Skip to content

epics-base/ci-scripts

Repository files navigation

Version Travis status AppVeyor status GitHub Actions status GitLab CI/CD status

Continuous Integration for EPICS Modules

The scripts inside this repository are intended to provide a common, easy-to-use and flexible way to add Continuous Integration to EPICS software modules, e.g. Device or Driver Support modules.

By including this repository as a Git Submodule, you will be able to use the same flexible, powerful CI setup that EPICS Bases uses, including a way to specify sets of dependent modules (with versions) that you want to compile your module against.

By using the submodule mechanism, your module will always use an explicit commit, i.e. a fixed version of the scripts. This ensures that any further development of the ci-scripts will never break your existing use.

This Repository

In addition to the script that runs the builds and tests, this repository contains service specific documentation and example configuration files (in the subdirectories), and a small test suite that is used to verify functionality and features of the ci-scripts module itself

The example files are your best reference. They are kept up-to-date and show a fully-featured and a minimal setup.

You are welcome to use the test suite as a secondary reference, but keep in mind that in your main module the path to the scripts has one level more (e.g., ./abc here would be ./.ci/abc in your module). Also, the test suite does not show the same quality and documentation levels as the example files.

Features

  • Compile against different branches or releases of EPICS Base and additional dependencies (modules like asyn, std, sequencer, etc.).

  • Define setup files that declare sets of dependencies with their versions and locations.

  • Define hooks for any dependency. Hooks are run on the dependency module before it is compiled, so the module can be patched or further configured.

  • Define shared (default) or static builds (for executables and libraries).

  • Define optimized (default) or debug builds.

  • Run tests (using the EPICS build system, i.e., make runtests and friends).

Supported CI Services

The listed properties and instructions for the CI services apply to their free tiers for open source projects, hosted in the cloud on their infrastructure.

The companies behind these services also offer "enterprise" installations on customer infrastructure, which will have different performance and limitations.

  • Two parallel runners on Linux/Windows (two runners on MacOS)
  • Ubuntu 14/16/18/20, MacOS 10.13/14/15, Windows Server v1809
  • Compile natively on Linux (different versions of gcc, clang)
  • Compile natively on MacOS (clang)
  • Compile natively on Windows (gcc/MinGW, Visual Studio 2017)
  • Cross-compile for Windows 32bit and 64bit using MinGW and WINE
  • Cross-compile for RTEMS 4.9 and 4.10 (pc386, Base >= 3.15)
  • Cross-compile for RTEMS 5 (10 BSPs, Base >= 7.0.5.1)
  • Built dependencies are cached (for faster builds).

See specific ci-scripts on Travis-CI README for more details.

  • One parallel runner (all builds are sequential)
  • Windows Server 2012/2016/2019
  • Compile using gcc/MinGW or different Visual Studio versions:
    2010, 2012, 2013, 2015, 2017, 2019, 2022
  • Compile for Windows 32bit and 64bit
  • No useful caching available.

See specific ci-scripts on AppVeyor README for more details.

  • 20 parallel runners on Linux/Windows (5 runners on MacOS)
  • Ubuntu 18/20/22, MacOS 11/12, Windows Server 2019/2022
  • Compile natively on Linux (gcc, clang)
  • Compile natively on MacOS (clang)
  • Compile natively on Windows (gcc/MinGW, Visual Studio 2019/2022)
  • Cross-compile for Windows 32bit and 64bit using MinGW and WINE
  • Cross-compile for RTEMS 4.9 and 4.10 (pc386, Base >= 3.15)
  • Cross-compile for RTEMS 5 (10 BSPs, Base >= 7.0.5.1)
  • Caching not supported by ci-scripts yet.

See specific ci-scripts on GitHub Actions README for more details.

