Skip to content

gerkey/ros1_external_use

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

78 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Overview

This repository demonstrates how to use ROS software in C++ and Python via CMake and make. The goal is to show that, while ROS itself is developed using catkin, package.xml meta-data files, and other tools that help with managing a workspace drawn from multiple repositories, you don't have to use those tools when developing your software that uses ROS.

It should be easy to use ROS just as you would any other software dependency, without giving consideration to the tools that are used internal to the development of ROS (though you might also find those tools useful for your own projects). We can do this because (thanks to catkin) ROS packages all provide CMake configuration files that allow you to find_package() and use them in your own CMake build just like most modern software.

ROS packages also provide pkg-config files, which opens the door to doing everything directly with make, not CMake.

Each directory demonstrates a different use case. Please explore and report any problems that you encounter. Also, if there are use cases that you'd like to see demonstrated, please open an issue.

Pre-installation on Ubuntu Linux

On Ubuntu, we provide binary packages for ROS software. To configure your system to use them, you need to add our repository and key to your system, like so (assuming that you're using ROS Indigo on Ubuntu trusty):

sudo sh -c 'echo "deb http://packages.ros.org/ros/ubuntu $(lsb_release -sc) main" > /etc/apt/sources.list.d/ros-latest.list'
sudo apt-key adv --keyserver hkp://pool.sks-keyservers.net --recv-key 0xB01FA116
sudo apt-get update

Now you're ready to sudo apt-get install ROS packages as indicated in the detailed instructions in each example in this repository.

Pre-installation on Mac OSX

We need to install a few system dependencies (on Ubuntu this step is done automatically by the dependencies encoded in the binary packages). We get those dependencies from a combination of brew and pip:

# <Install brew and configure your environment to include /usr/local (http://brew.sh)>
# Update brew and add a tap that will provide some ROS-specific dependencies
brew update
brew tap ros/deps
# brew install some stuff
brew install cmake libyaml console_bridge
# Also get pip
sudo easy_install pip
# pip install some stuff
sudo -H pip install -U wstool rosinstall rosinstall_generator rospkg catkin-pkg Distribute
# Initialize rosdep, which we'll use later to make sure that
# package-specific system dependencies are satisfied
sudo rosdep init
rosdep update

Get the code (all platforms)

Whatever platform you're on, you'll need to clone this repo:

git clone https://github.com/gerkey/ros1_external_use

Preliminary documentation

Here we begin to document in a definitive manner the tools and techniques that are being demonstrated in the examples.

Build system choices

We recommend using either CMake or make. But you should be able to port from make to pretty much any other build system.

Environment configuration for building

If your ROS packages are installed to a system standard location, such as /usr (this is where the new upstream Debian packages get installed), then you probably don't need any special configuration. However, if your packages are installed to somewhere like /opt/ros (using the Ubuntu packages from OSRF) or in your home directory (if you built from source), then you'll need to modify some environment variables to allow all the various CMake, C++, and Python assets to be located.

To get your environment configured, you have two choices:

  1. Source the setup file that's provided with your installation. This is the easiest way to get the environment configuration. E.g., if you are using ROS Indigo from the OSRF packages, then you would do:

    . /opt/ros/indigo/setup.sh
  2. Set the required variables manually. If your packages are installed to <prefix> (e.g., /opt/ros/indigo, or $HOME/ros1_ws/install_isolated), then the following commands will get you configured for building:

    # To find_package() things from CMake, you need your ROS installation on
    # your CMAKE_PREFIX_PATH.
    export CMAKE_PREFIX_PATH=<prefix>:$CMAKE_PREFIX_PATH
    # To pkg-config things from make, you need your ROS installation on
    # your PKG_CONFIG_PATH.
    export PKG_CONFIG_PATH=<prefix>/lib/pkgconfig:$PKG_CONFIG_PATH
    # To find headers installed to <prefix>/include, you need to modify
    # CPATH.
    export CPATH=<prefix>/include:$CPATH
    # The following line may require modification depending on your Python
    # version and your system type (`dist-packages` vs. `site-packages).
    export PYTHONPATH=<prefix>/lib/python2.7/dist-packages:$PYTHONPATH

Environment configuration for running

After building, you normally install your software somewhere before executing it. This isn't always necessary, but it usually eventually is as you start doing more complex things, like code generation. The requirements for run-time environment configuration are:

  1. Start with a subset of the build-time environment configuration. Specifically, you need: CMAKE_PREFIX_PATH and PYTHONPATH (again, you could just source the ROS setup file to get all the variables).

