Skip to content

Latest commit

 

History

History
460 lines (337 loc) · 22.7 KB

README.md

File metadata and controls

460 lines (337 loc) · 22.7 KB

ACPLT/RTE Runtime Environment

This is the source code of a runtime environment for the process control domain. It is developed since around 1999 at the Chair of Process Control Engineering at the RWTH Aachen University. It is well known in the German process control community under the name of the core technology ACPLT/OV.

The runtime environment is used at industry scale in the petro chemical plants at INEOS in Cologne.

Building ACPLT/OV

Prerquisites

To compile the ACPLT/RTE yourself you need to follow the steps presented here. Currently, ACPLT/RTE can be compiled on GNU/Linux and MinGW on Windows. Cross-compiling on Linux for Windows is also possible, using the MinGW compilers or (experimentally) the Microsoft Visual C compiler.

On any platform, you'll need a C compiler, make, cmake, tclsh, bison and flex.

Debian/Untuntu command line:

sudo apt install build-essential tcl bison flex cmake

On Windows, you can set up a Cygwin or MSys environment or download the required tools separately:

Instructions for MSys2 (mingw-w64):

Install the msys2 and update the base installation.

pacman -Syu

Follow the instructions that are provided.

Now install the required packages in the msys2 environment. Here we choose the 64-bit mingw-w64 toolchain.

pacman -S tcl flex bison mingw-w64-x86_64-toolchain

Add the $MSYS2_ROOT$\usr\bin directory to the PATH (either system wide or user). The Toolchain environment is places in MSYS2_ROOT\mingw64.

TODO Setup with Cygwin

Compilation

Clone the acplt/rte project from github. Create a build directory, run cmake configuration and build:

git clone https://github.com/acplt/rte.git
cd rte
mkdir build
cd build
cmake ..
make -j 4

On Windows with MinGW (without Cygwin), use mingw32-make.exe instead of make.

On Windows with Visual Studio / MSVC, you can either:

  • Start Visual Studio, open the project directory as a "local folder" or clone it from a Git repository. Visual Studio will automatically detect the CMakeLists.txt and run CMake to create a Solution file. If not, you can use ProjectGenerate Cache for {project_name} to run CMake. When finished, you should be able to start compilation via BuildInstall {project_name}. ("Install" means packaging to .\out\install by default.)

  • Start a development command line (or PowerShell) in the project directory and run cmake and msbuild.exe manually. See https://docs.microsoft.com/en-us/cpp/build/building-on-the-command-line for more details. Example procedure, assuming Visual Studio 2019 Community Edition:

    "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\Common7\Tools\VsDevCmd.bat"
    mkdir build
    cd build
    cmake ..
    msbuild.exe INSTALL.vcxproj

Keep in mind, that you will still need to install tclsh, bison and flex manually (see above) and add them to your %PATH%.

Packaging / Installing

Using the CMake install feature, we can pack or install the ACPLT/OV runtime executable together with the libraries for dynamic linking and templates/example files of the libraries. To do so, we must specify the install prefix (CMAKE_INSTALL_PREFIX) in the CMake configuration step and run the installation with make install. The install prefix may be set to /usr or /usr/local for direct install on Linux or to a relative directory path (e.g. export/) for packing the install tree in a local directory:

cd build
cmake -DCMAKE_INSTALL_PREFIX=export/ ..
make -j 4 install

Executables (like the ov_runtimeserver) will be placed in bin/ and the OV libraries inlib/, inside the prefix directory. Some OV libraries will install additional files like .fbd template files to the install prefix, typically in templates/. When building for Windows platforms, the OV libraries' DLL files will be added to the bin directory. This convention will ensure that the ov_runtimeserver will find the OV libraries for dynamic linking.

Cross Compiling

Cross compilation is possible and tested on Linux, especially using arm-linux-gnueabihf-gcc to target Raspberry Pi and similar platforms or using x86_64-w64-mingw32-cpp to target Windows systems. CMake toolchain files for those target platforms are provided in the repository. However, cross compilation requires additional steps to make the code generation tool available on the host platform before starting the build:

At first, a build tree for the host platform has to be set up. It is used to compile only the make target ov_codegen. Afterwards, a new build tree for the target platform is set up, using the CMake variables CMAKE_TOOLCHAIN_FILE to select the compilation toolchain and OV_CODEGEN_PATH to provide the path to the previously compiled ov_codegen tool.

git clone https://github.com/acplt/rte.git
cd rte
mkdir build
cd build
cmake ..
make -j 4 ov_codegen

cd ..
mkdir build-win32
cd build-win32
cmake -DCMAKE_TOOLCHAIN_FILE=../cmake/gcc-linux-armhf.cmake -DOV_CODEGEN_PATH="../build/core/codegen/ov_codegen" ..
make -j 4 ov_codegen

Static linking

The build system allows to link the ov_runtimeserver statically with all OV libraries of a project. This way, the ov_runtimeserver can be installed as a single binary without additional .dll/.so files required at runtime.