  • Docker-based runners on Linux (one VM instance per job)
  • Can use any Docker image from Dockerhub (the examples use ubuntu:bionic)
  • Compile natively using different compilers (gcc, clang)
  • Cross-compile for Windows 32bit and 64bit using MinGW and WINE
  • Cross-compile for RTEMS 4.9 and 4.10 (pc386, Base >= 3.15)
  • Cross-compile for RTEMS 5 (10 BSPs, Base >= 7.0.5.1)
  • Built dependencies are cached (for faster builds).

See specific ci-scripts on GitLab CI/CD README for more details.

How to Use the CI-Scripts

  1. Get an account on a supported CI service provider platform (e.g. Travis-CI, AppVeyor, ...). GitHub Actions does not require a separate account.

    (More details in the specific README of the subdirectory.)

  2. In your module, add this ci-scripts repository as a Git Submodule (name suggestion: .ci).

    git submodule add https://github.com/epics-base/ci-scripts .ci
  3. Create setup files for different sets of dependencies you want to compile against. (See below.)

    E.g., a setup file stable.set specifying

    MODULES=sncseq asyn
    
    BASE=3.15
    ASYN=R4-34
    SNCSEQ=R2-2-8
    

    will compile against the EPICS Base 3.15 branch, the Sequencer release 2.2.8 and release 4.34 of asyn. (Any settings can be overridden from the specific job line in the service configuration, e.g., .travis.yml.)

  4. Create a configuration for the CI service by copying one of the examples provided in the service specific subdirectory and editing it to include the jobs you want the service to run. Use your setup by defining e.g. SET=stable in the environment of a job.

  5. Push your changes and check the CI service for your build results.

Calling the cue.py Script

Independent from CI service and platform, the runner script is called from your main configuration as:

python .ci/cue.py <action>

where <action> is one of:

prepare
Prepare the build by cloning Base and the configured dependency modules, set up the EPICS build system, then compile Base and these modules in the order they appear in the MODULES setting.

build
Build your main module.

test
Run the tests of your main module.

test-results
Collect the results of your tests and print a summary.

exec
Execute the remainder of the line using the default command shell.

Extra arguments to make

You can add additional arguments to the make runs that the cue.py script starts. Put your additional arguments into environment variables named EXTRA, EXTRA1, ... EXTRA5.

The variables may contain multiple arguments, separated by whitespace. Use regular shell script quoting (single/double quotes, backslash escapes) if you need spaces inside an extra argument.

The YAML syntax needed to set environment variables depends on the CI service and platform. (See the full example configuration file.)

Setup Files

Your module might depend on EPICS Base and a few other support modules. (E.g., a specific driver might need StreamDevice, ASYN and the Sequencer.) In that case, building against every possible combination of released versions of those dependencies is not possible: Base (39) x StreamDevice (50) x ASYN (40) x Sequencer (52) would produce more than 4 million different combinations, i.e. build jobs.

A more reasonable approach is to create a few setups, each being a combination of dependency releases, that do a few scans of the available "version space". One for the oldest versions you want to support, one or two for stable versions that many of your users have in production, one for the latest released versions and one for the development branches.

A job uses a setup file if SET=<setup> (without the .set extension of the setup file) is set for the job in the main configuration file.

Setup File Syntax

Setup files are loaded by the build script. They are found by searching the locations in SETUP_PATH (space or colon separated list of directories, relative to your module's root directory).

Setup files can include other setup files by calling include <setup> (again omitting the .set extension of the setup file). The configured SETUP_PATH is searched for the include.

Any VAR=value setting of a variable in a setup file is only executed if VAR is unset or empty. That way any settings can be overridden by setting them in the job description inside the main configuration file (e.g., .travis.yml).

Empty lines or lines starting with # are ignored.

MODULES=<list of names> should list the dependencies (software modules) by using their well-known slugs, separated by spaces. EPICS Base (slug: base) will always be a dependency and will be added and compiled first. The other dependencies are added and compiled in the order they are defined in MODULES.

