The old code model expected the macros as C++ formatted text
("#define Foo 42) but newer targets like the Clang codemodel expect key
value arguments like "-DFoo=42". So instead of parsing the text again and
again we use an abstract data description.
Task-number: QTCREATORBUG-17915
Change-Id: I0179fd13c48a581e91ee79bba9d42d501c26f19f
Reviewed-by: Tobias Hunger <tobias.hunger@qt.io>
It is a first step and now a database is generated if you start QtCreator.
Some code is now shared with the PchManager which can be improved in the
future.
Change-Id: Ic267fe7960f6c455d91832859a673ce98f269aa2
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
...to free some memory.
The translation units of the 7 most recently used documents ("hot
documents", tracked by document visibility) are kept in memory.
Translation units of other documents are suspended and will be resumed
once they become visible again.
The resumption of a translation unit needs the same time as reparse
(since it is a reparse effectively).
The number of hot documents can be modified by the run time environment
variable QTC_CLANG_HOT_DOCUMENTS=N. Visible documents are always hot.
Task-number: QTCREATORBUG-11640
Change-Id: I68ecd2b1373e303372300203e42d90f65a4b39b3
Reviewed-by: Ivan Donchevskii <ivan.donchevskii@qt.io>
Reviewed-by: Marco Bubke <marco.bubke@qt.io>
We now support highlighting for an example text and for the query.
Change-Id: I88c415ff871cf3e4c2d4fc83d60a8555bf0ce08a
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
Because we can visit headers many times, we get results many times too.
Change-Id: I3bbe7d7a5d01c2580a4569bfe115f14a69edc8a7
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
Clang file ids are only unique for one query. Because we query in parallel
we have to manage our own unique ids.
Change-Id: I67d57d8b1766cab75ad252a14e57bbf9dc5fdb79
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
Every AST unit is created and queried asynchronously, like before, but
we don't wait anymore but use a timer to process new sources. So the server
will not be blocked and can process other messages like cancel.
Change-Id: If0e69466c78f628190f59fd32a03cab1c3a4d0a3
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
- Avoids the hassle of QRC files and manually registering mime types
- Avoids performance regressions because of mime types that are
registered after mime database has been used
- Makes it technically possible to detect that a disabled plugin could
handle a mime type if it was enabled
Change-Id: I373008b1b56e9c6b4853055f20b3eeb112a6eff9
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
Reviewed-by: hjk <hjk@qt.io>
${AnyProject}::updateCppCodeModel() did two potentially not that cheap
operations in the ui thread:
(1) Querying the MimeDatabase for the mime type for the source files of
the project. In 99.9% of the cases no files need to be read for
this as the file extension will resolve the type. The expensiveness
comes from the sheer number of files that can occur.
(2) Calling compilers with the "(sub)project's compiler command line"
to determine the macros. While the caches avoid redundant calls,
the number of the unique compiler calls makes this still a
ui-freezing experience.
These two operations are moved into a worker thread. For this, the
expensive compiler calls are encapsulated in thread safe lambdas
("runners") in order to keep the "mutexed" data minimal. The original
API calls of the toolchains are implemented in terms of the runners.
While adapting the project managers, remove also the calls to
setProjectLanguage(). These are redundant because all of the project
managers already set a proper value in the constructor. Also, currently
there is no need (client) to report back detection of C sources in
project parts. This also keeps CppProjectUpdater simple.
There is still room for improvement:
* Run the compiler calls in parallel instead of sequence.
* Ensure that the mime type for a file is determined exactly once.
Change-Id: I2efc4e132ee88e3c8f264012ec8fafe3d86c404f
Reviewed-by: Tobias Hunger <tobias.hunger@qt.io>
Overwriting QMAKE_CXXFLAGS will remove the needed isysroot flag
for finding the standard headers.
Explicitly settings the standard lib on macOS is also no longer
necessary.
Change-Id: I8a5ae5df485785b473a435c7feb9f8fb3295952b
Reviewed-by: Christian Stenger <christian.stenger@qt.io>
So far we only compiled the precompiled headers only sequentially. The
process generator is creating, managing a queue of processes to compile
the PCHs parallel.
Change-Id: I8075def4ef9e6b0191bbd51b3631d1c51ec7b361
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
QFileWatcher is only reporting one path per signal which is suboptimal if
you change many files at once. This patch is introducing a timer which is
waiting some time time to see if more path changes are reported and is
collecting them in a vector.
Change-Id: I50f7c21186353b199634e7b3cd5a41f8d581a31d
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
Compiling every header file again and again is quite time comsuming. There
are technics to improve this like preambles(a kind of automated
precompiled header) but they don't share their data between translation
units. This approach provides an automatically generated precompiled
header for every project and subproject to improve the loading time.
Change-Id: I34f5bd4db21951175920e2a9bbf6b97b1d705969
Reviewed-by: Eike Ziller <eike.ziller@qt.io>
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
Resulted in wrong import/export defines of Utils symbols. Also it needs
to be set before the project includes are be done.