To enable static linking, set the CMake variable OV_STATIC_LIBRARIES=ON. This will change the definition of all OV libraries in the project to static libraries (instead of dynamic libraries) and add them as linker dependencies to the ov_runtimeserver executable.

Usage

The main application of ACPLT/RTE is the ov_runtimeserver / ov_runtimeserver.exe. It manages the object database and dynamically links with OV libraries which provide functionality by adding object classes. With the help of TCPbind, ksbase and kshttp OV libraries (which are bundled in this repository), it allows to provide an ACPLT/KS server. Alternatively, the opcua OV library (also bundled in this repository) can be used to run an OPC UA / IEC 62541 server in OV.

The ov_runtimeserver can either be configured by command line arguments or using a configuration file. A simple invocation for starting an OV server with a plain new object database would be

ov_runtimeserver -f test.ovd -s MANAGER -l stdout -c 10000000 --force-create -w ksbase -w kshttp -w TCPbind -w fb

The parameters will make the ov_runtimeserver create a new database file (--force-create) of size 10 MB (-c) named test.ovd (-f), start a server named "MANAGER" (-s) with that database, and load the OV libraries ksbase, kshttp, TCPbind and fb, which will make it accessible via TCP port 7509 and provide basic functionality for function block networks.

As an alternative, these configuration options can be supplied via configuration file. An example configuration file is provided in misc/MANAGER.conf in this repository. The path to the configuration file is specified via the -cf command line option.

Running from build tree

On Linux, thanks to the RPATH set by CMake, the ov_runtimeserver can be started directly from the build tree and will find all the shared libraries which have been compiled in the same build setup.

core/runtimeserver/ov_runtimeserver -cf MANAGER.conf

Unfortunately, Windows executables don't support something like RPATH. To test ov from the build tree all build artefacts need to be moved to a single directory. This can be achieved by adding following lines to the CMakeList.txt in the project root directory.

set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/out CACHE STRING "" )
set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/out CACHE STRING "" )

This creates a directory out in the cmake build directory where all executables and shared libraries are moved.

Running from installed package

On unixoid systems, if ACPLT/RTE has been installed via make install without a special prefix, ov_runtimeserver should be part of the $PATH and be invokable like any other installed program. The OV libraries are placed in a system shared library directory, so they are automatically found for dynamic linking.

If, on the other hand, ACPLT/RTE has been packaged in a local directory, the ov_runtimeserver will need a proper LD_LIBRARY_PATH environment variable, pointing to the lib/ directory, to locate the OV libraries at runtime. On Windows, this is not necessary, since the OV library DLL files are placed in the bin/ directory together with the executable program.

Clients

Interaction with an ACPLT/OV server (including reading/writing values, creating, moving, deleting and linking objects, etc.) is typically done via the ACPLT/KS protocol. Different KS clients and KS client libraries for different purposes exist:

  • acplt-ks: a C++ KS library for writing KS clients and servers
  • fb_dbcommands: a simple command line client program for dumping and loading OV data into/from the "FBD" file format
  • iFBSpro: A full-featured KS graphical client, originally intended for function block engineering, written in Tcl and Tks
  • henson: A client-side web-based KS client for inspecting and modifying OV objects, written in JavaScript

An (incomplete) approach towards a Python client library for KS servers can be found in the tools/kshttp-python/ directory of this repository.

Miscellaneous Tools

  • TODO usage of ov_dbutil

Creating and building custom libraries

ACPLT/OV is meant to be extended with custom libraries, providing custom object classes. The object-oriented interface of these libraries is described using a domain specific language (DSL), the .ovm files. In these files, object classes with attributes, operations and inheritance are declared as well as relationship types.

From this declaration, C header files and some definitions are generated during compilation. The corresponding implementation code needs to provided as C source files. Templates for these source files can also be generated from the OVM model (see below).

Project structure

Due to the required build metadata for code generation an compilation, OV libraries must be compiled in a single CMake project with all their OV library dependencies, including the OV core libraries in this repository. This means, that the sources of library dependencies must be added as a subdirectory of the custom library project (preferably as a Git submodule). As an alternative, the CMake build metadata can be imported from another project's build tree, which uses export(EXPORT …). This allows to build dependency projects (like the OV core libraries) first and use find_package() in the dependent project's CMakeLists.txt for using the required build metadata, sources and build artifacts.

