A manual squash/merge of the changes below, plus a couple of subsequent
code fixes.
59085aa5fbb99e2d786cd2c1a06c24a111ccb49f:
Modify CppModel::ProjectInfo
Adding per project node information, to pass on the correct
defines/includes for each file, instead of aggregating them incorrectly.
Also split up SOURCES and OBJECTIVE_SOURCES.
Also ask the toolchain to convert the compilerflags to flags the
codemodel understands, for now only gcc and only c++11.
Also make the toolchain aware of the flags used to compile, so that it
can emit the correct defines.
Note: No header files are passed on.
74028802314cd4e75b41b46407433e07090a304d:
GCC: Evaluate cxxflags when checking for predefined macros
ebaaa4957e4c02cc9637a998eddae1d0acd74f83:
MSVC: Take cxxflags into account when checking for predefined macros
9bfce7e889bcf7bcc47bf880e3ea25945ca7d0d7:
Compile fixes
Change-Id: I9de94ad038dfc5dc1987732e84b13fb4419c96f5
Reviewed-by: Erik Verbruggen <erik.verbruggen@nokia.com>
Getting the #include directives ready for Qt5. This includes the
new-project wizards.
Change-Id: Ia9261f1e8faec06b9285b694d2b7e9a095978d2b
Reviewed-by: Eike Ziller <eike.ziller@nokia.com>
The order of actions and thus the state at which the signals are emitted
has changed. The aboutToUnloadSession comes before the actual saving
now. Also the aboutToLoadSession is after restoring values but before
restoring projects and editors.
Change-Id: I3eae66cffae970f00535d8d6e2bb11a6ca645456
Reviewed-by: hjk <qthjk@ovi.com>
Reviewed-by: Tobias Hunger <tobias.hunger@nokia.com>
It causes a conflict with the new QtConcurrent library
of Qt 5 and is quite small.
Change-Id: Ib97ab6daf8f67a88300cf03f044ba3593a4cd270
Reviewed-by: Eike Ziller <eike.ziller@nokia.com>
This way the editor does not need to know all the details of
instantiating or maintaining classes for highlighting and/or completion,
it can just ask the model manager. The change also enables different
highlighting- or completion-engines without changes to the cppeditor.
Change-Id: I8000d9d9fe446b292defddb2295493cf77d0f14a
Reviewed-by: Leandro Melo <leandro.melo@nokia.com>
Continuinng a previous commit... This is in order to make
the interface clearer.
Change-Id: Ic1b05217f878578bc84a064927507b75981c47d6
Reviewed-by: Roberto Raggi <roberto.raggi@nokia.com>
This is not really a direct mapping to the Microsoft specifc
ones, but I guess defining such macros is an acceptable solution.
Change-Id: I17a9414364ef664f2e8afa7ececc6a87f663049e
Reviewed-on: http://codereview.qt.nokia.com/4277
Reviewed-by: Qt Sanity Bot <qt_sanity_bot@ovi.com>
Reviewed-by: Erik Verbruggen <erik.verbruggen@nokia.com>
There is logic in the preprocessor that assumes a normalized line
ending. Other patches have already fixed the issue in other parts.
Change-Id: Iac262da361b6528cc9466a6c87e83da95af9ea04
Reviewed-on: http://codereview.qt.nokia.com/4252
Reviewed-by: Qt Sanity Bot <qt_sanity_bot@ovi.com>
Reviewed-by: Christian Kamm <christian.d.kamm@nokia.com>
It now lives in qmljstools/qmljsfindexportedcpptypes, all in one place.
Also ensures that the source code is available when a file is being
scanned for QML exports. This will enable checking comments for
annotations about the URI a plugin is usually imported as.
Change-Id: I1da36d0678e0a8d34b171dbe0f6b5690d89eb18b
Reviewed-on: http://codereview.qt.nokia.com/3392
Reviewed-by: Fawzi Mohamed <fawzi.mohamed@nokia.com>
Having a duplicate prototype chain - once in FakeMetaObjects and once
in QmlObjectValues was unnecessary. Now FMOs don't contain references
which may allow other simplifications.
Previously they were leaked when a qmldump or the C++ exported QML
type list updated.
Just deleting the previous FakeMetaObjects is not an option, as they
might still be used in a QmlObjectValue owned by an Engine.
Reviewed-by: Erik Verbruggen
The scanning phase was in some cases insanely long.
Reason was that Creator feeds wrong paths to the
CppModelManager, for example "/".
The scanning phase did what it was told to: scan that whole
Harddrive.
This patch makes sure that the scanning is not done for "/"
Done-With: Roberto
(The other Samba is meant. The slow one, not the nice one.)
Creator manages to feed the CppModelManager with all kinds of
wrong include paths. That is usually not a big issue as long
as that path is not a simple "/".
CppManager appended a separator to all incoming include paths.
And when the above "/" was passed in, we had (on my machine)
7 seconds of Samba Network lookup, multiple times during a
parse. And the sad thing was that my dear colleagues in the
Intranet did not even have the searched-for header files on
their shared folders.
This patch makes sure that a separator is only appended if
the path does not already end with one.
The real core of the issue (wrong include paths from Creator)
still needs to be fixed.
Done-with: Roberto Raggi
