| ID: | 4612 |
| From: | rn...@igor2.repo.hu |
| Date: | Tue, 8 Dec 2020 08:36:48 +0100 (CET) |
| Subject: | Re: [pcb-rnd] compile/test farm - update + plans |
| in-reply-to: | 4580 from rn...@igor2.repo.hu |
Hi all,
I've finished the initial implementation of the autobuild client and
tested it succesfully on a modern linux, a ~20 years old minix, a ~20
years old BSD and ~25 years old UNIX.
The server is up and running too, handing out jobs, collecting reports.
The only missing part now is the web based status summary, so I can keep
track on the progress only on the server file system. But this part will
be easy to add.
Next steps:
First I will contact those who offered CEC hosts and will try to set up
the original (IRC based) CEC client on those systems. These are the most
important systems, as they will likely catch majority of the bugs. I
expect this step to take a few weeks.
The second wave will be setting up ST systems. It seems even if my client
software is portable, setting up the OS image still takes 3..6 hours
because of random, unexpected difficulties (e.g. openbsd 2.9 came with a
getty that didn't support auto-login, so I had to spend hours figuring
how to work it around!) I expect this to take many months to complete,
mainly because I will be doing this in the background. I have a list of
hosting offers, so whenever I have the next OS image ready, I will contact
our testfarm donors in private to figure where we can place it.
At the moment my plan for the final setup is this:
- 3..5 CEC systems: linux-x86_64 and linux-i386 for the common cases and
fast results, succesful builds here would trigger ST hosts; then a few
more exotic systems to catch majority of portability issues fast
- Looking at the *NIX family tree, I expect it'd take about 8..10 more
ST/MS systems to cover all important branches
- Modern qemu has support for more and more strange CPU architectures, but
it seems it's very often hard to get original UNIX systems to boot in
them, mainly for missing firmware ("BIOS") code and missing simulated
hardware; on the other hand those qemu archs often boot linux fine. So I
expect to have 3..5 more ST/MS linux systems for the archs qemu has access
to but we didn't cover with the above two.
Project classes to compile:
- minilibs: since I didn't try to compile pcb-rnd on old UNIX for the past
2+ years, it probably won't work out-of-the-box on very old systems. So I
expect we will start the testfarm with scconfig and a few minilibs being
tested. These should compile everywhere and they are equally needed for a
succesful pcb-rnd compilation
- then over months we will gradually ramp up compiling all minilibs and
repo.hu hosted dependencies; meanwhile I will manually test pcb-rnd on all
systems and fix up the porting
- later, by the time all the minilibs are compiling on all testfarm
hosts and by the time pcb-rnd manual porting tests are done, I will enable
librnd/pcb-rnd testing
- then if that already works well, I could enable camv-rnd too, and then
gradually all other repo.hu projects that don't have too heavy external
dependencies
Best regards,
Igor2
Reply subtree:
4612 Re: [pcb-rnd] compile/test farm - update + plans from rn...@igor2.repo.hu