Modules needed only for specific jobs (e.g., on specific architectures) can be added from the main configuration file by setting ADD_MODULES for the specific job(s).

REPOOWNER=<name> sets the default GitHub owner (or organization) for all dependency modules. Useful if you want to compile against a complete set of dependencies forked into your private GitHub area.

For any module mentioned as foo in the MODULES setting (and for BASE), the following settings can be configured:

FOO=<version> Set version of the module that should be used. Must either be a tag name or a branch name. [default: master]

FOO_REPONAME=<name> Set the name of the remote repository as <name>.git. [default is the slug in lower case: foo]

FOO_REPOOWNER=<name> Set the name of the GitHub owner (or organization) that the module repository can be found under.

FOO_REPOURL="<url>" Set the complete URL of the remote repository. Useful for dependencies that are not hosted on GitHub.

The default URL for the repository is pointing to GitHub, under $FOO_REPOOWNER else $REPOOWNER else epics-modules, using $FOO_REPONAME else foo and the extension.git.

FOO_DEPTH=<number> Set the depth of the git clone operation. Use 0 for a full clone. [default: 5]

FOO_RECURSIVE=YES/NO Set to NO (or 0) for a flat clone without recursing into submodules. [default is including submodules: YES]

FOO_DIRNAME=<name> Set the local directory name for the checkout. This will be always be extended by the release or branch name as <name>-<version>. [default is the slug in lower case: foo]

FOO_HOOK=<hook> Set the name of a .patch file, a .zip or .7z archive or a script that will be applied (using -p1), extracted or run after cloning the module, before compiling it. Working directory is the root of the targeted module, e.g., .../.cache/foo-1.2). [default: no hook]

FOO_VARNAME=<name> Set the name that is used for the module when creating the RELEASE.local files. [default is the slug in upper case: FOO]

The ci-scripts module contains default settings for widely used modules, so that usually it is sufficient to set FOO=<version>. You can find the list of supported (and tested) modules in defaults.set. Feel free to suggest more default settings using a Pull Request.

RTEMS

Cross-compiling to RTEMS versions 4.9, 4.10 or 5 is supported on supported CI services. For configuration see below. Tests can also be run cross-platform, using qemu.

The RTEMS 5 builds now include most of the BSPs with configuration in Base:

  • beatnik
  • gen68360
  • mcp750
  • mvme167
  • mvme2100
  • mvme3100
  • pc686 w/ libbsd
  • qoriq_e500 w/ libbsd
  • uC5282
  • xilinx_zynq_a9_qemu w/ libbsd

Build configuration can be found here. Set RSB_BUILD to select the RTEMS toolchain release name/data from https://github.com/mdavidsaver/rsb/releases.

RTEMS 5 builds need to be switched to ubuntu version >= 20 (aka. os: ubuntu-20.04 with GitHub Actions, dist: focal with Travis-CI or image: ubuntu:focal with GitLab CI/CD).

Cross Compilation

Setting the CI_CROSS_TARGETS environment variable enables cross-compiling from Linux to the provided targets architectures. The value of the environment variable must contain the EPICS architecture and - depending on the target - may contain additional information like the compiler prefix to be used or the version of the target OS.

Multiple cross-targets can be added to the CI_CROSS_TARGETS variable by separating them with a colon (:) character.

For example, possible values are:

  • linux-aarch64
  • linux-arm@arm-linux-gnueabi
  • linux-arm@arm-linux-gnueabihf
  • linux-ppc
  • linux-ppc64
  • win32-x86-mingw
  • windows-x64-mingw
  • RTEMS-pc386-qemu@4.9
  • RTEMS-pc386-qemu@4.10
  • RTEMS-pc686-qemu@5

Debugging

Setting VV=1 in your service configuration (e.g., .travis.yml) for a specific job will run the job with high verbosity, printing every command as it is being executed and switching the dependency builds to higher verbosity.

