Re: Default to fetching from CDN first?

[Greg Troxel <gdt%lexort.com@localhost>]

Roland Illig <roland.illig%gmx.de@localhost> writes:

> Am 13.09.2019 um 20:33 schrieb Greg Troxel:
>> <rillig-baiting>Now, if pkglint could optionally warn when the upstream
>> distfile is no longer fetchable, and cache this so it never checks more
>> than once every 30-90 days, that would be cool.   Of course, this is
>> disconnected from the choice of where to fetch first.</>
>
> I intentionally wrote pkglint to operate in offline mode only.
>
> Pkglint doesn't try to download anything. The nearest thing it does is
> to compute the checksums of distfiles that are already downloaded to
> $PKGSRCDIR/distfiles, to fill in the missing distinfo lines. I always
> thought that having pkglint talk to the internet would be unexpected and
> could be seen as a danger of phoning home, wherever this "home" would be
> in the case of pkglint.
>
> Pkglint also doesn't extract the distfiles. I thought about this, for
> checking whether the GNU configure scripts run "test $var == yes", which
> is not portable; this would make the actual builds a bit faster. Since
> extracting the distfiles is expensive, this would need to be hidden
> behind a new command line option, such as --expensive or -Wexpensive.
> Another possibility would be to check the files in WRKDIR _iff_ they are
> already extracted. But that's a whole new topic, and pkglint would grow
> much larger than it already is. Another possible pkglint check would be
> to cross-check the license declaration in the distfile, which is
> currently done by wip/ninka.
>
> Pkglint also doesn't run bmake in any way. Running "bmake clean" would
> interfere with the package developer, who may have some unsaved patches
> lying around.
>
> Because of the above reasons, I would rather delegate this task to a
> completely separate tool. It could be based on the pkglint code, sure,
> but I'd rather call it distlint.

I find your rationale and decisions to be wise.  I really did mean
baiting.  And in particular, I really like avoiding the network in a
tool that isn't explicitly about using the network.

Perhaps Ottavio can write a tool to audit pkgsrc for
fetchable-from-MASTER_SITES distfiles.  (Really; I dimly remember doing
a make fetch run on wip to cache things and look for issues about 10
years ago.)