Since my previous article about a continous integration service to track OpenBSD ports contribution I made a simple proof of concept that allowed me to track what works and what doesn't work.
A first step for the CI service would be to create a database of diffs sent to ports. This would allow people to track what has been sent and not yet committed and what the state of the contribution is (build/don't built, apply/don't apply). I would proceed following this logic:
Step 1 is easy, it could be mail dumped into a directory that get scanned every X minutes.
Step 2 is already done in my POC using a shell script. It's quite hard and required tuning. Submitted diffs are done with diff(1), cvs diff or git diff. The important part is to retrieve the pkgpath like "lang/php/7.4". This allow testing the port exists.
Step 3 is important, I found three cases so far when applying a diff:
Being able to check if a diff is applied is really useful. When building the contributions database, a daily check of patches that are known to apply can be done. If a reverse patch is detected, this mean it's committed and the entry could be delete from the database. This would be rather useful to keep the database clean automatically over time.
Step 4 is an inexpensive extra check to be sure the distfiles can be downloaded over the internet.
Step 5 is also an inexpensive check, running portinfo can reports easy to fix mistakes.
All the steps only require a ports tree. Only the step 4 could be tricked by someone malicious, using a patch to make the system download very huge files or files with some legal concerns, but that message would also appear on the mailing list so the risk is quite limited.
To go further in the automation, building the port is required but it must be done in a clean virtual machine. We could then report into the database if the diff has been producing a package correctly, if not, provide the compilation log.
Automatically creating an OpenBSD-current virtual machine was tricky but I've been able to sort this out using vmm, rsync and upobsd.
The script download the last sets using rsync, that directory is served from a mail server. I use upobsd to create an automatic installation with bsd.rd including my autoinstall file. Then it gets tricky :)
vmm must be started with its storage disk AND the bsd.rd, as it's an auto install, it will reboot after the install finishes and then will install again and again.
I found that using the parameters "-B disk" would make the vm to shutdown after installation for some reasons. I can then wait for the vm to stop and then start it without bsd.rd.
My vmm VM creation sequence:
upobsd -i autoinstall-vmm-openbsd -m http://localhost:8080/pub/OpenBSD/ vmctl stop -f -w integration vmctl start -B disk -m 1G -L -i 1 -d main.qcow2 -b autobuild_vm/bsd.rd integration vmctl wait integration vmctl start -m 1G -L -i 1 -d main.qcow2 integration
The whole process is long though. A derivated qcow image could be used after creation to try each port faster until we want to update the VM again.
Multplies vm could be used at once to make parallel testing and make good use of host ressources.
I'm currently able to deposite email as files in a directory and run a script that will extract the pkgpath, try to apply the patch, download distfiles, run portcheck and run the build on the host using PORTS_PRIVSEP. If the ports compiled fine, the email file is deleted and a proper diff is made from the port and moved into a staging directory where I'll review the diffs known to work.
This script would stop on blocking error and write a short text report for each port. I intended to sent this as a reply to the mailing at first, but maintaining a parallel website for people working on ports seems a better idea.