  2. Extend your library path to find libraries from installed ROS packages, extend PATH to find top-level ROS executables, set the ROS_MASTER_URI, and set the ROS_PACKAGE_PATH (again, this would be handled for you by the ROS setup file):

    # Add the shared library location.
    #   Linux version: 
    export LD_LIBRARY_PATH=<prefix>/lib:$LD_LIBRARY_PATH
    #   OSX version:
    #export DYLD_LIBRARY_PATH=<prefix>/lib:$DYLD_LIBRARY_PATH
    # Extend PATH to include things like roslaunch
    export PATH=<prefix>/bin:$PATH
    # We need ROS_MASTER_URI to find the roscore
    export ROS_MASTER_URI=http://localhost:11311
    # We need ROS_PACKAGE_PATH to point to the installed ROS packages
    export ROS_PACKAGE_PATH=<prefix>/share
  3. Extend the environment configuration for your installed software, assuming that it's installed at <install_prefix>:

    export CMAKE_PREFIX_PATH=<install_prefix>:$CMAKE_PREFIX_PATH
    # Modify this line as needed for your installation choices:
    export PYTHONPATH=<install_prefix>/lib/python2.7/dist-packages:$PYTHONPATH
    export ROS_PACKAGE_PATH=<install_prefix>/share:$ROS_PACKAGE_PATH

Building C++ programs

To build your C++ application code against ROS packages, you need to assemble the right flags to pass to the compiler and linker.

CMake

ROS packages follow the CMake configuration protocol, which means that you can just find_package() each one and then use the resulting variables that are defined. To build an executable that relies on package foo, you would refer to foo_INCLUDE_DIRS and foo_LIBRARIES:

find_package(foo REQUIRED)
include_directories(${foo_INCLUDE_DIRS})
add_executable(myprogram myprogram.cpp)
target_link_libraries(myprogram ${foo_LIBRARIES})

Notes:

  • You don't have to call link_directories(${foo_LIBRARY_DIRS}) because ROS packages follow the recommended practice of returning absolute paths in ${foo_LIBRARIES}.
  • Another variable of interst in ${foo_DIR}, which points to <prefix>/share/foo; you might start there when locating package-specific assets.

make

ROS packages provide pkg-config files that let you get build flags from make (or from a shell script or the command line, or wherever). The equivalent of the CMake example using package foo looks like this in make:

foo_cflags = $(shell pkg-config --cflags foo)
foo_libs = $(shell pkg-config --libs foo)
# Work around a known issue with new linkers that don't accept -l:/path/to/lib
foo_libs_nocolon = $(subst -l:,,$(foo_libs))
myprogram: myprogram.cpp
	$(CXX) -Wall -o $@ $(foo_cflags) $< $(foo_libs_nocolon)

Notes:

  • You can get just the include dirs by calling pkg-config --cflags-only-I foo.
  • You might also call pkg-config --variable=prefix foo if you need to get the directory containing the package to locate other assets.

Using pre-existing ROS messages in C++

If you're using messages defined in an installed ROS package from your C++ program, then you just need to follow the steps from the previous section to build your program with the flags provided by that package.

Using pre-existing ROS messages in Python

Using messages defined in an installed ROS package from your Python program doesn't require any special configuration. As long as your PYTHONPATH points to your ROS installation, things like from std_msgs.msg import String will just work.

