Re: [gentoo-dev] last rites: sys-fs/eudev

[orbea <orbea@riseup.net>]

On Tue, 12 Sep 2023 20:06:32 +0100
"Eddie Chapman" <eddie@ehuk.net> wrote:

> orbea wrote:
> > On Tue, 12 Sep 2023 20:23:49 +0300
> > Alexe Stefan <stefanalexe48@gmail.com> wrote:
> >  
> >> All this makes me wonder, what really is the reason for this
> >> shitshow. Something tells me systemd and it's shims will never be
> >> without a maintainer, regardless of how "popular" they are among
> >> gentoo folks. All this seems like intentional crippling of systemd
> >> alternatives. I don't use eudev, but I don't like seeing  choice
> >> being taken away for such paper-thin reasons. The "reasons" listed
> >> for the removal are "dead upstream", which is false, and open
> >> "bugs", most of which are at most differences in the default
> >> behavior.  
> >
> > I see 9 issues listed for sys-fs/eudev on the Gentoo tracker.
> >
> > I closed 1 as invalid.
> >
> > https://bugs.gentoo.org/904741
> >
> > And submitted an upstream PR for another.
> >
> > https://bugs.gentoo.org/771705
> > https://github.com/eudev-project/eudev/pull/261
> >
> > As for the rest...
> >
> > Possibly invalid?
> >
> > https://bugs.gentoo.org/667686 (Outdated?)
> > https://bugs.gentoo.org/711462
> >
> > Possibly outdated?
> >
> > https://bugs.gentoo.org/713106
> >
> > And the last 4 need to further investigation.
> >
> > https://bugs.gentoo.org/851255
> > https://bugs.gentoo.org/770358
> > https://bugs.gentoo.org/668880
> > https://bugs.gentoo.org/753134
> >
> >
> > Surprisingly I don't see an issue for sticky-tags.
> >  
> >> I use a static /dev. That won't just stop working after an update,
> >> regardless of how much money changes hands.
> >>
> >> What does eudev need to do and doesn't do? From discussion in
> >> various places, I understand that it must set permissions for a
> >> devtmpfs and maybe create some symlinks. Does it not do that? I
> >> know that Lennart wants to do everything and do it poorly, but
> >> eudev doesn't have to do that too. What's the point of a for then?
> >>
> >> Overlays were mentioned in this thread. If we remove everything
> >> from ::gentoo in favor of overlays, what is the point of ::gentoo
> >> then? Do devs receive prizes based on how many useful packages
> >> they remove? Don't answer that, we all already know the answer.
> >>
> >> There is this quote from "the doctor" on the forums that sums up
> >> all the insanity:
> >>  
> >>> As for software depending on what /dev you use, maybe he hasn't
> >>> been paying >attention but there is no sane reason any userspace
> >>> application should care how >the entries in /dev are made. There
> >>> is also no sane reason to break your API >every few months when
> >>> the good idea fairy comes to call.  
> >>
> >> As for this:
> >>  
> >>> Alexe Stefan <stefanalexe48@gmail.com> writes:
> >>>  
> >>>> Must eudev be 100% compatible with all the garbage that gets
> >>>> shoved into udev to stay in ::gentoo? I don't see mdev being
> >>>> held to that standard.  
> >>  
> >>> Please don't top-post.  
> >>  
> >>> mdev is not a provider of virtual/libudev and doesn't pretend to
> >>> be via its pkgconfig file.  
> >>
> >> What if eudev has to pretend, not because of build or runtime
> >> failures, but because of needless pesky pkgconfig checks? Should
> >> the default eudev setup include virtual/libudev in
> >> package.provided? I think it's better the way it is.  
> 
> Number of open bugs on its own is really not a good argument for
> removing a package. sys-fs/udev has about 15 open bugs currently so
> go figure. But the
> sticky-tags API issue, to be fair to those who argue for eudev
> removal, is the main issue, rather than the open bugs.

Agreed, that does seem the most pressing issue as far as I can tell,
the followup would probably be the cross-compile issue I submitted an
upstream PR for.

> 
> But I want to ask: what are the obstacles to keeping eudev in tree but
> effectively only for non-desktop use cases? I would love to hear
> specific reasons from those who are pro-removal why eudev can't exist
> at least for the server use case.
> 
> Because the sticky-tags issue only really affects desktop users. And
> if some important server package comes along in future and wants to
> use a new udev API feature, then implementing individual features in
> eudev is more of a realistic proposition than the continual burden of
> implementing everything.

Even with a desktop its not necessarily an issue for someone using a
minimal window manager. Everything I want works just fine on my eudev
gaming system including Steam.

The only thing in the ::gentoo repo that requires sticky-tags is
dev-libs/libgudev which I believe is mostly required by desktop
managers such as Gnome. It appears to be optional in even XFCE, but I
am not sure of the ramifications of disabling the system-info USE flag
in xfce-base/libxfce4ui.

Additionally the workaround PR proposed for the upstream eudev repo
would make libgudev happy while potentially working satisfactorily in
many cases? This would require testing I can't accomplish.

> 
> I have many Gentoo server instances out there and I really can't see
> any sensible reason why eudev can't continue being the udev provider
> in those scenarios, and surely portage can easily handle marking
> eudev as not compatible with desktop package installations. Then for
> desktop users the choice is between eudev or a desktop. Granted it's
> not ideal but it's better than no eudev at all in tree, and I'm sure
> there are other similar situations in tree currently where the user
> has to make a choice between one or the other thing.

I think this is effectively accomplished here?

https://github.com/gentoo/gentoo/blob/1ba10d93746ee934fc5bd0a5089e87d897d77eee/virtual/libudev/libudev-251-r1.ebuild#L15

> 
> Now I know the argument that might come back is "well sure, but who's
> going to do the work needed to be able to make the choice possible?".
> Well, let's see, maybe someone will volunteer, but I just want to know
> first is there any insurmountable obstacle that makes that scenario
> not even possible?
> 
>