prevent fonts such as Palatino and Times to automatically adapt the math font (IOW it lets you load the text font only for a bunch of fonts where this is easily possible).
Furthermore it adds an interface to select a specific math font, which is defined in latexfonts. Currently, this is only euler (the only one I know), but if there are other math-only tex fonts, they can be added easily (but note that this changes the file format).
Non-TeX math fonts are not yet supported. Eventually, unicode-math support can use the existing UI, but this is not on my agenda.
The LaTeX font now do not specify simply alternative packages or packages for OT1 encoding etc., but they refer to complete AltFonts (which are not directly accessible via the GUI). This way, alternative fonts can also have options (osf, sc etc.), and they can use all sorts of initializing methods (\usepackage, \setrmfamily etc.).
This assures that refs to existing labels are not being touched if a new label with a duplicate name is created, and the name then changed by the duplicate checker.
Branch candidate.
Fix thinko in Undo::endUndoGroup.
Do not reset undo groups in Undo::clean
The underlying problem remains: the various VC functions should not
always reload the buffer or at least should use the parameter
clearUndo=false.
Filenames embraced in <...> can occur anywhere on the line and multiple times. This fixes for me the case that graphics included via ERT were not tracked. It probably also fixes#8336.
This is a candidate for branch.
GuiTabular.cpp: a cell can either be a multicolumn or multirow, not both the same time, this is already respected by the toolbar and context menu, only the dialog allowed to set this although this was correctly never applied
should also go to branch
The stderr message "Object::disconnect: Unexpected null parameter"
often appeared when closing and opening buffers.
GuiView::on_currentWorkAreaChanged should not pass a null pointer
to Object::disconnect.
Fixes 2 issues:
1. LyX uses for a decimal alignment a multicolumn and having for a cell a multicolumn _and_ a multirow is invalid LaTeX.
2. It was impossible to unset a decimal alignment via the context menu or toolbar button.
The previous scheme of loading all possible translations and checking
whether the work is a bit too much "brute force" and causes problems
on Mac OS X (documents loaded with the wrong language).
Now there is an helper static method in Messages class that checks
whether a readable .mo file exist for the language. There should be an
API in gettext for doing that, but alas it is not possible.
As a consequence the method Language::translated() has been removed,
along with its cache.
This is the same protection used for \cite{}. Although "ulem" commands
don't strictly need this protection with \ref{} and friends, this helps
when one wants to use another package souch as "soul".
Indeed, ulem's \uline{} doesn't hyphenate words, while soul's \ul{} does,
and thus the output is much nicer.
With this patch, replacing ulem with soul is as simple as adding
\usepackage{soul}
\renewcommand{\uline}[1]{\ul{#1}}
to the preamble, without worrying about manually enclosing in \mbox the
\ref commands. In any case, the output from ulem remains unchanged.
Another possibility offered by soul is that of typesetting as
"yellow highlighted" the underlined text. This is achieved by adding
\usepackage{color,soul}
\renewcommand{\uline}[1]{\hl{#1}}
to the preamble.
The mhchem package treats the caret both as a shorthand for \uparrow or
as a superscript operator according to whether it is surrounded by
spaces or not. The \ce and \cf insets allow inserting spaces but there
is no provision for inserting a space after the caret, which is always
considered by LyX as a superscript operator. The solution here is to
insert a space after the caret if the superscript is empty or an empty
brace inset.
Mathed does not allow empty superscripts, so an empty brace has to be
inserted when working in LyX. On the other hand, when importing latex
code, an empty superscript is retained.
This has no effect whatsoever for normal latex code, as a space after
the caret is ignored. In any case, the output is only changed if an
empty brace inset is used as superscript. Specifically, the output is
changed from "^{{}}" to "^ {}".
The code in DocIterator::sanitize now follows more closely the
previous StableDocIterator::asDocIterator code. In particular, it adds
the slices one by one, since fixIfBroken will chop the cursor
otherwise.
In order to interact with native osx applications, AppleScript support is a plus.
Here is a patch that makes LyX respond to a simple command (run) and that allows to communicate with LyX as with the LyX client.
Example of use:
tell application "LyX" to run "server-get-filename" with argument ""'
returns
message:/Users/bpiwowar/newfile1.lyx, code:0
with a message and the error code
This bug occur only at certain Qt versions and systems and we do
not really understand what is going on.
Quoting Stephan - the parent path of "." e.g. never should be ".".
The inconsistency between the name of FileName and the internal QFileInfo
state looks very dangerous.