Change-Id: I950155a92c9a8ea21d5d91a57bd42058980d3971
Reviewed-by: Marco Bubke <marco.bubke@qt.io>
Slow and very slow tests have now their own test category. We add SlowTest
for tests which are slower than ~5ms and VerySlowTest if they are slower
than ~100ms. They are disabled them by "-*SlowTest.*". If you have a faster
machine than most developers simply try lower values. The aim is that most
developers can execute the tests in under ~2s.
In the long run we should use dependency breaking and data sharing to
reduce the count of the slow tests.
Change-Id: I8578071258d7f89b2052709f3dd526ced811483f
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@qt.io>
So we can easily verify if other changes break anything.
The disabled tests should be all related to the by default enabled delayed
template instantiation on Windows, which is necessary to parse windows headers.
The disabled tests can be run with --gtest_also_run_disabled_tests or by
setting the env variable GTEST_ALSO_RUN_DISABLED_TESTS=1.
Change-Id: I040d881ff0a3fbf5fe85c0b3d1b8e7b0959d1b9c
Reviewed-by: Marco Bubke <marco.bubke@qt.io>
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
...and add some basic tests.
Introduce the abstractions ProjectInterface and ToolChainInterface in
order to break the dependency to the ProjectExplorer. Also, some simple
logic can go there to simplify the (Base)ProjectPartBuilder.
Change-Id: I6c50a1804ce62098b87109931eb171f5c2542937
Reviewed-by: David Schulz <david.schulz@qt.io>
This applies for all project managers, except qmake. The qmake project
manager will make use of this in follow up changes.
Before, "foo.h" was always recognized as a CXXHeader. Now, it depends on
the other files. E.g. in a file list {"foo.h", "foo.c"} foo.h is now a
CHeader. In {"foo.h", "foo.c", "bar.cpp"} the file "foo.h" is ambiguous
and we will create two project parts, one where it is a CHeader, the
other where it is a CXXHeader.
Change-Id: I50505163368742584b1380c284d42cbe07cb4fc9
Reviewed-by: David Schulz <david.schulz@qt.io>
Offset are better to compute the length of a text range. The only problem
is that we get them for utf8 because that is the text representation of
source file. QTextDocument is Utf16 based, so it can not binary represent
the source file. Actually I do not see a simple performing workaround for
it.
Change-Id: Id615e1ee6a6e85c6ecc4f8044e275142409d9b46
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
Clang query is mechanism to use AST matcher to search for code. Think
about regular expression but in the context of AST. So you get a semantic
search tool for C++.
Change-Id: I72e882c5b53a0c52f352a3664847c4c3e4f6fc2e
Reviewed-by: Tim Jenssen <tim.jenssen@qt.io>
This adds a mix-in for writable functions arguments.
Change-Id: I758f7fef77d992ea25395db550571ccb081fd5fd
Reviewed-by: Nikolai Kosjar <nikolai.kosjar@qt.io>
...because those errors can lead to a substantial performance/functional
regression.
The actual diagnostics (possibly with children) are shown as details in
the info bar.
The info bar can be hidden with the "Do Not Show Again" button.
Re-enabling the info bar is possible with the new editor tool bar
button.
Change-Id: I03394ff8e3c84127946b0b791930b28a385f5a46
Reviewed-by: David Schulz <david.schulz@qt.io>
Parsing happens rotationally on the translation units.
The recently parsed translation unit is used for completion jobs while
the older version is used for parse jobs.
Advantages:
A1. A completion job cannot be blocked anymore by currently running
parse job.
A2. Faster triggering of parse jobs. A reparse was triggered about
1650ms after the last keystroke. This is down to 500ms now since we
do not have a blocking translation unit for the completion anymore.
Disadvantages:
D1. Memory consumption is doubled for an edited document.
This could be addressed by suspending the second translation unit
after some time of inactivity.
D2. Setup of the supportive translation unit takes some time.
Change-Id: I958c883c01f274530f5482c788c15cd38d6f4c3e
Reviewed-by: David Schulz <david.schulz@qt.io>
A TranslationUnit is owned by TranslationUnits now. TranslationUnits
allows to add another TranslationUnit and to update/query the recently
and previously parsed translation unit.
This does not change any behavior yet.
Change-Id: I8a2f0cc05d3e51bf739dd5d7c4da14b54147f3ab
Reviewed-by: David Schulz <david.schulz@qt.io>
...in preparation for follow-up changes. This will enable e.g. a timer
per document.
This does not change any behavior yet.
Change-Id: Ic1dc06de602373c666d47ce7a95ab99e56d389d5
Reviewed-by: David Schulz <david.schulz@qt.io>
The "TranslationUnit parts" moved already to TranslationUnitCore.
Change-Id: I2bea7847e2b3e84fbfacc3d2dc43f180873349ac
Reviewed-by: David Schulz <david.schulz@qt.io>
If you have no LLVM installed it was impossible to compile the unit test.
But some unit test don't depend on LLVM. With this change it is now
possible to compile them.
Change-Id: Iac0c1b3cdf6c317e6ba4755acd5f8458db5a7451
Reviewed-by: David Schulz <david.schulz@qt.io>
It was bumped to 14 in merge commit 1853f01a, due to a conflict in
unittest.pro which I confused with qtcreator.pri.
Change-Id: I1ec15d9d20d90c38849eea799a724f4c85fe36b5
Reviewed-by: Eike Ziller <eike.ziller@qt.io>