Doing code generation for custom messages

If you define your own ROS messages by writing .msg files in your own application, then you'll need to run the code generator(s) to produce the code required to instantiate the corresponding structures in memory.

Notes:

  • If you're using custom messages in Python, you should always make install and get your runtime environment configuration set up before trying to run any programs. Otherwise, you won't be able to import your generated Python classes.

CMake

Calling message code generators from CMake is simplified because the genmsg package provides macros to help. Let's say that you have two custom messages, Foo.msg and Bar.msg, and that they in turn use messages from two installed ROS packages foo_msgs and bar_msgs. Then you would do message generation like so:

# Give the project a name (good practice in general and required by the code
# generators to decide how to namespace and where to produce their outputs).
project(myproject)
# This line is temporarily required because of a bug in genmsg that will be fixed soon:
# https://github.com/ros/genmsg/pull/61
find_package(catkin REQUIRED)
# The genmsg package gives us the code generator macros
find_package(genmsg REQUIRED)
# We need to find_package() each ROS package that defines messages that our
# message use.
find_package(foo_msgs REQUIRED)
find_package(bar_msgs REQUIRED)
# Enumerate our custom message files
add_message_files(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} FILES Foo.msg Bar.msg)
# Do code generation, specifying which other message packages we depend on.
generate_messages(DEPENDENCIES foo_msgs bar_msgs)

That's code generation all set up. Now let's build a program that uses the output:

# Build a program that uses the custom messages
add_executable(use_custom_msg use_custom_msg.cpp)
# Ensure that code generation happens before building use_custom_msg by
# depending on a special target created by generate_messages().
add_dependencies(use_custom_msg ${PROJECT_NAME}_generate_messages)

Notes:

  • The C++ code for Foo.msg will be in devel/include/myproject/Foo.h and will define a message class myproject::Foo. Your include directories are automatically modified to allow your code to #include <myproject/Foo.h>.
  • The Python code for Foo.msg will be in devel/lib/python2.7/dist-packages/myproject (or similar) and will define a Python class Foo that can be imported from the module myproject.msg.
  • Depending on what ROS client libraries you have installed, code generators for other languages might run (e.g., for LISP). We don't cover those here.
  • All the generated code is automatically registered for installation when you later do make install.

make

Compared to CMake, it takes a bit more effort to call the message code generators from make, but it's definitely doable. Taking the example from above (Foo.msg and Bar.msg depend on messages from foo_msgs and bar_msgs), here's what you would do:

# Give the project a name (good practice in general and required by the code
# generators to decide how to namespace and where to produce their outputs).
project = myproject
# Get flags for the message packages that we depend on
foo_msgs_includes = `pkg-config --cflags-only-I foo_msgs`
bar_msgs_includes = `pkg-config --cflags-only-I bar_msgs`
# Enumerate our custom messages
msgs = Foo.msg Bar.msg
# Compute the expected outputs from the code generators
msgs_cpp = $(foreach msg, $(msgs), $(project)/$(basename $(msg)).h)
msgs_py = $(foreach msg, $(msgs), $(project)/msg/_$(basename $(msg)).py)
msgs_py_init = $(project)/msg/__init__.py
# Find C++ and Python code generators
gencpp_dir = $(shell pkg-config --variable=prefix gencpp)
genpy_dir = $(shell pkg-config --variable=prefix genpy)
# Compute include directories for all message packages. This is an overinclusive
# hack because it would be painful to recursively gather the message package
# dependencies (though in principle it can be done by calling
# `pkg-config --print-requires <pkg>` for each one). The goal is to end up with
# a set of strings of the form:
#  -Ifoo_msgs:/path/to/foo_msgs/msg -Ibar_msgs:/path/to/bar_msgs/msg
pkg_msg_dirs = $(wildcard $(gencpp_dir)/share/*/msg)
pkg_msg_includes = $(foreach dir, $(pkg_msg_dirs), -I$(shell basename `dirname $(dir)`):$(dir))
# General rule for doing C++ code generation. Args:
#  $(pkg_msg_includes) : collection of args of the form `-I:foo_msgs:/path/to/foo_msgs`
#  -p $(project) : the "package" name; determines the namespace used in the code
#  -o $(project) : the output directory name
#  -e $(gencpp_dir)/share/gencpp : location of templates needed by gencpp
$(project)/%.h: %.msg
	$(gencpp_dir)/lib/gencpp/gen_cpp.py $(pkg_msg_includes) -p $(project) -o $(project) -e $(gencpp_dir)/share/gencpp $<
# General rule for doing Python code generation. Args:
#  $(pkg_msg_includes) : collection of args of the form `-I:foo_msgs:/path/to/foo_msgs`
#  -p $(project) : the "package" name; determines the namespace used in the code
#  -o $(project) : the output directory name
$(project)/msg/_%.py: %.msg
	$(gencpp_dir)/lib/genpy/genmsg_py.py $(pkg_msg_includes) -p $(project) -o $(project)/msg $<
# Extra rule for generating the __init__.py module file
$(msgs_py_init): $(msgs_py)
	$(gencpp_dir)/lib/genpy/genmsg_py.py --initpy -p $(project) -o $(project)/msg

