| Age | Commit message (Collapse) | Author |
|---|
|
Summary:
The current KConfigXT compiler is in a sad state:
It's a massive file with loads of global variables that handle state, the generator is done within the main() function and it seems to have grown organically. There are no classes to separate logic / state / generation, what exists is code that generates code from a xml / ini pair, but it's hard to even discover what a bit of code is doing. The code istyle is C++ / Java from the nineties, which is not bad per see but it also uses quite a few things that are going to be deprecated in Qt 6 so I'm also taking the time make the code more streamlined with newer code style (no iterators, lambdas, auto usage, etc).
The code that generates the files simplly pushes strings to a text stream, and it's hard to figure out when something starts or something ends: for instance, the code that generates the Constructor has more than sixty lines of code englobing some nested if - for - if - for constructs.
Currently the code is "done" - there's one bug that I still need to find & fix regarding Translations, but the rest seems sane.
The current testcode generates incorrect *whitespaces* regarding the old code (there's some parts that I feel that it's important to fix before merging, but overall, the whitespace changes are not bad and easier to handle, old code had a hand-counted amount of spaces before each line, new code has a function whitespace() that adds the current-and-correct amount of whitespaces based on indentation level that you start by startScope() and ends with endScope(). rest of the code still needs to be ported to it.
I plan to fix the testcases whitespace by manually adding them, I'v fougth with the code for a while and added a few hacks there but I don't want to make the code hackish again.
New code is not perfect by any means, but is a good step in the right direction.
This code tries to Separate the compiler code into many different files / classes to be more obvious what's happening, and each class also has many helper methods to minimize copypaste.
- CodeGenerator: Has base code for the header and source files that can be shared
- HeaderGenerator: Logic for generating the header file
- SourceGenerator: Logic for generating the source file
- KcfgParser: Logic for parsing the kcfg file and extracting the information from the Xml file
- CommonStructs: a header that contains the structs that are currently used everywhere.
- KConfigParameters: (was CfgConfig - ConfigConfig, wat) - Has information passed via the kcfgc file
- kcfg_compiler - will be renamed to main - start the other classes and generates the files.
This code here currently has the begining of this separation, with the CodeGenerator and the HeaderGenerator in a ~good~ state, but unfinished.
Test Plan:
- Run the test cases,
- Compare the diffs generated by the testcases and fix in the code the errors / differences
- Run and compare real kde source with the new and old generators to look for errors
Reviewers: #frameworks, ervin, bport, dfaure
Reviewed By: dfaure
Subscribers: davidre, bcooksley, cgiboudeaux, kossebau, bport, ngraham, kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D26202
|
|
This reverts commit 98c32e29f50465d4d4e16bafdf0491edbfb422b0.
This broke kdevelop and kmymoney. I'm workin on the fixes and more
unittests for them.
|
|
Summary:
The current KConfigXT compiler is in a sad state:
It's a massive file with loads of global variables that handle state, the generator is done within the main() function and it seems to have grown organically. There are no classes to separate logic / state / generation, what exists is code that generates code from a xml / ini pair, but it's hard to even discover what a bit of code is doing. The code istyle is C++ / Java from the nineties, which is not bad per see but it also uses quite a few things that are going to be deprecated in Qt 6 so I'm also taking the time make the code more streamlined with newer code style (no iterators, lambdas, auto usage, etc).
The code that generates the files simplly pushes strings to a text stream, and it's hard to figure out when something starts or something ends: for instance, the code that generates the Constructor has more than sixty lines of code englobing some nested if - for - if - for constructs.
Currently the code is "done" - there's one bug that I still need to find & fix regarding Translations, but the rest seems sane.
The current testcode generates incorrect *whitespaces* regarding the old code (there's some parts that I feel that it's important to fix before merging, but overall, the whitespace changes are not bad and easier to handle, old code had a hand-counted amount of spaces before each line, new code has a function whitespace() that adds the current-and-correct amount of whitespaces based on indentation level that you start by startScope() and ends with endScope(). rest of the code still needs to be ported to it.
I plan to fix the testcases whitespace by manually adding them, I'v fougth with the code for a while and added a few hacks there but I don't want to make the code hackish again.
New code is not perfect by any means, but is a good step in the right direction.
This code tries to Separate the compiler code into many different files / classes to be more obvious what's happening, and each class also has many helper methods to minimize copypaste.
- CodeGenerator: Has base code for the header and source files that can be shared
- HeaderGenerator: Logic for generating the header file
- SourceGenerator: Logic for generating the source file
- KcfgParser: Logic for parsing the kcfg file and extracting the information from the Xml file
- CommonStructs: a header that contains the structs that are currently used everywhere.
- KConfigParameters: (was CfgConfig - ConfigConfig, wat) - Has information passed via the kcfgc file
- kcfg_compiler - will be renamed to main - start the other classes and generates the files.
This code here currently has the begining of this separation, with the CodeGenerator and the HeaderGenerator in a ~good~ state, but unfinished.
Test Plan:
- Run the test cases,
- Compare the diffs generated by the testcases and fix in the code the errors / differences
- Run and compare real kde source with the new and old generators to look for errors
Reviewers: #frameworks, ervin, bport, dfaure
Reviewed By: dfaure
Subscribers: bport, ngraham, kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D26202
|
|
Summary:
The generated classes of kconfig_compiler take a KSharedConfig::Ptr by
value, one possibility would be to make them const & but that is BIC so
instead what we do is just move them to the only place the config is
used so the cheaper move constructor can be used instead of the copy
constructor
Reviewers: ervin, apol
Reviewed By: ervin, apol
Subscribers: kde-frameworks-devel, ervin, apol
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D25061
|
|
Reviewers: #plasma, #frameworks, dfaure, mart, apol
Reviewed By: apol
Subscribers: kossebau, apol, kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D24490
|
|
Summary: We were not adjusting the property name to the getter letter-casing.
Test Plan: See added test, also fixes the issue that made me realize this issue.
Reviewers: #frameworks, davidedmundson
Reviewed By: davidedmundson
Subscribers: ngraham, aacid, kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D24010
|
|
|
|
Lets you specify Notifiers= in .kcfg for config entries that should be written with Notify flag,
i.e. announce the change to KConfigWatcher
Differential Revision: https://phabricator.kde.org/D20196
|
|
Summary:
When using kconfig_compiler generated sources in projects where the used
file extensions for C++ files are not ".h" and/or ".cpp", the include for
the generated header looks a bit alien to the project, as well as the
generated source file if one looks closer at it.
This code adds new optional flags HeaderExtension & SourceExtension which
can be used to control the file extensions used for the generated files.
Test Plan:
All unit tests and the new test_fileextensions pass, existing projects using
kcfg without & with the new flags build fine.
Reviewers: #frameworks, apol
Reviewed By: apol
Subscribers: apol, kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D19565
|
|
Summary:
of the generated class that has a pointer and raw copy would be bad.
Those generated classes are internal and nobody would probably have this, but being safe never hurts
Reviewers: vkrause
Reviewed By: vkrause
Subscribers: kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D18136
|
|
Summary:
Since "Use override" (3985f2a0) kconfig_compiler.cpp
generates usrSave with override instead of Q_DECL_OVERRIDE. This causes
the KConfigCompiler_Test to fail [1], as the reference file of the test
was not updated.
[1]: https://build.kde.org/job/Frameworks%20kconfig%20kf5-qt5%20SUSEQt5.10/32/
Reviewers: #frameworks, mlaurent
Reviewed By: mlaurent
Subscribers: kde-frameworks-devel
Tags: #frameworks
Differential Revision: https://phabricator.kde.org/D12892
|
|
|
|
|
|
|
|
This way, different newline endings in OSes are handled as \n when
reading, and thus fixing the comparison on non-\n OSes.
|
|
Use the QProcess::start() variant with explicit (empty, in this case)
arguments, so the program string is not parsed as shell command, thus
preserving paths with spaces as such.
|
|
|
|
|
|
Summary:
In case a kcfg with arg="true" was used and singleton the static
instance method only accepted a QString config name. This made it
impossible to combine a singleton config with an already existing and
open KSharedConfig::Ptr.
With this change an overloaded instance method is added which takes a
KSharedConfig::Ptr as argument. The private ctor, though, only takes a
KSharedConfig::Ptr and the instance method taking a QString argument
uses KSharedConfig::openConfig on the config file name.
The change is source-incompatible in the following situation:
* kcfgfile arg="true"
* Singleton = true
* Inherits is specified
In this situation the previous revision created an instance method with
a QString argument and passed that to the parent constructor. This is
not in accordance with the documentation. Any user of this behavior was
relying on a bug. With this change now the call to the parent
constructor carries a KSharedConfigPtr.
Test Plan:
kconfigcompiler tests still pass and a config with singleton
and arg="true" generates the code as I need it
Reviewers: #frameworks, dfaure, mdawson
Differential Revision: https://phabricator.kde.org/D3690
|
|
Reviewers: #frameworks, davidedmundson
Reviewed By: davidedmundson
Differential Revision: https://phabricator.kde.org/D3702
|
|
This reverts commit cd4e6504dfbdface00037625f0cedda511e6d839.
As suggested by Martin on release-team@kde.org, given that it breaks SC.
|
|
This reverts commit 71f16741a0288d8587876dcc7dbb33ba8f00546a.
|
|
Sorry.
|
|
Summary:
In case a kcfg with arg="true" was used and singleton the static
instance method only accepted a QString config name. This made it
impossible to combine a singleton config with an already existing and
open KSharedConfig::Ptr.
With this change an overloaded instance method is added which takes a
KSharedConfig::Ptr as argument. The private ctor, though, only takes a
KSharedConfig::Ptr and the instance method taking a QString argument
uses KSharedConfig::openConfig on the config file name.
This provides full API compatibility and at the same time allows to use
KSharedConfig in addition to the arg name based variant.
Test Plan:
kconfigcompiler tests still pass and a config with singleton
and arg="true" generates the code as I need it
Reviewers: #frameworks
Differential Revision: https://phabricator.kde.org/D3386
|
|
REVIEW: 129382
|
|
Summary:
Instead of having an odd fake of it. Will help some required
refactorings and already showed some issues, fixed by this patch,
namely:
- don't use string(regex replace) to extract a string from another
string. in case it doesn't match it will offer the whole content which
is never what we want.
- messages(ERROR), the correct parameter is FATAL_ERROR, cmake
understands "ERROR" as mere output string
- turn the macro into a function, otherwise 2 calls in the same
subdirectory are dangerous.
CCBUG: 371562
Test Plan: tests still pass, projects that use the macro still build
Reviewers: #frameworks, dfaure
Reviewed By: dfaure
Subscribers: dfaure
Differential Revision: https://phabricator.kde.org/D3178
|
|
|
|
|
|
REVIEW: 125833
|
|
|
|
::usrWriteConfig is deprecated, use ::usrSave as recommended by the
documentation.
REVIEW: 124467
|
|
In applications translations can be looked up in the globally set
translation domain, but in libraries it is necessary to link every
i18n call to the library's own translation domain. A new code
generation option TranslationDomain= is added to enable this.
It has effect only if TranslationSystem=kde is set.
Added unit tests to check generated translation calls.
CHANGELOG: New code generation option TranslationDomain=, for use with TranslationSystem=kde; normally needed in libraries.
REVIEW: 123872
|
|
|
|
|
|
It adds a configuration setting that makes it possible to generate
Q_PROPERTY instances out of each variable exposed by the configuration
class.
Especially useful when it comes to exposing these classes to QtQuick
interfaces.
REVIEW: 123367
CHANGELOG: Generate QML-proof classes using the kconfigcompiler
|
|
Add new variable to specify it in *.kcfgc : "CategoryLoggingName"
CHANGELOG: Allow to generate file with qloggingcategories support.
REVIEW: 122931
|
|
gcc 4.5 doesn't have nullptr.
|
|
- use Q_NULLPTR instead of 0 or NULL
- simplify some foreach loops
- use QStringLiteral in a few places
- added a few consts
REVIEW: 118666
|
|
This reverts commit 88c9a9664e99b639df4cde2be22dbe44df228879.
|
|
|
|
The MOC files for generated output from kconfig_compiler are not automatically
generated. Re-add the code that does this, fixing the build.
I'm assuming this isn't an issue on Windows, as I don't have any machines to
test on. Please let me know if this isn't the case.
CC: arichardson.kde@gmail.com
|
|
We have to generate the files in the same directory as the test executables
otherwise QFINDTESTDATA will fail
|
|
This also fixes running 'nmake test' on Windows.
|
|
The extra load of values in KCoreConfigSkeleton is not documented anywhere
that it happens, and in normal circumstances re-loading new values isn't
expecterd during a save operation.
Update various mentions of readConfig/writeConfig to read/save.
Update documentation to match the new names of functions.
Rename writeConfig() to save() and usrWriteConfig() to usrSave()
Rename functions to match the new naming conventions. Also create old deprecated
instances of both functions to ensure that old code continues to operate
correctly. Also make save() non-virtual, for the same reasons read() is
now non-virtual.
Rename usrReadConfig to usrRead, to be consistent with the new name for readConfig.
Rename usrReadConfig to usrRead, and mark the former as deprecated. To maintain
compatibility, usrRead still calls usrReadConfig in its default implementation.
usrReadConfig remains empty.
REVIEW: 117010
|
|
Just because the executable has the "_kf5" suffix, that does not mean
the target should have it. This is ugly API, and will be unnecessary
porting effort for KF6.
REVIEW: 116995
|
|
Call it from generated singletons, since the constructor creates
a KConfig from a filename, which already loads from disk.
This removes the need for using DelayedParsing.
REVIEW: 116845
|
|
Previously the classes generated by kconfig_compiler would only emit
the defined signals when using the setters provided by that class.
However, when using e.g. KConfigDialog which uses
KConfigSkeletonItem::setProperty() to change the items no signal was
generated.
This patch fixes this by using a wrapper KConfigSkeletonItem
subclass that calls a private itemChanged() method in the generated
class which updates the set of changed properties. As soon as the item
is saved (usrWriteConfig() in the generated class is called) the signal
will be emitted
REVIEW: 115635
REVIEW: 115634
|
|
Previously the output of diff -u was written to a file when the
generated file did not match the expectations. Having the output printed
to stdout makes it easier to see the exact error without having to know
that a diff exists in a certain file somewhere in the build directory.
REVIEW: 115832
|
|
|
|
|