The LaTeX font information are now centralized and outsourced. This removes a lot of hardcoding and duplication and makes it easier to support new LaTeX fonts.
Normally, we do not load babel when English is the only language. However, if we check if babel is used from Font::validate, we need to surpass this test, else no secondary language is registered if English is the main language.
The forward flag is used to place the cursor behind the replaced text if it's true.
But it's not correct to move the cursor if it's false. The cursor is in front of the
replacement already after the replaceSelectionWithString() was done.
* Assure that really no language package is called when none is selected.
* Only call global custom package if global is desired.
This is a candidate for branch as well.
The previous scheme of loading all possible translations and checking whether the work
is a bit too much "brute force" and causes problems on Mac OS X (documents loaded
with the wrong language).
In the new scheme, autotools install a file lib/installed_translations that contains a list of installed languages (the .gmo files that got installed). This file is read
in Languages::readInstalledTranslations and allows to set the translated() property
of each language.
Fixes bug #7111: Assertion with undo and InsetBibitem
* add Paragraph::brokenBiblio(), with tells whether there is something to fix.
* rename Paragraph::checkBiblio to fixBiblio; simplify it greatly by using
InsetList methods
* In TextMetrics::redoParagraph, call recordUndo before fixBiblio (only when
there is something to fix obviously)
Special care is taken to update cursor only when it makes sense. Note
that there are cases where undo information is not recorded (current
cursor not in the slice where action happens)
Babel settings might contain @ characters. Check for this and wrap the settings into \makeatletter ... \makeatother switches if necessary (see discussion at #8265).
The code does a reload on the buffer to avoid loose pointers if the
file changes directory. This commit adds a bool parameter to
Buffer::reload to allow for keeping the undo stack intact. This is of
couse only wanted in this case, not when reloading an earlier version
of the file.
In the existing code for setting layout of pragraph parameters,
Text::undoSpan is used to "guess" which paragraphs should be saved in Undo.
With this patch the approach is more precise: before every explicit change
to a paragraph, a Cursor::recordUndo call is inserted. This is much more robust than trying to guess.
In particular, we do not look at depth changes at all, since they are now
handled in updateBuffer since #8159 has been fixed.
The goal here is to get rid of the old code that modified variables
LANGUAGE and LC_ALL, therefore creating the problems mentionned
in the ticket.
In the new system, there is no explicit "GUI" message handler, that
needs to be reset at each language change. Instead, getGuiMessages
calls getMessages with the correct parameter. This allows to simplify
greatly the code and to remove a lot of old cruft.
This commit does a bit more than fix selection, since it saves the full cursor state
in the undo stack. This means that undo now restores:
* the selection
* the current font
* transient mark (shall we keep this one?), logical position...
In order to do that, it introduces an intermediate class between Cursor and DotIterator: CursorData.
The new inheritance diagram is thus
DocIteraator -> CursorData -> Cursor
CursorData contains all the members of Cursor that define the current position, but not the stuff
related to current view of dispatch mechanism. It may make sense in the future to move members
between CursorData and Cursor and to move some member functions to CursorData.
Now UndoElement uses CursorData for cur_before and cur_after, but not for the cell. The undo API uses
also CursorData instead of DocIterator.
If there was an encoding set by the inputenc package, it must not be ignored.
Now all tex2lyx test cases can be exported to .tex again by LyX (although
there are still some differences in the .lyx output).
It is invoked by 'make check' (automake only, it would be nice if someone
could add it to cmake as well), or by calling
python src/tex2lyx/test/runtests.py <path to tex2lyx binary>
by hand. Currently, it does not compare the output (this comes later).
The added .lyx files are from tex2lyx around mid of april, so that you
can see the regressions of the current version if you run the test
yourself (simply run git diff afterwards).
The home made test runner is quite stupid, but better than nothing.
Feel free to improve it or replace it with something better, as long
as running it stays as simple as now.
- tex2lyx/text.cpp:
- fix bug that swallowed valid braces
- the encoding is "Bg5" not "BIG5"
- add and update comments
- CJK.tex: add an example for the Bg5 encoding
- Preamble.cpp/:
- move list definition out of anon namespace
- declare lists and a function in the class
- text.cpp: remove the now already defined list
If someone can tell me how I can tell QtCreator to actually use tabs, not spaces, I would be most grateful (yes, I have set Tab Policy to Tabs Only in Prefs)
- Parser.cpp: \verb can have any character as delimiter (except of ASCII letters) not only '+', therefore partly revert [3943b887/lyxgit] and fix it for all cases
The idea is to record undo at the place where the document is modified:
1/ in Buffer::updateBuffer, add a recordUndo, with the caveat that a
const_cast has to be used (because updateBuffer is const but
modifies the document, go figure).
