:: Re: [Dng] OT: separate GUI from com…
トップ ページ
このメッセージを削除
このメッセージに返信
著者: Laurent Bercot
日付:  
To: dng
題目: Re: [Dng] OT: separate GUI from commands
On 01/06/2015 11:25, Didier Kryn wrote:
> I know the argument that init should be kept small to be rock-solid.
> Does it mean that it must be written in C and not in Lua or Ash?


It all depends on what you call "init". This word encompasses at
least three different things. Explanation and details here (first
half of the document) :
http://skarnet.org/software/s6/s6-svscan-1.html

That said, everything is possible - and everything also has a cost.
Writing stage 1 or stage 2 in lua means that you need to have the lua
interpreter available in your root filesystem, and that you need to
trust it. Which is probably fine, but still, it's a cost that needs to
be acknowledged. Writing init in shell is perfectly fine for stage 1
and stage 3, but suboptimal for stage 2 - you definitely don't want
to program a supervision system in shell.

Maybe I misunderstood you: it sounded like you wanted to make init
scriptable, as in including support for plugins written in a scripting
language, or something of the kind. That's what I objected to. But if
it's about writing init itself in a scripting language, depending on
the stage you're talking about, it's definitely reasonable.


> Anyway Sysvinit is not so small, invokes a great deal of shell
> scripts to do the job, and leaves service supervision to the admin.


Oh, I never said I liked sysvinit. The best thing I can say in favor
of sysvinit is that it's not systemd. But there are a lot of alternatives,
runit being the simplest one to install - and I'm working on a stage 1
binary for s6 so it becomes just as simple.


> Please take it easy. I'm not saying this would be the fastest or the
> most secure of all init programs. But it would be a base for
> experimentation and hacking and I think it would be pretty
> educative.


Oh, sure - I'm all for learning by experimenting and hacking away.
It's just that this mailing-list is about a distribution that aims
to be Debian-sized, so I thought you were suggesting it for Devuan.
I think you'll agree that on a distribution, stability, not grounds
for experimentation, is the primary concern.

(PM: I tried to send you an e-mail and the in2p3.fr MX rejected it,
for unknown reasons, and the postmaster isn't answering me. Do you
have another e-mail address I could write you to ?)

--
Laurent