That's code generation all set up. Now let's build a program that uses the output:

# Build an executable that uses locally defined messages. Note that it depends
# on the output from the C++ generator.
use_custom_msg: use_custom_msg.cpp $(msgs_cpp)
	$(CXX) -Wall -o $@ $(foo_msgs_includes) $(bar_msgs_includes) -I. $<

Further, let's show how to install our custom messages, both the raw input and the generated output:

install_prefix ?= /tmp/$(project)
install: all
	# As usual, the python install location might vary from platform to platform
	mkdir -p $(install_prefix)/include/$(project) $(install_prefix)/lib/python2.7/site-packages/$(project)/msg $(install_prefix)/bin $(install_prefix)/share/$(project)/msg
        # Put the .msg files in share/myproject/msg
	cp -a $(msgs) $(install_prefix)/share/$(project)/msg
	# Install our executable that uses custom messages
	cp -a use_custom_msg $(install_prefix)/bin
	# Install the C++ generated code
	cp -a $(msgs_cpp) $(install_prefix)/include/$(project)
	# Install the Python generated code
	cp -a $(msgs_py) $(msgs_py_init) $(install_prefix)/lib/python2.7/site-packages/$(project)/msg
	# Drop an empty `__init__.py` file to make `myproject` into a Python module
	touch $(install_prefix)/lib/python2.7/site-packages/$(project)/__init__.py

Doing code generation for custom services

Code generation for services is very similar to how it's done for messges. Here we'll just note the differences.

CMake

Differences from message generation:

  • Instead of calling add_message_files(), you call add_service_files().

make

Differences from message generation:

  • The C++ generator will produce 3 files instead of one. Here's how to compute the names of those output files, given two input .srv files:

    srvs = Bat.srv Baz.srv
    srvs_cpp = $(foreach srv, $(srvs), $(project)/$(basename $(srv)).h $(project)/$(basename $(srv))Request.h $(project)/$(basename $(srv))Response.h)
  • The Python generator script is called gensrv_py.py instead of genmsg_py.py. Here are the update rules for calling that script (arguments are the same as before):

    # General rule for doing Python code generation
    $(project)/srv/_%.py: %.srv
            $(gencpp_dir)/lib/genpy/gensrv_py.py $(pkg_msg_includes) -p $(project) -o $(project)/srv $<
    # Extra rule for generating the __init__.py module file
    $(srvs_py_init): $(srvs_py)
            $(gencpp_dir)/lib/genpy/gensrv_py.py --initpy -p $(project) -o $(project)/srv

Doing code generation for custom actions

Code generation for actions is two steps:

  1. Run the action code generator to process an .action file into multiple .msg files.
  2. Run the usual message code generator on the resulting .msg files.

CMake

Differences from message generation:

  • Always find_package(actionlib_msgs) because the auto-generated .msg files will refer to messages that are in that package.
  • Instead of calling add_message_files(), call add_action_files().

