install_dependencies.sh continues on error

[nandlab]

Without the rapidjson-dev package, CMake exits with an error. Can you please add it to install_dependencies.sh?

[joern274]

Thanks for the comment. Which operating system / distribution / version are you using?

[nandlab]

My OS is Devuan GNU/Linux 4 (chimaera) x86_64.
It is a Debian-based distribution.

In install_dependencies.sh I changed
if [ "$distribution" == 'Ubuntu' ] || [ "$distribution" == 'LinuxMint' ]
to:
if [ "$distribution" == 'Ubuntu' ] || [ "$distribution" == 'LinuxMint' ] || [ "$distribution" == 'Devuan' ]

[RenWal]

rapidjson-dev is in

hal/install_dependencies.sh

Line 78 in 90b72d4

pybind11-dev python3-pybind11 rapidjson-dev libspdlog-dev libz3-dev z3 \

In fact, it has been a dependency since d648355, so unless you're building a version that is almost 4 years old, or you're building something other than master (in this case, please indicate the commit you're building from), rapidjson-dev should be installed as a dependency.

Could you please verify that:

• your modification to install_dependencies.sh is working correctly
• rapidjson-dev wasn't deleted from install_dependencies.sh as part of that modification
• rapidjson-dev is available in your package repository
• running install_dependencies.sh isn't reporting any kind of error when you run it

If all of the above apply, are you able to manually install rapidjson-dev, and does that fix the build issue?

[nandlab]

Thank you for the reply!

Sorry, rapidjson-dev was indeed already in install_dependencies.sh, I somehow overlooked it :/
When I execute the install_dependecies.sh script, the apt command shows the following message and no apt packages are installed:

Package apport is not available, but is referred to by another package.
This may mean that the package is missing, has been obsoleted, or
is only available from another source

E: Package 'apport' has no installation candidate`

Apport seems to be an Ubuntu-specific crash-report software and is not available on Debian-based distributions.
But the script continued anyway (installed pip requirements) and exited with a result code of 0.

When I remove apport from install_dependencies.sh, install_dependencies.sh runs fine and installs all the required apt packages (except apport of course).

I have two questions:

1. Is it possible to make install_dependencies.sh exit with non-zero if any subcommand fails?
2. Is apport really required or is it possible to use HAL without it?

[RenWal]

As a general word of caution; this is the reason why we currently only support builds on Ubuntu (and macOS to some degree). I agree though that the error handling in the install_dependency.sh script could be improved. Would you be willing to send us a pull request?

Regarding apport: Looks like this was introduced in fd5b00e and its primary use is crash dump collection in our CI.

@joern274 do we need to depend on apport for a local build at all, or is this really just for GitHub CI?

[joern274]

Must apologize that I haven't fixed this issue already but the plan is to have the modification in install_dependency.sh tested on Debian based virtual machine before updating the master. In case it works flawlessly it might also be a good idea to add an appropriate workflow to GitHub CI.

And yes, apport is the tool to uppload core dump from GitHub CI in case a build problem cannot be reproduced on the local machine.

[nandlab]

You can set -e at the top of the script to "exit immediately if a command exits with a non-zero status" (see help set).
I tested it and it did abort with a non-zero exit status when the apt command failed.