2/ in GuiApplication::dispatch, add an extra undo group that
encompasses the updateBuffer call. Some other undo groups may be
redundant now, but it is not a problem since they do not cost
anything.
The button text of InsetInclude insets shows whether the child document is
included or excluded from compilation. Changing this for a child document
in the document settings does not get reflected on screen. This patch
updates the button text on the updateBuffer() call.
Preamble.cpp:
- support the font scaling
- improve coding style
text.cpp: - improve logic/coding style
XeTeX-polyglossia.tex:
- add scaling
- change document so that it is compilable using the Libertine font
Instead of adding the extra '\n' to tex2lyx, we should remove it from
BufferParams.cpp because it was introduced there unintendedly in d7eaaba9
(Juergen Spitzmueller; "support for CJK font argument"; Jun 5 2008).
This partly reverts commit c2810f922665c591a7c32d4c679e58e55132505a.
Fix commit [aa9fed93/lyxgit]: it is never a good idea to call substr(4) on a string without checking its length. Moreover, it is even better in this case to check that the first letters are "text" as the reminder of the code assumes...
- tex2lyx/Preamble.cpp/h: not all new header definitions since fileformat 413 have yet been added to texl2yx, so do it now; also change some new default values
- bugfix for the parsing of polyglossia's paragraph environments
- new support for polyglossia's language change commands \textxxx
- XeTeX-polyglossia.tex: add more testcases and add a missing \end
This allows to workaround the problem in #8198 - on some systems
we get dead caron as a single character instead of whole combined
character. I suppose this has rather something to do with X settings
than LyX per se.
In this version, the idea is to record undo at the place where the document is modified, which is definitely cleaner.
1/ in Buffer::updateBuffer, add a recordUndo, with the caveat that a
const_cast has to be used
2/ in GuiApplication::dispatch, add an extra undo group that
encompasses the updateBuffer call. Some other undo groups may be
redundant now, but it is not a problem since they do not cost
anything.
http://marc.info/?l=lyx-devel&m=133876924408431&w=2
The problem here is that the copy_params() routine in FindAndReplace.cpp
created a new DocumentClass, but it never updated its Buffer to reflect
that new DocumentClass. So its Paragraphs still contained points to the
Layouts in the old DocumentClass which, since ead697d4b6, gets garbage
collected once it is no longer needed. So the Layout doesn't exist, and
we crash.
For some unknown reason, the paste code likes to call Cursor::setSelection(), which is not appropriate here since we are not trying to set a selection at this point.
objects. The problem that led to the leak is that these objects can be held in
memory long after the Buffer that created them is gone, mostly due to their
use in the CutStack. So they were previously held in a storage facility, the
DocumentClassBundle. Unfortunately, they were now being created too often,
especially by cloning. It's not really a leak, because they're accessible, but
we weren't ever destroying them.
This new approach uses a shared_ptr instead.
Thanks to Vincent for pointing out const_pointer_cast.
While cppcheck did not turn out any suspicious error messages, using
the "performance" flag highlighted several nitpicks in three categories
* do not use it++ for iterators, ++it is better
* do not use size() to test for emptyness, empty() is here
* do not use "const T" as a function parameter, "const & T" is better
I doubt that any of these is a real performance problem, but the code is cleaner anyway.
When there is a single document open, the tabbar gets hidden. The only and selected tab is still visible though as a break in the line delining the tabwidget. To fix this we paint the tabwidget ourselves and remove the presence of the hidden tab.
When the socket does not exist anymore, we should not try to access it. In
trying to do so, std::map will create a new shared_ptr but this pointer
doesn't point at anything. To prevent this, we explicitly check whether
the socket is available.
This reverts the previous fix in [0a33374c/lyxgit] and fixes it
differently.
We always have to call 'notifyCursorLeaves', but we only have to make sure
that we call the 'fixIfBroken()' functions first.
We rely on the 'or' operator to prevent us from calling
'notifyCursorLeaves' if one of the two cursors is broken. This doesn't
work when using the '|' operator. The compiler 'optimizes' the code in
such a way that we always call notifyCursorLeaves anyway. Using the '||'
operator fixes this.
The undo range is extended to encompass all the paragraph after the end that
are a non-zero depth. The reason is that these paragraphs may see their depth
reduced if the last paragraph sees its depth reduced.