If you further want to use the actionlib libraries, which help you to build action clients and servers, then you'll also want to find_package(actionlib) and use the resulting variables in building your executables.

Make

Differences from message generation:

  • You need to enumerate your .action files, then compute multiple .msg outputs that will be produced, e.g.:

    actions = Foo.action Bar.action
    msgs = $(foreach msg, $(actions), $(basename $(msg))Action.msg $(basename $(msg))ActionGoal.msg $(basename $(msg))ActionResult.msg $(basename $(msg))ActionFeedback.msg $(basename $(msg))Goal.msg $(basename $(msg))Result.msg $(basename $(msg))Feedback.msg)
  • With the .msg targets known, do the usual computation of their language-specific targets:

    msgs_cpp = $(foreach msg, $(msgs), $(project)/$(basename $(msg)).h)
    msgs_py = $(foreach msg, $(msgs), $(project)/msg/_$(basename $(msg)).py)
    msgs_py_init = $(project)/msg/__init__.py
  • Find and use the action code generator:

    actionlib_msgs_dir = $(shell pkg-config --variable=prefix actionlib_msgs)
    # General rule for doing .action -> .msg generation
    %Action.msg %ActionGoal.msg %ActionResult.msg %ActionFeedback.msg %Goal.msg %Result.msg %Feedback.msg: %.action
    	$(actionlib_msgs_dir)/lib/actionlib_msgs/genaction.py -o . $<

If you further want to use the actionlib libraries, which help you to build action clients and servers, then you'll also want to use pkg-config to get flags for actionlib and use them in building your executables.

Installing for use by tools like roslaunch

If you've ever used ROS, you know how useful tools like roslaunch, rosmsg, and rostopic are (maybe also rosrun, if you like to invoke package-specific executables). These tools share a requirement, which is the ability to locate package-specific assets at run time. To use them with your own software, you need to make your code look a little like a ROS package. This is easy to do during installation.

For example, let's say that you build two nodes, talker, and listener, and you have a launch file, talker_listener.launch that refers to them like so:

<launch>
  <node pkg="myproject" type="talker" name="talker" output="screen"/>
  <node pkg="myproject" type="listener" name="listener" output="screen"/>
</launch>

What's required to run that launch file, so that roslaunch can find your programs? We need to do a few things:

  1. Have a simple package.xml file in <install_prefix>/share/myproject.
  2. Create an empty marker file <install_prefix>/.catkin. If <install_prefix> is included in the CMAKE_PREFIX_PATH, then this marker file will cause roslaunch and friends to look in there for package-specific executables.
  3. Install the executables to <install_prefix>/lib/myproject. That's where roslaunch will look.

make

Here's what the install rule could look like:

# Assume that you have earlier rules that will build the talker and listener
# executables
install: talker listener
	# Install the executables to <prefix>/lib/<project>, so that
	# they can be found by rosrun/roslaunch.
	mkdir -p $(install_prefix)/lib/$(project)
	cp -a talker listener $(install_prefix)/lib/$(project)
	# Install the .launch file; this could go anywhere, but `launch` is a
	# good pattern to follow.
	mkdir -p $(install_prefix)/share/$(project)/launch
	cp -a talker_listener.launch $(install_prefix)/share/$(project)/launch
	# Create a simple package.xml, which is necessary to be treated as
	# a ROS package.
	echo "<package><name>$(project)</name></package>" > $(install_prefix)/share/$(project)/package.xml
	# Create an empty .catkin marker file, which causes
	# rosrun/roslaunch to look in <prefix>/lib/project for executables
	# for each <prefix> in CMAKE_PREFIX_PATH
	touch $(install_prefix)/.catkin

Then, after you make install, if you have the run-time environment configuration set up properly, you should be able to do:

roslaunch myproject talker_listener.launch

Or, equivalently:

roslaunch <install_prefix>/share/myproject/launch/talker_listener.launch

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •