:: [Dng] eudev (Was: useful blog post…
Startseite
Nachricht löschen
Nachricht beantworten
Autor: Matteo Panella
Datum:  
To: dng
Alte Treads: Re: [Dng] useful blog post
Betreff: [Dng] eudev (Was: useful blog post)
On 29/11/2014 18:28, Haines Brown wrote:
> While I'm at it, I'd also like to see udev cease being default and
> replaced by eudev. I'm interested in people's thoughts about this.


Usual disclaimer: I speak only for myself.

TL;DR: it can't coexist in the archive with systemd-udev without
introducing ugly hacks with metapackages, sorting out and assuring
ABI-compatibility for libudev and libgudev is a pain in the neck, but
that aside it could be done.

Now for the longer explanation.

eudev is developed mostly in lockstep with systemd-udev, so each release
of eudev corresponds to a given systemd-udev release. This is good,
because we'd only need to package the version of eudev which corresponds
to the version of systemd-udev in the main archive.

The bad part is ensuring ABI-compatibility for libudev and libgudev.
Both have a *massive* amount of rdeps, and if ABI-compatibility cannot
be assured all rdeps must be rebuilt. That's not easy.

The ugly part is coexistence in the archive: simply put, you can't have
both without rewriting the control file for all rdeps or introducing an
insane number of metapackages _in the main Debian archive_ (I guess
ftpmaster would have a thing or two to say about it...).

The last part is controversial, because if Devuan adopts eudev it would
have to effectively mask out systemd-udev by shipping packages with the
same names but higher priority, negating the freedom of choice to end
users (quite ironic, isn't it?).

Again, this cannot be avoided: it's either one implementation or the other.

(and that's the reason every ITP for eudev usually becomes stale after
some time: you need cooperation from udev maintainers, fleshing out a
policy for the switch and in the end you could always have ftpmaster say
something along the lines of "you're out of your collective minds,
choose one implementation and be done with it!" - and probably they'd be
right)

Regards,
--
Matteo Panella