bug#27943: tar complains about too-long names (guix release)

[Ludovic Courtès]

Leo Famulari <address@hidden> skribis:

>> On Thu, Nov 30, 2017 at 02:55:52PM +0100, Ludovic Courtès wrote:
>> > I thought about it, but since it’s an unsual case, what about adding a
>> > special property to packages instead?  You’d write:
>> > 
>> >   (package
>> >     ;; …
>> >     (properties '((fixed-vulnerabilities "CVE-123-4567" "CVE-123-4568"))))
>> > 
>> > ‘guix lint’ would honor this property, and that would address both cases
>> > like this and situations where a CVE is known to no longer apply, as is
>> > the case with unversioned CVEs¹.
>> > 
>> > Thoughts?
>
> I'd rather the property's name more clearly reflect that it doesn't
> actually fix the vulnerability, but just prevents the linter from
> complaining about it.
>
> Someone who sees this property used in a package could reasonably assume
> that it's required to list all fixed CVEs in a 'fixed-vulnerabilities'
> list, and that it is the "single source of truth" for which bugs apply
> to a package. But, it would not actually have anything to do with that,
> just being a way to silence the linter.

Yes, I see it as a last resort, and thus rarely used.  When used, it
should be accompanied by a comment clearly explaining what we’re doing.

I think people are unlikely to see it as a “single source of truth”
because it’ll be used in a handful of packages only, and because
comments there should make it clear that it’s really just to placate the
linter.

> However, I can't think of a good idea for another name...

Maybe ‘lint-hidden-vulnerabilities’ or ‘hidden-vulnerabilities’, or
‘ignored-vulnerabilities’, or…?  What’s you preference?  :-)

> On Thu, Nov 30, 2017 at 11:49:01PM +0200, Efraim Flashner wrote:
>> I like that idea. It also allows us to mitigate a CVE without needing to
>> specifically add a patch. I've attached my first attempt at implementing
>> it.
>
> I think of `guix lint -c cve` as one of many tools for discovering
> important problems in our packages, but I don't think that we must
> absolutely silence the linter. It's always going to be imprecise, with
> both false negative and positive results.

I agree.  Like patch file names, I view this new property as a way to
silence the reader when we have reliable info to do that.

Would you be OK with a more appropriate name and the understanding that
it’s there to address rare cases like this one?

Thanks for your feedback!

Ludo’.