Comment # 2 on bug 1042558 from Oliver Kurz

(In reply to Lukas Ocilka from comment #1)
> ## Observation
> 
> There are two sets of tests in this bugreport
> 
> ## Expectations
> 
> One set per bugreport

Sounds like you want to mock me. I wanted to give a reference to multiple test
cases to clarify the differing situation for "clearly not enough space" and
"probably not enough space"

> ## More details
> 
> ### Test Run 6558
> 
> There is an error reported to the user: Not enough disk space
> http://lord.arch/tests/6558#step/installation_overview/3
> 
> No, you can't continue before fixing the problem
> http://lord.arch/tests/6558#step/start_install/3
> 
> --> IMO behaves correctly, error does not let you continue

yes, everything as expected here

> ### Test Run 6559
> 
> There is a warning that it might not fit, but openQA test continues
> http://lord.arch/tests/6559#step/installation_overview/3
> 
> RPM starts reporting errors, but that a common stderr, so it might be
> warning, it might be anything, it's reported to the user, but RPM does not
> report a failure (exit code)
> http://lord.arch/tests/6559#step/install_and_reboot/17
> Or is this from libzypp running scripts directly?
> 
> Then it fails hard, when RPMDB2SOLV finally reports an error with a
> different exit code http://lord.arch/tests/6559#step/install_and_reboot/21
> 
> --> IMO Yast behaves correctly, you have been warned, you decided to
>     continue, but then it failed

Maybe I am misreading the warning in
http://lord.arch/tests/6559#step/installation_overview/3 but to me it reads
like "If you continue the installation it will succeed but you will only have
about 200MB in the installed system" which is hardly a usable setup. But the
problem is that the installation can not continue successfully. This is why I
argue that the calculation is wrong and the warning should be an error instead.

> --> Yes, we could change Yast to start screaming when there is only, let's
>     say 10% left, or 15% left, or 20% or 55% ... the fact is, there is
>     nothing like a good number, and even btrfs experts don't know

I know it's not easy, especially not with btrfs, but I would like to find a
more conservative estimate, show an error and prevent installation. In the end
when we can't find a better way, let an installation run empirically like I did
but more fine grained HDD sizes.

> --> The only thing that should be fixed is IMO RPM (scripts) to throw
>     a real error when "no space left on device" appears somehow, but that
>     might be also just the "xterm" post-script issue (ignoring errors) see
>     http://lord.arch/tests/6559#step/install_and_reboot/17
> 
> So, what do you feel needs to be fixed?

1. "no space left on device" should be a hard error while the installation is
running
2. Show an error preventing installation when it can't succeed (like in
http://lord.arch/tests/6558) but also for bigger HDD sizes where it *still*
fails

Maybe it's better to understand my reasoning when I explain that I came across
bug 1039504 first.