[BusyBox] Building Linux from Scratch 5.0 with busybox: how I didit.
rob at landley.net
Sun May 9 23:33:23 UTC 2004
On Sunday 09 May 2004 11:01, Larry Doolittle wrote:
> Rob -
> On Sun, May 09, 2004 at 03:52:15AM -0500, Rob Landley wrote:
> > On Thursday 06 May 2004 16:03, tom at ceisystems.com wrote:
> > > > Alternatively, you could attempt to call these things bugs in
> > > > the build scripts, patch them to depend on more widely
> > > > available features, and see if the upstream maintainers are
> > > > non-parochial enough to accept said patches.
> > >
> > > In my opinion, I feel that scripts of any
> > > sort should rely on the most generic methods possible. This makes them
> > > more portable from system to system and environment to environment, and
> > > much less susceptible to error due to changes in features within a
> > > certain program.
> > So rather than finishing busybox, everybody in the world should change
> > all existing scripts. While this is an Interesting argument, I don't
> > find it particularly compelling somehow...
> Ideally, everyone would code their scripts to the applicable standards,
> and those standards would not be bloated.
> In practice, people do whatever they want as long as their scripts
> work on their computer; and the commonly used GNU toolkit includes
> outrageously bloated features. The result is needlessly bloated
> scripts, which is a Bad [TM] thing. When it happens in widely
> published build scripts, I would like to see it fixed. YMMV.
I don't implement a lot of GNU extensions. Just the ones that are small, easy
to do, and actually used.
Mostly the problems I'm finding are stuff that SuS3 says we should do, but we
> > > I've actually worked very hard to make my own boot
> > > scripts generic in this manner. In doing so, they can be used on any
> > > standard Linux system...and probably most UNIX systems as well.
> > 1) A compile time option to be SuS3 compliant is a good thing. Our sort
> > is not even _close_ to a full implementation. (I'm working on it...)
> I won't argue against working toward full SuS3 compliance.
> In fact, a configuration option named SuS3_COMPLIANCE would be
> cool, so people don't have to turn on a zillion options one
> at a time.
> I also don't want Busybox to lose its roots: it should encourage
> people to build a minimal size system that "gets the job done".
> If changing a few lines of a script saves them 10K of Busybox
> binary, they should do it.
I agree. For example, the rewrite of sort I'm doing will have a config option
that lets you compile the initial (tiny but feature-free) sort.
The -i option of patch (specify input file instead of stdin) shouldn't add
enough space to need a config option, and the code may be shareable in libbb
to actually make stuff smaller. Support for fuzzy matches would definitely
require a config option.
> - Larry
More information about the busybox