bug#26339: [PATCH v2 01/12] system: Pass <bootloader-parameter> to grub.

[Ludovic Courtès]

Hi,

Danny Milosavljevic <address@hidden> skribis:

> On Mon, 08 May 2017 21:47:29 +0200
> address@hidden (Ludovic Courtès) wrote:
>
>> This does not remove the circular dependency, it just expresses it
>> differently, so I don’t think it will help.  Using
>> 
>>   (module-ref (resolve-interface '(gnu system grub))
>>               'grub-configuration-file)
>> 
>> would have achieved the effect you wanted.
>
> Hmm.... I've tested the @@-version locally by removing all the .go files and 
> then doing guix system reconfigure ... which took ages every time.  And it 
> worked just fine, no errors, no warnings, nothing.  Did I test it wrong?

No apparently it worked on Hydra (but as I wrote before, the way
build-aux/compile-all.scm builds things doesn’t catch
macro-used-before-defined errors, so testing locally doesn’t help much).

I was concerned that @@ would effectively cause the same problem as
#:use-module, but apparently that’s fine.

>> However, that’s pretty much a band-aid, so we should look for a way to
>> improve this.
>
> Yes, but as soon as the other bootloader patches are merged the band-aid 
> would be gone.
>
> We have multiple choices here, 
> (1) revert all ~12 commits, or
> (2) do the band-aid, leave the commits in and review and merge the remainder 
> later, or
> (3) review and merge all the other bootloader patches now...

I think we’re fine with the band-aid for now.  ;-)

We’ll keep reviewing/merging the rest of the series, which apparently
provides a nicer bootloader abstraction.

> I don't have a strong preference.  If I knew how flaky even simple changes in 
> these modules are I wouldn't have been in favor of refactoring them at all... 
> (really, just having u-boot with Grub as payload - and grub-efi as part of 
> regular grub - would have been much simpler than this, in retrospect)

There’s some flakiness, but I think the circular dependency in this case
was the consequence of a half-baked abstraction.  Fortunately the rest
of the patch series appears to address this, so we’re all good!

I think it remains easy to refactor GuixSD, especially now that we have
system tests.  Let’s not throw the baby out with the bath water.  ;-)

Thanks,
Ludo’.