Re: udev fork

[Allin Cottrell <cottrell@wfu.edu>]

On Wed, 12 Sep 2012, Bruce Dubbs wrote:

> Allin Cottrell wrote:
>> On Wed, 12 Sep 2012, Bruce Dubbs wrote:
>> 
>>> Greg KH wrote:
>>>> On Wed, Sep 12, 2012 at 03:56:33PM -0500, Bruce Dubbs wrote:
>>>>> Greg KH wrote:
>>>>> 
>>>>>> What dependencies?  Run time?  Build time?  And why are dependencies
>>>>>> bad?  Do you have no ram in your system for them?
>>>>> 
>>>>> The configure scripts require packages that are not in LFS.
>>>> 
>>>> Like what?  Can't you add them?
>>> 
>>> intltool, glib, gperf, gobject-introspection.
>>> 
>>> intl needs XML::Parser.  glib needs libffi and python and can use
>>> pcre, attr, d-bus, gamin, and gtk-doc.  gobject-introspection also
>>> needs glib and can use cairo and gtk-doc.  cairo needs libpng, glib,
>>> and pixman and can use fontconfig, gtk+, xorg libraries (and on and on).
>> 
>> Pkg X "can use" pkg Y (where Y is something that one might or might not
>> want to install) is not an argument against requiring pkg X.
>
> It is for LFS.  Every user builds every package from source.  That's the 
> purpose of LFS.

I don't see it. "Can use" Y means it doesn't have to use Y: if 
you're building X from scratch and don't have a (pressing) use for 
whatever pkg Y provides, then say --disable-Y when configuring X (or 
maybe that's not even needed, if Y is auto-detected). Not a problem.

>> I'm one who thinks (on the basis of experience with home-rolled
>> systems), that systemd really is a smarter, faster, more comprehensible,
>> and more user-manageable way to get a Linux system up and running than
>> sysvinit plus a big mess of shell scripts.
>
> After dealing with LFS users for 10 years, my experience is different. If we 
> were building a binary distro to distribute to users, I might agree with you, 
> but we try to make things easy to understand.  The base LFS system has about 
> 2000 lines of shell scripts.  Compare to about 150K of C code in systemd.  If 
> a script has a problem, there are typically about 5 lines in a start or stop. 
> Plowing through all the C code is a lot more difficult.

OK, opinions may differ on this. But I'm not talking about making a 
distro either, just running a DIY Linux system (not strictly LFS, 
but making grateful use of LFS from time to time).

I've found that the C code of systemd "just works"; the only thing I 
have to worry about is the *.service files, which are easier to 
manage than shell scripts plus the menagerie of shell functions they 
call. I'm no more required to concern myself with systemd's C code 
that I was with sysvinit's C code.

>> However, I take your point about some of the systemd dependencies,
>> direct and indirect (even though systemd's configure script has a fair
>> number of useful --disable-whatever options).
>
> They have rejected patches that fix the problem.

That's relevant. Any specifics? Did you have a patch to really 
disable nls?

>> Why intltool, for instance? Systemd has a --disable-nls option in its
>> configure script. But this is in fact just automake fraud; there's
>> really no way to disable nls (and everything it brings in, including
>> intltool), so far as I can tell.
>
> That's why we have a hand crafted Makefile.  I don't understand 
> why autotools are needed for a package that only has one target 
> architecture.

For my part I like autoconf but consider automake the spawn of 
Satan, so I'm part-way with you on that. Makefiles that can be read 
by human beings -- and don't contain 95% irrelevant repetitive 
numbskull boilerplate -- are certainly my preference.

--
Allin Cottrell
Department of Economics
Wake Forest University