To allow reusing of custom OV libraries in different application scenarios, we propose implementing them in separate "library projects" that can then be used as a dependency in application-specific projects. Application-specific projects can then include all the required library projects as well as this rte repository subdirectories (Git submodules) to build the runtime environment and all required libraries in a single build step. To allow any combination of library projects in an application-specific project, library projects should not include the rte core project. Instead they should use the find_package() import method, as described above, to be build independently.

Application-specific project

According to the description above, the structure of a library project should look like this:

project_example/
├── project_specific_library/
│   └── ...                      [see below]
├── rte/                         [preferrably as Git submodule]
│   └── ...
├── rte_fblib/                   ["official" library project, preferrably as Git submodule]
│   └── ...
├── rte_example_libs/            [preferrably as Git submodule]
│   └── ...
└── CMakeLists.txt

With a CMakeLists.txt, similar to this:

cmake_minimum_required(VERSION 3.13)

project(project_example C)

# Disable default build of unrequired libraries from library projects
# (can be re-enabled during CMake build using -DBUILD_OV_LIBRARY_UDPbind=ON etc.
#  Libraries can also still be build explicitly with `make UDPbind` etc.)
SET(BUILD_OV_LIBRARY_UDPbind OFF CACHE BOOL "")
SET(BUILD_OV_LIBRARY_SSChelper OFF CACHE BOOL "")

# Include ov core project from subdirectory
add_subdirectory(rte)

# Include ov helper functions and definitions
# (required for `add_ov_library()` etc.)
get_target_property(RTE_CMAKE_DIR ov RTE_CMAKE_DIR)
include(${RTE_CMAKE_DIR}/ov_functions.cmake)
include(${RTE_CMAKE_DIR}/ov_definitions.cmake)

# Include library projects
add_subdirectory(rte_fblib)
add_subdirectory(rte_example_libs)

# Include libraries of this project
add_subdirectory(project_specific_library)

# Finish up build file generation, especially for statically linked builds
# Must be called after last OV library definition!
ov_finish_project()

Library project

In contrast, a library project should only include the OV library sources, not the rte project or any other library project:

rte_example_libs/
├── example_library/
│   └── ...                      [see below]
├── another_example_library/
│   └── ...
└── CMakeLists.txt

However, we need some additional tricks in the CMakeLists.txt to locate the RTE project's build metadata (and maybe other library projects' build metadata) and export this project's build metadata to be included by other library projects:

cmake_minimum_required(VERSION 3.13)

project(rte_example_libs C)

# Find and include RTE CMake project (if it isn't already included by a parent project)
find_package(RTE)

# Include ov helper functions and definitions
get_target_property(RTE_CMAKE_DIR ov RTE_CMAKE_DIR)
include(${RTE_CMAKE_DIR}/ov_functions.cmake)
include(${RTE_CMAKE_DIR}/ov_definitions.cmake)

# Add libraries of this project
add_subdirectory(example_library)
add_subdirectory(another_example_library)

# Export all targets to allow importing from build tree via find_package(RTEExampleLibs)
# This does only make sense if this is the root project
get_directory_property(hasParent PARENT_DIRECTORY)
if(NOT hasParent)
    export(EXPORT ${PROJECT_NAME}
         FILE RTEExampleLibsConfig.cmake)
endif()

If we want to build this library project as a stand-alone project (i.e. not as a subdirectory of an application-specific project), we need to build the rte project before and make sure that find_package(RTE) is able to find the rte project's metadata in its build directory. This can be achieved by setting the cmake variable CMAKE_PREFIX_PATH to the path to rte's build directory when invoking CMake for the library project:

...
cmake -DCMAKE_PREFIX_PATH="/path/to/rte/build" ..
make -j4

Creating a new library

For the provided tools (CMake build scripts, acplt_builder for templates) to work, OV libraries must follow the following file structure:

<library_name>/
├── include/                     [optional, for custom C header files]
│   └── example_custom_header.h
├── model/
│   ├── <library_name>.ovm
│   ├── <library_name>.ovf       [optional, C header for function type decl.]
│   └── <library_name>.ovt       [optional, C header for data type declarations]
├── source/
│   ├── className1.c
│   ├── className2.c
│   └── example_custom_code.c
└── CMakeLists.txt

For simple OV libraries, the CMakeLists.txt file only requires two lines:

add_ov_library(<library_name>)
ov_library_includes(<library_name> <dependency1> <dependency2>)

The add_ov_library() function defines a CMake library target for the OV library, which is compiled from all .c files in the source/ directory as well as the sources generated from the OVM file in the model/ directory. The include/ directory is automatically added as include path for compilation (if present). The ov_library_includes() defines OV library dependencies which must be considered for code generation and linking. It also sets up the OV code generation build step for the library.