For debugging on your local machine, you may set CACHEDIR to change the location for the dependency builds. [default is $HOME/.cache]

Set PARALLEL_MAKE to the number of parallel make jobs that you want your build to use. [default is the number of CPUs on the runner]

Set CLEAN_DEPS to NO if you want to leave the object file directories (**/O.*) in the cached dependencies. [default is to run make clean after building a dependency]

Service specific options are described in the README files in the service specific subdirectories:

References: EPICS Modules Using ci-scripts

EPICS Base and its submodules pvData, pvAccess, pva2pva, PVXS

EPICS Modules: ASYN, autosave, busy, devlib2, ecmc, gtest, ip, lua, MCoreUtils, modbus, motor, mrfioc2, OPCUA, PCAS, softGlueZync, sscan, std, vac, xxx

ESS: EtherCAT MC Motor Driver

Migration Hints

Look for changes in the example configuration files, and check how they apply to your module.

If comments in the example have changed, copy them to your configuration to always have up-to-date documentation in your file.

2.x to 3.x Migration

Update the script and test settings in your configuration to call the new script, following the example file.

python .ci/cue.py <action>

AppVeyor

The configuration: setting options have changed; they are now default, static, debug and static-debug.

MinGW builds are now using the CMP: gcc compiler setting.

Adding arguments to make is supported through the EXTRA .. EXTRA5 variables. Each variable value will be passed as one argument.

Travis

The new BCFG (build configuration) variable accepts the same options as the AppVeyor configuration: setting. Replace anySTATIC=YES settings with BCFG=static.

Remove bash in the homebrew: section of addons:. There are no more bash scripts.

MinGW builds (cross-builds using WINE as well as native builds on Windows) are now using the gcc compiler setting. Since gcc is the default, you can simply remove compiler: mingw lines.

For Windows, Travis offers native MinGW and Visual Studio 2017 compilers. Use os: windows and set compiler: to gcc or vs2017 for those builds.

Chocolatey packages to be installed for the Windows jobs are set by adding them to the environment variable CHOCO.

Frequently Asked Questions

How can I see what the dependency building jobs are actually doing?

Set VV=1 in the configuration line of the job you are interested in. This will make all builds (not just for your module) verbose.

How do I update my module to use a newer minor release of ci-scripts?

Update the submodule in .ci first, then change your CI configuration (if needed) and commit both to your module. E.g., to update your Travis setup to release 3.3.0 of ci-scripts:

cd .ci
git pull origin v3.3.0
cd -
git add .ci
  # if needed:
  edit .travis.yml     # and/or other CI service configurations
  git add .travis.yml
git commit -m "Update ci-scripts submodule to v3.3.0"

Check the example configuration files inside ci-scripts (and their changes) to see what might be needed and/or interesting to change in your configuration.

Depending on the changes contained in the ci-scripts update, it might be advisable to clear the CI caches after updating ci-scripts. E.g., a change in setting up EPICS Base will not be applied if Base is found in the cache.

How do I add a dependency module only for a specific job?

Add the additional dependency in the main configuration file by setting ADD_MODULES for the specific job(s).

Why the name cue?

The noun cue is defined as "a signal (such as a word, phrase, or bit of stage business) to a performer to begin a specific speech or action". (Merriam-Webster)

Release Numbering of this Module

The module tries to apply Semantic Versioning.

Major release numbers refer to the API, which is more or less defined by the full configuration examples in the service specific subdirectories. If one of these files has to be changed for the existing configuration options or important new options are being added, a new major release is created.

Minor release numbers refer to additions and enhancements that do not require the configuration inside an existing user module to be changed. (Unless for using a new feature.)

Again: using the git submodule mechanism to include these scripts means that user modules always work with a fixed, frozen version. I.e., developments in the ci-scripts repository will never break an existing application. These release numbering considerations are just a hint to assess the risks when updating the submodule.

License

This module is distributed subject to a Software License Agreement found in file LICENSE that is included with this distribution.