Revision control
Copy as Markdown
// Copyright (C) 2017 Ford Motor Company
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR GFDL-1.3-no-invariants-only
/*!
\title Domain Specific Language (DSL)
*/
/*!
\page qtremoteobjects-repc.html
\title Qt Remote Objects Compiler
\brief The Qt Remote Objects Compiler creates \l {Source} and \l {Replica} header files
\ingroup overviews
\keyword repc
\section1 REPC Overview
The \underline {Rep}lica \underline {C}ompiler (repc) generates QObject
header files based on an API definition file. The file (called a "rep"
file) uses a specific (text) syntax to describe the API. By convention,
these files are given a .rep file extension, short for Replica. When these
files are processed by repc, repc generates both \l {Source} and \l
{Replica} header files.
The Qt Remote Objects module also includes \l {CMake functions} and
\l {qmake variables} that can be added to your project file to automatically run
repc, and add the resulting files to the list of files processed by
\l{moc}{Meta Object Compiler} during the build process, making use of Qt
Remote Objects in your projects simple.
While Qt Remote Objects supports sharing any QObject over the network (using
enableRemoting on the Source side and acquireDynamic on the Replica side),
there are a couple of advantages to letting repc define your objects. First
of all, while \l {QRemoteObjectDynamicReplica} {DynamicReplicas} are
useful, they are more cumbersome to work with. The API is not known until
the object is initialized, and using the API from C++ requires string
lookups through QMetaObject's methods. Secondly, having the interface known
at compile time finds any issues at compile vs. at runtime. Thirdly, the rep
format supports default values, which can be handy if you are unable to
ensure the Source is available when the Replica is instantiated.
See the documentation \l {Source} {here} for information on using
the generated files in your code. Here we will focus on the repc format and
options.
\section1 The rep file format
The rep file format is a simple \l {Domain Specific Language (DSL)} for
describing an interface supported over Qt Remote Objects (QtRO). Since QtRO
is an object based system, these interfaces are defined by APIs available
through objects, that is, classes with properties, signals, and slots.
\section2 Comments
Single-line comments in C++ style (\c{/}\c{/ text}) as well as multi-line comments
in C style (\c{/}\c{* text *}\c{/}) are supported.
\section2 The class type
Each class defined in a rep file becomes a QObject in the generated header
files, with the described API generated for you.
To define a class use the \c class keyword, followed by the name you
want for your type, and then enclose your API in brackets like so
\code
class MyType
{
//PROP/CLASS/MODEL/SIGNAL/SLOT/ENUM declarations to define your API
};
\endcode
When using the generated header files inside a library, it might be needed
to define class attributes to set the symbol visibility. This can be done
similar to C++ by defining the attribute after the \c class keyword.
In the following example the \c MYSHAREDLIB_EXPORT macro is used, which
is defined in \c "mysharedlib_global.h". See \l{Creating Shared Libraries}
for more information how this works.
\code
#include "mysharedlib_global.h"
class MYSHAREDLIB_EXPORT MyType
{
...
};
\endcode
\section3 PROP
Q_PROPERTY elements are created by using the PROP keyword in the rep
file. The syntax is the \c PROP keyword followed by the definition enclosed
in parentheses, where the definition is the type, the name, and (optionally) a
default value or attributes.
\code
PROP(bool simpleBool) // boolean named simpleBool
PROP(bool defaultFalseBool=false) // boolean named defaultFalseBool, with false
// as the default value
PROP(int lifeUniverseEverything=42) // int value that defaults to 42
PROP(QByteArray myBinaryInfo) // Qt types are fine, may need #include
// additional headers in your rep file
PROP(QString name CONSTANT) // Property with the CONSTANT attribute
PROP(QString setable READWRITE) // Property with the READWRITE attribute
// note: Properties default to READPUSH
// (see description below)
PROP(SomeOtherType myCustomType) // Custom types work. Needs #include for the
// appropriate header for your type, make
// sure your type is known to the metabject
// system, and make sure it supports Queued
// Connections (see Q_DECLARE_METATYPE and
// qRegisterMetaType)
\endcode
More information about creating custom types can be found \l {Creating
Custom Qt Types} {here}.