Note that there is a memory cost, since we store paragraphs than may not be
modified in practice. This may matter for some pathological files (people
putting all their document at depth 1 for a weird reason?)
I tried to be careful, but this can introduce new bugs. Please test.
Only user of this function is code that has been commented out, comment
out this function as well.
Signed-off-by: Lars Gullik Bjønnes <larsbj@gullik.org>
The check whether fname != oldname does not work for unnamed buffers.
Therefore, we check explicitly whether the Buffer that already exists with
the same name isn't really the buffer we are trying to save.
Text::outerFont looks recursively for paragraphs before the current one
which has a lower depth. If such a paragraph cannot be found, depthHook
and outerHook return the current paragraph. As such, we end up in an
infinite loop. So, if we find a par_depth that was the same as the
previous one, we apparently can't find a suitable paragraph and we should
quit the loop.
In this case, the working dir has already a slash at the end.
Appending another one would result in a path ending with a double slash,
which has a special meaning when used in TEXINPUTS (all subdirs would
be recursively scanned). So, avoid doing that.
We want the key as id, not the label (which is optional).
We also need a kind of namespace for the citation ids.
We should also clean the id tag before using it.
As long as leftMargin() is not correctly implemented for
MARGIN_RIGHT_ADDRESS_BOX, we should also not do this here. Otherwise, long
rows will be painted off the screen and will not be editable.
This is a part of bug #8081: Translate cells in
\begin{sideways}...\end{sideways} to native syntax. The turn environment
comes in a later commit, since this one os suitable for backporting.
This is not needed, since LyX supports comments in math. Data loss with math
comments containing a backslash in LyX has been fixed as well.
The test case was found in bug #8104.
When using, e.g., a 'mathcal' inset in math, the inline completion and
other special characters like '\#', '{..}' are are painted in the
'mathcal' font as well. This is overcome by setting the mathnormal font
before painted these characters.
If loading a basic layout fails, we try to load a basic layout without
stdclass.inc instead. However, this still 'implicitly' requires
stdinsets.inc. If the load failed because stdclass.inc couldn't be loaded,
we expect that stdinsets.inc can't be loaded as well, so we should exclude
stdinsets.inc as well.
This prevents an assertion if stdinsets.inc can't be found or can't be
loaded. This can happen when the layouts have the wrong format and python
is not available.
If no mask is supplied or the mask is attached to the end of the filename, we end up with unexpected names like
<system-temp-dir>\lyx_tmpdir.qHp780.vcr780_<mask>
instead of a temporary file in the lyx temporary directory like
<system-temp-dir>\lyx_tmpdir.qHp780\<mask>.vcr780.
is intended for us, i.e., if the first argument is "box". Maybe we
should also check to make sure that the rest of the arguments are
good, but we used just to return true no matter what the argument
was, so this is an improvement.
LFUN_INSET_MODIFY is really kind of a mess.
The LATEX_LIST_ENVIRONMENT and LATEX_BIB_ENVIRONMENT type paragraphs are
not yet implemented in DocBook export. However, we need to actively skip
these paragraphs in order not to end up in an endless loop.
If the stream is good (i.e. there are still tokens) and we expect an
argument, we call getArg(). However, if there are only spaces, the stream
suddenly isn't good anymore after 'skipSpaces' and we would get an error
when calling 'getChar'. Therefore we have to check whether the stream is
still good.
If the stream is not good, we don't need to 'putback', because we didn't
read anything yet. If we now do rewind the stream, we are asking for
problems as in bug #8089.
This was introduced in [3cafb856\lyxgit] to fix bug #4318.
If we are in rtl text, we *need* to check whether the first character is a
space. We can't use the rtl variable for this because this is only
computed on the last row.
This bug was introduced in [0a137e31/lyxgit] to fix the computation for
freespacing paragraphs. We better can just check whether the paragraph is
freespacing or not.
The signature of the character in cache should take into account the
change tracking baseColor (author) and mergeColor (added/deleted). If not,
we will mixup different cached pixmaps.
When an InsetText is deleted, all lines in the inset are painted as
deleted. This status should be restored afterwards. If not, insets will be
painted in the wrong color afterwards (e.g. tables).
I had already noticed the above mentioned thinko and this commit
should have been part of [3285ce1d5/lygit] but somehow I managed
to lose it in the git branches game: do a branch for that, than
try a different thing in another branch, then rebase to avoid
merge conficts, and finally delete all unneeded branches. Well it
is too much easy to do the wrong thing and pick the wrong branch,
deleting the right one. I will never follow the git way anymore
and will do I was used to do with svn.
