Continuous Integration Testing

It is important to ensure that any change made to framework code does not introduce additional bugs. In practice, this is impossible to guarantee but we can at least perform a full build of the entire framework with all samples (both in the framework any any associated libraries).

This is done using the integration testing framework using linux and Windows build environments.

In addition, a number of integration tests are run using Host builds which verify the logic of a large proportion of the code. Testing low-level operation requires real hardware and this must be done manually, but in general libraries and samples can be largely tested using Host builds and carefully constructed tests.

The SmingTest library should be used for such test applications to ensure they are supported on all architectures. It also provides a mechanism for logging test results.

Github Actions

We use Github Actions to manage all test builds. This service is free of charge for open-source projects.


Appveyor has been removed in favour of GitHub Actions.

We used to use Travis but this is no longer free of charge.

Sming performs the build and test logic is handled using scripts, which are intended to be easily portable to other CI services if necessary. Mostly batch scripts (.cmd) are used for Windows, and bash scripts (.sh) for GNU/Linux but where practical powershell core is used as this runs on either.


Sming doesn’t perform CI builds for MacOS.

Library CI support

Sming libraries may be separately built and tested whether or not they are included as part of the Sming repository (or a fork).

There are two mechanisms available.

GitHub Actions

The library.yml reusable workflow is provided, which takes care of these tasks:

  • Checking in the library to test

  • Checking in the Sming framework

  • Installing build tools

  • Builds all applications within the library’s samples directory, for all supported architectures

  • If a test application is provided then that should be located in a test directory. This is built for all architectures, and also executed for Host.

Builds are handled using Tools/ci/library/Makefile.

See also

To use this in a project, add a suitable workflow to the .github/workflows directory. Templates are provided in the .github/workflows/library directory.

Here is the basic push scenario:

name: CI Push
on: [push]
    uses: SmingHub/Sming/.github/workflows/library.yml@develop
    # Inputs are all optional, defaults are shown
      # Repository to fetch Sming from
      sming_repo: ''
      # Sming branch to run against
      sming_branch: 'develop'
      # Library alias
      alias: ''

The sming_repo and sming_branch inputs are provided if your library requires modifications to Sming which are not (yet) in the main repository.

The alias input is required where the library repository name does not correspond with the working title. For example, the jerryscript library is in a repository called Sming-jerryscript, so must be checked out using a different name. If Sming contains a library (or Component) with the same name then it will be overridden, with a warning Multiple matches found for Component 'jerryscript' in the build log.

The ci-dispatch.yml example demonstrates manual triggering, which allows these inputs to be easily changed. See

Note that the workflow must be available in the library’s default branch, or it will not appear in the github web page.


Appveyor may be configured to test a Sming library separately. Steps to enable:

Add project to appveyor account

Projects -> New Project and select from list

Set Custom Configuration


Set Project URL slug

If the library under test already exists in the Sming framework then the test directory MUST have the same name to ensure it gets picked up.

For example, testing the Sming-jerryscript library requires this value to be set to jerryscript to match the Sming library name. Build logs should then report a warning Multiple matches found for Component 'jerryscript'.

Set sming fork/branch

By default builds use the main Sming develop branch. If testing a library which requires changes to the framework, you’ll need to use a fork and add SMING_REPO and SMING_BRANCH environment variables to the project settings.

Note that environment variables set here will override any values set in appveyor.txt.

The provided default makefile builds all applications within the library’s samples directory. If a test application is provided then that should be located in a test directory. This is built for all architectures, and also executed for Host.

Build on your own ‘cloud’

Resources are limited to one concurrent build job per appveyor account. Each build gets two virtual CPUs but they’re not particular fast. Network connectivity is, on the other hand, excellent!

One very useful feature that appveyor provides is Bring Your Own Cloud or Computer. This allows the actual builds to be run on other hardware.

Builds can have up to 5 concurrent jobs and as many CPUs as are available. In addition, build images can be pre-installed with toolchains. This can reduce total run times from 5+ hours to around 30 minutes.


Full support requires a Windows server with Hyper-V, WSL2 and Docker installed. Hyper-V is built into Windows 10/11 professional edition. WSL2 should be available on all Windows versions.

Linux/MacOS are supported but only for GNU/Linux images.

Note that whilst Docker supports both Windows and Linux images, both cannot be used at the same time: it is necessary to manually switch between Linux/Windows containers. However, testing shows much better performance using Hyper-V for Windows builds.

  1. Add Docker build cloud for Linux builds:

    • Appveyor -> BYOC -> Add Cloud

    • Cloud Provider: Docker

    • Operating system: Windows

    • Base Image: Ubuntu 20.04 Minimal

    • Shell commands:

      git clone --branch develop --depth 1 /tmp/sming
      pwsh /tmp/sming/Tools/Docker/appveyor/setup.ps1
    • Image name: linux

    Execute commands as indicated in the resulting screen.

    Wait for the image to be built.

    The final stage updates the cloud information in your appveyor account. Customise as follows:

    • Name

      Change this so it contains only letters, numbers and dash (-). Default names contain a space, e.g. COMPUTER Docker so change to COMPUTER-Docker

    • Custom Docker command arguments

      Customise CPU resources, RAM usage, etc. For example:



    • Failover strategy

      Default values will fail a job if no worker is available to service it. The following settings are suggested:

      Job start timeout: 60
      Provisioning attempts: 100
  2. Add Hyper-V build cloud for Windows builds:

Same as (1) above except:

  • Cloud Provider: Hyper-V

  • Base Image: Windows Server Core 2019 Minimal

  • Image name: windows

When complete, fix the build cloud name as previously, e.g. COMPUTER-HyperV. Also check CPU cores, RAM allocation, failover strategy.

  1. Fix authorization token

The above steps will also install the Appveyor Host Agent software on your computer. This is the software which communicates with the Appveyor server and directs the build jobs.

The authorization token used by the agent can be found in the registry:


Make sure that both clouds have the same token.

  1. Configure BYOC images

    Select BYOC -> Images and amend mappings as follows:

      • Image Name: Ubuntu2004

      • OS Type: Linux

      • Build cloud: “COMPUTER-Docker” (as configured above)

      • Image Name: Visual Studio 2019

      • OS Type: Windows

      • Build cloud: “COMPUTER-HyperV” (as configured above)

    Now, when a build is started it should use your own server. To revert back to normal operation change the Image Name fields in the entries. It’s not necessary to delete them: just add, say, “X” to the name so they’re not recognised.


    Clouds may also be configured on a per-project basis by setting the APPVEYOR_BUILD_WORKER_CLOUD environment variable to the appropriate cloud name.

    To get both Linux and Windows builds working concurrently using this approach would require a single cloud to support dual images.

Rebuilding docker images

Appveyor images are customised by pre-installing Sming build tools. When these are updated images must be re-built.

The easiest way to do this is using the provided dockerfiles:

cd $SMING_HOME/../Tools/Docker/appveyor
docker build --no-cache -t linux -f Dockerfile-Linux .
docker build --no-cache -t windows -f Dockerfile-Windows .

Custom images

To use a Sming fork for building the image simply replace the repo URL and branch in the Shell Commands given above.

These may also be passed to docker build as follows:

docker build -t linux-test -f Dockerfile-Linux --build-arg SMING_REPO= --build-arg SMING_BRANCH=feature/appveyor-revisions .


If you get error image not supported by cloud this probably means an image has been mapped to the wrong clould. Goto Appveyor -> BYOC -> Images and update/delete the offending entries.

If either cloud is shown as offline then check the authorization token (step 4 above). It may be necessary to restart the Appveyor Host Agent service (via Windows service manager).