By default, properties will have getters and a "push" slot defined, as well
as a notify signal emitted when the value is changed. Qt Remote Objects
requires the notify signal on the Source object to trigger sending updates
to the attached Replicas. In earlier versions of QtRO, properties defaulted
to being read/write, that is, having getters and setters. However, due to the
asynchronous nature of QtRO, this led to unintuitive behavior at times.
Setting the READWRITE attribute on the PROP will provide the old (getter
and setter) behavior.
\code
// In .rep file, old (setter) behavior
PROP(int myVal READWRITE) // Old behavior with setMyVal(int myVal) method
// In code... Assume myVal is initially set to 0 in Source
int originalValue = rep->myVal(); // Will be 0
rep->setMyVal(10); // Call setter, expecting a blocking/
// non-asynchronous return
if (rep->myVal() == 10) ... // Test will usually fail
\endcode
If it is necessary to block until the value is changed, something like the
following is necessary.
\code
// In .rep file, old (setter) behavior
PROP(int myVal READWRITE) // Old behavior with setMyVal(int myVal) method
// In code... Assume myVal is initially set to 0 in Source
bool originalValue = rep->myVal(); // Will be 0
// We can wait for the change using \l QSignalSpy
QSignalSpy spy(rep, SIGNAL(myValChanged(int)));
rep->setMyVal(10); // Call setter, expecting a blocking/
// non-asynchronous return
spy.wait(); // spy.wait() blocks until changed signal
// is received
if (rep->myVal() == 10) ... // Test will succeed assuming
// 1. Source object is connected
// 2. Nobody else (Source or other Replica)
// sets the myVal to something else (race
// condition)
// Rather than use QSignalSpy, the event-driven practice would be to connect the
// myValChanged notify signal to a slot that responds to the changes.
\endcode
QtRO now defaults to READPUSH, which provides an automatically generated
slot for requesting a property change.
\code
// In .rep file, defaults to READPUSH
PROP(bool myVal) // No setMyVal(int myVal) on Replica, has
// pushMyVal(int myVal) instead
// In code... Assume myVal is initially set to 0 in Source
bool originalValue = rep->myVal(); // Will be 0
// We can wait for the change using \l QSignalSpy
QSignalSpy spy(rep, SIGNAL(myValChanged(int)));
rep->pushMyVal(10); // Call push method, no expectation that change
// is applied upon method completion.
// Some way of waiting for change to be received by the Replica is still necessary,
// but hopefully not a surprise with the new pushMyVal() Slot.
spy.wait(); // spy.wait() blocks until changed signal
// is received
if (rep->myVal() == 10) ... // Test will succeed assuming
// 1. Source object is connected
// 2. Nobody else (Source or other Replica)
// set the myVal to something else (race
// condition)
\endcode
You can also use the \c CONSTANT, \c READONLY, \c PERSISTED, \c READWRITE,
\c READPUSH, or \c SOURCEONLYSETTER keywords in the PROP declaration, which
affects how the property is implemented. READPUSH is the default value if no value
used.
\code
PROP(int lifeUniverseEverything=42 CONSTANT)
PROP(QString name READONLY)
\endcode
Please note there are some subtleties here. A CONSTANT PROP has a
Q_PROPERTY declared as CONSTANT on the SOURCE side. However, replicas
cannot know the correct value until they are initialized, which means the
property value has to be allowed to change during initialization. For
READONLY, the source will have neither a setter nor a push slot, and the
replica side will not have a push slot generated. Adding the PERSISTED
trait to a PROP will have the PROP use the \l QRemoteObjectAbstractPersistedStore
instance set on a Node (if any) to save/restore PROP values.