I introduced this bug in r40091 where support for empty dates was added.
I also noticed that the title_layout_found flag was not properly passed down
to all variants of parse_text*. Since this is rather a global variable I
placed it into the preamble class.
The test case of bug #8104 shows other problems as well, but those are no
regressions and will be fixed separately.
OK for branch?
This is a patch from Scott Kostyshak. The problem it solves is as follows:
1. enable continuous spell check.
2. type a misspelled word and press space so that it has a wavy red underline.
3. right-click and choose a suggested replacement word.
4. press the backspace button.
Result: nothing happens. If you press the backspace button again, then it works as normal.
The selection code was added for the benefit of the spellchecker, but the code has been rewritten since then.
* Avoid undo step when using backspace in macro mode
* Use recordUndoInset when entering macro mode (if one enters something like \hline, the outer inset itself will be modified)
* Use recordUndoInset when pasting in an INsetMathGrid (same reason as above)
Both bugs above were due to a missing screen update. This patch
updates the current view after dispatching a lyxserver command
and thus solves both.
The patch is quite strightforward and the only difficulty was due
to the fact that the lyxserver needs the result of the dispatched
command. Now, GuiApplication::dispatch(FuncRequest const &) does
right this, but does not return the result, while
GuiApplication::dispatch(FuncRequest const &, DispatchResult &)
is called by the former but does not return the result. So, I
split the first one, isolating the code performing the update,
such that the second one can also update the current view when
the caller is the lyx server. When the action is initiated by
anything different from the lyx server, the behavior is unchanged.
It seems that parameterless macros are not wrapped into a MathAtom.
Rather than touching the macro code (which is tantamount to opening
a can of worms, IMHO), I prefer this solution.
If a converter specifies the needaux flag, latex (or xelatex) is
always run to produce the needed auxiliary files. This is wrong
because there are documents that can only be compiled with a specific
backend and thus the conversion may fail. On the other hand, even if
the document specifies the backend to be used, LyX ignores this info.
This commit rectifies this behavior by letting LyX run the same flavor
of the latex backend that shall be used for previewing the document
also for producing the auxiliary files.
The listings inset and include inset of type listings use two english terms
that are not localized yet: "Listing" for the caption and "Listings" for the
list of listings (not supported natively by LyX yet). The existing layout
translation mechanism has been extended to translate these terms as well:
1) Support [[stuff]] context in lib/layouttranslations
2) Support BabelPreamble and LangPreamble in InsetLayout
3) Use a InsetLayout for InsetInclude of type listings
4) Define BabelPreamble and LangPreamble in the layouts for InsetInclude
and InsetListings
LyX fails to read the bounding box from an EPS file if it has
negative values. Adjusting the regex will overcome this problem.
Negative values do not pose big problems later on, but the GUI
doesn't handle it correctly yet (see bug #5718).
It is broken since the latest file format update (neither did Uwe run a test
after updating the file format, nor did I after undoing my accidental change).
The reason it was broken: the bool version of write_attribute() is chosen
for "0" instead of the string version, so each table used rotate="true".
Math commands need it as well as text commands. At the same time, this
further unifies the checking for termination and fixes cases of wrong
output (e.g. for 0x2005).
If \hline is entered, do not create an unknown inset, but increase the number
of hlines of the current row if that is allowed. The same idea is applied to
copy-paste (not part of the bug report).
This is also a test for committing via git.
The crash was due to the otherwise innocent looking 07924ac300. The cast
Vincent replaced was already wrong---it seems to have been a thinko in
7bbd67eb2679---but it worked kind of by accident before. Now it crashes.
Setting up .gitignore or .git/info/excludes is something that should
be done. Not doing it makes it a lot harder to see actual new files
that should be added.
Signed-off-by: Lars Gullik Bjønnes <larsbj@gullik.org>
Multirows were introduced in 8bb69f24 (Uwe Stoehr, 11 Feb 2010). In the
computation of the nearest cell, it was forgotten to account for the
vertical offset. tabular.cellHeight is the full height of the cell, while
the point that comes from the coordCache is offsetted by VOffset.
Therefore, we have to subtract the VOffset from o.y_.
We only copied the error list to the cloned buffer when makeLatexfile
failed. This means that a second try will never clear the error list and
the user will be presented the same list again.
Therefore, we unconditionally copy the error list such that we also clear
the list on success.
These warnings have been (accidentally) commented out in
42aeab2a (uwestoehr; support for table cell rotations;
16-03-2012). In the next commit the warning will be replaced
by a form which is also understood by MSVC.
