[Nix-dev] [***SPAM***] Re: Stable NixOS releases

[phreedom at yandex.ru]

On Вторник 14 мая 2013 17:02:17 Marc Weber wrote:
> @phreedom
>   Do you still remember what exactly broke for new users?

No. This is hearsay as in users complain about livecd failures and that's all 
what I know.

> By accident I went offlist (sorry), copy pasting the text here:
> 
> Marc Weber
>     So again: the real problem is nixos-install not allowing to use
>     different revision. If new users trying to build "games" they may run
>     into trouble using maintained versions, too (because some are not used
>     often)
> 
>     So how to improve this?
>     if nixos-rebuilds fails an its live-cd environment give users a hint
>     about how to proceed, such as stripping options from configuration.nix?
> 
>     Because the "minimal set of packages" should always work due to
>     automatic tests.

We could try this.

> Mathijs Kwik:
> 
>     Personally, when I try something new and it fails, I just abort and
>     wait "until things become more stable".
>     Even if I do investigate and manage to fix/move on, it at the very
>     least gives me a bad first impression.
> 
>     But maybe that's just me.
> 
> my new reply (to the list this time):
> 
>   So we should try to understand what went wrong - and start writing test
>   cases?
> 
>   So what do you want to do do improve? Start masking packages like gentoo
>   does? 

We already can prioritize older packages. I think this is enough. If someone 
commits something really nasty, they could require some config to be set in 
order to build.

>   Or start with standard setups new users won't fail with which we
>   can test? Eg write a full blown kde enabling configuration.nix to
>   /etc/nixos for new users by default? Because new comers won't care, they
>   want to see "X" somehow and then start fiddling with the distro?
>   And it should not fail at that first "provide X to me"?

Also, this.

>   We simply don't have the man power to retest all packages over and over
>   again. We have to focus on the most widely used ones. Its sad, but the
>   truth.

Not even debian does. Still maintaining a couple of versions of core packages 
can be a good thing. I wonder how far we can go with tests.

>   New users will use live cds - so does it mean that the revisions
>   creating live cds don't depend on important tests (such as does kde
>   work?)
> 
>   Thus would make the live cd depend on a "is X running" test improve the
>   situation? - because if a revisions fails, no live cd would be created?
>   (I don't know exactly when the live cds succeed and fail -
>   just trying to understand what we really can do about it)

no idea.

>   Do you know what typical "new users" are going to put into their
>   configuration.nix?

KDE, some wireless manager, firefox, will usually test suspend even if by 
accident.

>   Having some tested "starting points" would eventually be the way to go -
>   so that we know what to test.