Another nuanced value is SOURCEONLYSETTER, which provides another way of
specifying asymmetric behavior, where the \l Source (specifically the helper
class, \c SimpleSource) will have a public getter and setter for the
property, but it will be ReadOnly (with a notify signal) on the \l Replica
side. Thus the property can be fully controlled from the \l Source side,
but only observed from the \l Replica side. SOURCEONLYSETTER is the mode
used by repc for MODEL and CLASS instances, meaning the \l Source can
change the pointed to object, but the \l Replica can not provide a new
object, as no set<Prop> or push<Prop> method is generated. Note, this does
not impact the behavior of the pointed to type's properties, just the
ability to change the pointer itself.
\sa QRemoteObjectAbstractPersistedStore
\section3 CLASS
The CLASS keyword generates special Q_PROPERTY elements for objects derived from QObject.
These properties have the same semantics as SOURCEONLYSETTER. The syntax is the \c CLASS
keyword followed by the property name and then the type of subobject enclosed in parentheses.
\code
// In .rep file
class OtherClass
{
PROP(int value)
}
class MainClass
{
CLASS subObject(OtherClass)
}
\endcode
\section3 MODEL
The MODEL keyword generates special Q_PROPERTY elements for objects derived from QAbstractItemModel.
These properties have the same semantics as SOURCEONLYSETTER. The syntax is the \c MODEL
keyword followed by the property name and then parentheses enclosing the (comma-separated) roles
that should be exposed to the replica.
\code
// In .rep file
class CdClass
{
PROP(QString title READONLY)
MODEL tracks(title, artist, length)
}
\endcode
\section3 SIGNAL
\l [DOC QtCore] {Signals} {Signal} methods are created by using the SIGNAL keyword in the rep file.
Usage is to declare \c SIGNAL followed by the desired signature wrapped in
parentheses. The void return value should be skipped.
\code
SIGNAL(test())
SIGNAL(test(QString foo, int bar))
SIGNAL(test(QMap<QString,int> foo))
SIGNAL(test(const QString &foo))
SIGNAL(test(QString &foo))
\endcode
Just like in Qt \l {Qt::ConnectionType} {queued connections}, parameters in signals that are
references will be copied when being passed to replicas.
\section3 SLOT
\l [DOC QtCore] {Slots} {Slot} methods are created by using the SLOT keyword in the rep file.
Usage is to declare \c SLOT followed by the desired signature wrapped in
parentheses. The return value can be included in the declaration. If the
return value is skipped, void will be used in the generated files.
\code
SLOT(test())
SLOT(void test(QString foo, int bar))
SLOT(test(QMap<QString,int> foo))
SLOT(test(QMap<QString,int> foo, QMap<QString,int> bar))
SLOT(test(QMap<QList<QString>,int> foo))
SLOT(test(const QString &foo))
SLOT(test(QString &foo))
SLOT(test(const QMap<QList<QString>,int> &foo))
SLOT(test(const QString &foo, int bar))
\endcode
Just like in Qt \l {Qt::ConnectionType} {queued connections} and QtRO
SIGNALS, parameters in slots that are references will be copied when being
passed to Replicas.
\section3 ENUM
Enumerations (which use a combination of C++ enum and Qt's Q_ENUM in QtRO)
are described using the ENUM keyword.
\code
ENUM MyEnum {Foo}
ENUM MyEnum {Foo, Bar}
ENUM MyEnum {Foo, Bar = -1}
ENUM MyEnum {Foo=-1, Bar}
ENUM MyEnum {Foo=0xf, Bar}
ENUM MyEnum {Foo=1, Bar=3, Bas=5}
\endcode
Related topics: \l {The ENUM type}, \l {USE_ENUM keyword}
\section2 The POD type
Plain Old Data (POD) is a term to describe a simple data collection, along
the lines of a C++ struct. For example, if you have an API for a phonebook,
you may want to use the concept of an "address" in its interface (where
address might include street, city, state, country, and postal code). You
can use the POD keyword to define objects like this, which can then be used
by in PROP/SIGNAL/SLOT definitions in your class definitions.