Notes:

  • ov_library_includes() must name all OV library dependencies that are listed as #include in the library's OVM file.
  • ov_library_includes() must always be present. Typically there's at least one dependency, e.g. on one of the core libraries like ov or fb.
  • ov_library_includes() must not be used to link against other libraries/targets than OV libraries.

Additional settings for the compilation or linking of the library can be added through the usual CMake commands, e.g.

  • adding compiler flags like target_compile_definitions(<library_name> PRIVATE -DMY_PREPROCESSOR_DEF=1)
  • linking additional libraries like
    if (WIN32)
        target_link_libraries(<library_name> PRIVATE winmm.lib ws2_32.lib)
    endif()
    (If compiled libraries are provided with the OV library sources, they should by convention be placed in a lib/ directory.)
  • Adding additional build steps for preparation of additional sources like
    add_custom_command (
        OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/generated_foo.c
        COMMAND tclsh ${CMAKE_CURRENT_SOURCE_DIR}/prebuild.tcl foo.txt ${CMAKE_CURRENT_BINARY_DIR}/generated_foo.c
        DEPENDS foo.txt
        COMMENT "Using tclsh to generate foo for <library_name>"
        )
    target_sources(<library_name> PRIVATE ${CMAKE_CURRENT_BINARY_DIR}/generated_foo.c)

Generating library source templates

As explained above, the C header files and some definitions for the object-oriented model of the library are generated from the OVM file during build, using the ov_codegen program. However, the implementation of getter- and setter functions and operations, including inherited operations, must be provided in the user-provided C code. Their signature and naming must match the declarations in the generated C headers exactly to pass compilation.

The acplt_builder program can be used to help with this. It generates template C files from the OVM definition and places them in a subdirectory source/sourcetemplates/ in the library's directory.

The usage is really simple:

acplt_builder -f -l path/to/library/model/library.ovm

However, to generated these templates, acplt_builder must be able to find and read OVM files of all dependency libraries, that have been included into the target library's OVM file. In contrast to ov_codegen, which gets the paths of included OVM files from the CMake build system, acplt_builder searches a list of search paths for the included libraries. These search paths include the current working directory, its parent directory and the target library's parent directory. Other search paths can be added via the OV_LIBRARY_PATH environment variable. Similar to the $PATH/%PATH% variable, it should contain a list of directories, which contain libraries, separated by colon : on *NIX systems resp. semicolon ; on Windows.

To use acplt_builder for building the project_specific_library from the application-specific project example above, you would have to use it as follows (assuming, that you use ./build as the build directory and already compiled acplt_builder):

cd project_example/
export OV_LIBRARY_PATH="$(pwd)/rte/core:$(pwd)/rte/syslibs:$(pwd)/rte_fblib:$(pwd)/rte_example_libs"
build/rte/core/acplt_builder -f -l project_specific_library/model/project_specific_library.ovm

You may want to create a shell script for each project to setup a development environment correctly, e.g.:

#!/bin/bash
# to be placed in project root ...
SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"

export OV_LIBRARY_PATH="${SCRIPT_DIR}/rte/core:${SCRIPT_DIR}/rte/syslibs:${SCRIPT_DIR}/rte_fblib:${SCRIPT_DIR}/rte_example_libs"
export PATH="${PATH}:${SCRIPT_DIR}/build/rte/core/"

You may also want to configure your IDE to invoke acplt_builder every time the model files are edited. An example configuration for CLion is provided in tools/CLion.

Documentation

Debugging

ACPLT/RTE can generally be debugged like any other C program. One feature of ACPLT/RTE is the persistent database which is used to preserve structures and states between server starts.

On its own, the software already provides some debugging functionality. In debug builds the server will check the consistency of the database during shutdown. This includes the allocated data blocks in the database. The server will report allocated blocks that are no longer reachable through the OV data model and thus will be lost on the next start. The origin of lost blocks can be investigated with valgrind.

A list of allocated pointers in a database can be obtained with the ov_dbutil executable and the --print-pointer option.

Pointer found in the data model are listed with the offset from the start of the database, the object id of the containing ov object and the path to the object. Lost pointers are presented with the offset from the start of the database, the content of the block as a string, its length and the block number.

Valgrind

ACPLT/RTE also provides extended valgrind support for debugging the persistent database. This support can be activated via the OV_VALGRIND cmake option. Instances build with this option and invoked with valgrind, will show mem leaks and access warnings for the database as well as for the default memory pool (malloc).

This however decreases the performance of the custom memory allocator dramatically, to the point where the webinterface may no longer be usable. For such cases, allocation in the database can be redirected to malloc. To start redirecting allocations, change the variable /vendortree/use_malloc to True. This step cannot be reversed during runtime, and the database cannot be started again.