I think we should keep the implementation of each tool as minimal as
possible, but POSIX-complete, and of course common tools such as
install(1) and tar(1). However, actually using a system that is
nothing more than POSIX is very cumbersome. And I think it is a better
solution to implement non-standard tools when possible to address
usability issues, e.g. implementing sponge(1) instead of -i for sed(1).
However, if the system isn't actually intended for be used
interactively via the command line, e.g. on embedded devices
or a service running in a container, there is no for non-standard
tools such as sponge(1), and it ought to be easy to select what
you want on your system. I suggest that tools be group into a
few categories (unless they are organised into separate directories
I see no reason one tool couldn't be in multiple categories).
These categories could for example:
- "posix" for all POSIX tools,
- "lsb" for all LSB tools
- "users" for account management
- "extra" for other common tools
- "common" for "posix", "lsb", "users", and "extra"
- "interative" for tools that only make since of when
a lot via the terminal
- "all" for every implemented tool
Maybe these should also be subdivided into "portable" and "linux".
The user could then specify the tools to include either by
setting BIN when running make(1) or by saying "yes" or "no" to
each category (of course each category would have a default
option), e.g. POSIX=yes INTERACTIVE=no.
Best regards,
Mattias Andrée
On Fri, 08 Mar 2024 11:36:27 +0100
Elie Le Vaillant <eolie...@disroot.org> wrote:
> Hi,
>
> I think one of the main current issues with the current
> organization of sbase's and ubase's code, is that while
> they share parts of code (some parts of libutil are shared),
> they do not actually have it in common. As a result, changes
> to shared parts of libutil in sbase are not reflected in ubase,
> and vice-versa.
>
> Some parts of ubase's libutil are not portable, so indeed it
> makes sense that they are ubase-specific. But some, such as
> recurse, strtonum, strl*, ealloc, eprinf, and maybe others,
> serve the same exact function as in sbase, but sometimes
> vary in implementation, because they didn't receive the same
> patches.
>
> So I wonder:
> - Is this a problem that needs fixing? (I think yes)
> - How do we fix it?
>
> We could sync both periodically, applying whichever patch change
> both *base's libutil to both.
>
> Another idea could be to have both in the same git repository,
> allowing libutil (and possibily more code, like libutf if we
> ever need to) to be shared between them both without syncing
> them back and forth. My idea would be something like this:
>
> sbase/
> portable/
> ls.c
> cols.c
> ...
> unportable/
> ps.c
> kill.c
> ...
> libutf
> libutil/
> portable
> unportable
> Makefile
>
> This could fix the "multiple -box" problems. This would require
> rewriting some parts of the Makefile (for example, having PORTABLEBIN
> and UNPORTABLEBIN to select whether or not we want the unportable
> utilities; the mkbox script also), and could also provide a solution for
> the "moretools" repo by having it being a separate directory in this
> hypothetical repository.
>
> Also I'm not sure whether we should keep the goal of being POSIX-compliant.
> ls doesn't columnate, we have (non-standard) cols to do this. sed doesn't
> have the -i flag, we have sponge for this. cron isn't specified by POSIX,
> only crontab is. Maybe toybox roadmap's section on POSIX is relevant:
> https://landley.net/toybox/roadmap.html
>
> I think we should try and implement a minimal Unix-like userspace,
> and allow ourselves some freedom on what to implement. We already
> do this with sponge and cols. On ubase it is true also, with ctrlaltdel
> for example. I do not see why not do it more.
>
> Overall I think bringing everything in the same repository, with
> what is now sbase and ubase in separate directories rather than
> separate repositories, would fix both the current situation, and
> allow for a "sextra"/"uextra" directory for supplementary tools.
>
> Mattias Andrée already proposed this back when he proposed a patch
> for shuf(1):
> > No, we don't really need shuf(1) in sbase, but I think we
> > should have a suckless implementation available, it can be
> > a useful utility. I have a few more utilities I fund useful
> > but I haven't bothered to set up a repository yet. [...]
> > I think it might be a good idea to have sextra for portable
> > utilities and uextra for unportable utilities, if you have
> > any other suggestions I would like to hear them.
>
> I think this could fix the current situation, with code on
> different versions split between 2 repositories and ultimately
> 2 -box binaries, and allow a broader scope without impeding the
> goals of minimalness of sbase/ubase.
>
> Regards,
> Elie Le Vaillant
>
