Mail archive
alpine-devel

Re: [alpine-devel] Default DHCP server on Alpine Linux v3.3

From: Laurent Bercot <ska-devel_at_skarnet.org>
Date: Sat, 21 Nov 2015 14:40:15 +0100

On 21/11/2015 13:56, Christie Taylor wrote:
> ISC DHCP is maintained by a team that focuses on DHCP and has been
> tested over and over on giant networks and is pretty solid.

  Yeah, yeah. And ISC BIND is maintained by a team that focuses on DNS
and has been tested over and over on giant networks, and is still a
complete piece of crap.
  And ISC NTP is maintained by a team that focuses on NTP and has been
tested over and over on giant networks and is a horrible mess.
  etc. etc.

  ISC has produced exactly one good thing, and it's the ISC license.
Every piece of software they pump out is a poorly engineered
monolithic piece of spaghetti code, broken both by design *and* by the
amount of bugs still present in it after countless releases.

  udhcpc/udhcpd, on the other hand, do exactly what they are supposed
to do, and nothing more: DHCP client and DHCP server. It's easier to
understand, easier to maintain, and the people who maintain it actually
know their heads from their asses code-wise and engineering-wise.
Don't fall for the sirens of marketing - udhcp is by far the better
product.


> BusyBox DHCP (udhcp) is just another BusyBox tool that shouldn't be
> (in my opinion) coupled with BusyBox.

  udhcp actually started as a separate project, and was merged into
busybox for convenience. Whether it's a good thing or a bad thing is
a political question; a lot of embedded engineers will swear that it's
a good thing.


> It looks like they try to do everything on what should be just the
> coreutilities. Bite off more than can chew.

  It's a political question, and a good one; I suggest you take it to the
busybox mailing-list. I also tend to think that busybox has overstepped
its initial goal and is now trying to integrate too many utilities.
However, it does not mean that it's a bad project. Technically, the
busybox code is very good, often miles above the projects it reimplements;
and that's the most important thing.


> I think the more Alpine distances from BusyBox the better. A single
> binary doing all these things, from coreutilities to device manager
> and dhcp server?! Crazy. Programs should be separate.

  You don't have to use a single binary for all this. You can, if you
so desire, compile a binary named "mdev", a binary named "udhcpc", a
binary named "cp", and so on; the fact all those binaries are part of
the Busybox project should be pretty much invisible to you. Having
everything as a single binary is a choice; historically, that choice
made perfect sense, and when you're trying to use as few resources as
possible, it still makes sense today.

  You can agree or disagree with the way Busybox is organized. I'm not
a fan of the "everything in a single binary" approach, nor of the Kbuild
approach. But the fact remains that Busybox works, is pretty small, and
is very maintainable. If you discover a bug in udhcpc and you report it,
several people (some of them among the Alpine maintainers) can usually
understand what it's about very quickly, and get a fix out in 24 hours,
probably much less. I'm not sure you'd get so quick a reaction time with
ISC dhcp.

  You recommend utilities from the suckless project. That's another
valid approach, and suckless utilities are usually good. However,
"I don't like that Busybox is one project encompassing tons of
different utilities" is not enough of a reason to switch - at least,
not without research: what would be the benefits of switching from busybox to suckless?
  * quality of code-wise (probably a wash)
  * maintainability-wise (probably a wash - separate projects doesn't automatically
mean that the code is more maintainable or that the bug-fixing cycle is shorter)
  * resource-wise (probably a loss - having one binary is good for something)

  But please, no ISC software as the default. I'm all for having ISC DHCP
packaged for people who want it (and as long as I'm not the package maintainer),
but making it the default would be a major regression.

-- 
  Laurent
---
Unsubscribe:  alpine-devel+unsubscribe_at_lists.alpinelinux.org
Help:         alpine-devel+help_at_lists.alpinelinux.org
---
Received on Sat Nov 21 2015 - 14:40:15 UTC