Criterion proposal - keyboard layouts

[drago01]

On Tue, Feb 21, 2012 at 12:42 AM, Adam Williamson <awilliam at redhat.com> wrote:
> On Mon, 2012-02-20 at 20:59 +0100, drago01 wrote:
>> We currently don't have any explicit criterion that mandates that
>> keyboard layouts choose in the installer should work in the installed
>> system.
>> As a maintainer of one of the packages involved there
>> (system-setup-keyboard) I see such bugs in basically every release,
>> during the development cycle.
>>
>> We should pay more attention to such issues as a system that uses a
>> different keyboard layout as the one physically present can not only
>> be very annoying (we shouldn't release in that state) but
>> can be useless when you have special characters in your password(s).
>> (Can't easily decrypt / login).
>>
>> As seen in  https://lists.fedoraproject.org/pipermail/test/2011-March/097859.html
>> we do currently have a paragraph that mentions this but is IMO way to
>> vague.
>>
>> So I propose something like: "The keyboard layout selected in the
>> installer must be in use after rebooting the installer (plymouth and
>> X)".
>
> I wish I had better background on it, but I'm pretty sure we did discuss
> having an explicit criterion before going with the 'vague' paragraph
> instead. All that mail (by me) says is that I didn't think a specific
> criterion 'worked well', which is pretty useless looking back - sigh.
>
> One problem I do remember is that, in practice, we don't really want to
> block the release if a single extremely obscure keyboard layout turns
> out to be broken at the last minute; that's one factor in favor of the
> vague hand-wavey judgment call paragraph. We want to block only on
> reasonably popularly-used layouts, your Frenches and Germans and
> whatevers, but it's a bit hard to write a criterion that properly
> restricts the list without being too restrictive.

Well the proposed criterion is explicit and narrow if it works in
anaconda it ought to work in the installed system. That means we have
proper support for it but something went wrong when writing out the
configuration. This is pretty straight forward to test and we
shouldn't really release with such a bug hence why it should block.