Usage is to declare \c POD followed by the name for the generated type,
followed by type and name pairs separated by commas, where the type/name
pairs are wrapped in parentheses.
\code
POD Foo(int bar)
POD Foo(int bar, double bas)
POD Foo(QMap<QString,int> bar)
POD Foo(QList<QString> bar)
POD Foo(QMap<QString,int> bar, QMap<double,int> bas)
\endcode
A full example would look like this
\code
POD Foo(QList<QString> bar)
class MyType
{
SIGNAL(sendCustom(Foo foo));
};
\endcode
The code generated by repc creates a Q_GADGET class for each POD, with
corresponding Q_PROPERTY members for each type defined for the POD.
When using the generated header files inside a library, it might be needed
to define class attributes to set the symbol visibility. This can be done
by defining the attribute after the \c POD keyword.
In the following example the \c MYSHAREDLIB_EXPORT macro is used, which
is defined in \c "mysharedlib_global.h". See \l{Creating Shared Libraries}
for more information how this works.
\code
#include "mysharedlib_global.h"
POD MYSHAREDLIB_EXPORT Foo(int bar)
\endcode
\section2 The ENUM type
It is often easier and cleaner to define an ENUM inside a class (see \l ENUM),
but if you need a standalone enum type, using the ENUM keyword outside of a
class definition can be helpful. This will generate a new class in your
header files that handles marshalling, etc.. The syntax is identical to \l
ENUM, with the exception that the declaration in this case is not contained
in a \c class declaration.
Related topics: \l {ENUM}, \l {USE_ENUM keyword}
\section2 USE_ENUM keyword
The USE_ENUM keyword was implemented before autogeneration via the ENUM
keyword was added. It is kept for backwards compatibility.
Related topics: \l {ENUM}, \l {The ENUM type}
\section2 Directives
The rep file defines an interface, but interfaces often require external
elements. In order to support this, repc will include any (single line)
directives at the top of the generated files. This allows you to, for
instance, use #include or #define directives that support the logic or
datatypes needed.
The repc tool currently ignores everything from the "#" symbol to the
end-of-line and adds that to the generated files. So multi-line
#if/#else/#endif statements and multi-line macros are not supported.
\section3 #HEADER and #FOOTER Directives
There are two special directives, \c #HEADER and \c #FOOTER. Those directives
can be used to define content which should be put as-is into the generated
code, either before (HEADER) or after (FOOTER) the interfaces declaration.
The leading \c #HEADER and \c #FOOTER tokens plus one white-space character are
stripped.
In the following example, the generated repc classes are put inside a namespace.
\code
#HEADER namespace MyNamespace {
class MyType
{
...
};
#FOOTER } // namespace MyNamespace
\endcode
\section1 CMake functions
The CMake functions for generating source and replica types are listed below.
\annotatedlist cmake-macros-qtremoteobjects
\section1 qmake variables
\section2 REPC_REPLICA
Specifies the names of all rep files in the project that should be used to generate
replica header files.
For example:
\code
REPC_REPLICA = media.rep \
location.rep
\endcode
The generated file(s) will be of the form \c {rep_<replica file base>_replica.h}.
\section2 REPC_SOURCE
Specifies the names of all rep files in the project that should be used to generate
source header files.
For example:
\code
REPC_SOURCE = media.rep \
location.rep
\endcode
The generated file(s) will be of the form \c {rep_<replica file base>_source.h}.
\section2 REPC_MERGED
Specifies the names of all rep files in the project that should be used to generate
combined (source and replica) header files.
For example:
\code
REPC_MERGED = media.rep \
location.rep
\endcode
The generated file(s) will be of the form \c {rep_<replica file base>_merged.h}.
\note Typically sources and replicas live in separate processes or devices, so this variable
is not commonly used.
\section2 QOBJECT_REP
Specifies the names of existing QObject header files that should be used to generate corresponding
.rep files.
*/