Re: [Debconf-discuss] Writeup after the "Cleaning up the /bin/sh mess" lunch discussion
Goswin von Brederlow writes ("Writeup after the "Cleaning up the /bin/sh mess" lunch discussion"):
> the writeup is largely based on what was already in gobby, what was
> discussed during and what I figured out later with Jonathan Nieder when
> refactoring the cleanup solution to have /bin/sh in base-files instead
> of a virtual package.
Thanks for doing this. It looks good on the whole but I have some
small comments.
> 2) Requirements and goals
...
> - Failures of shells as /bin/sh are whishlist bugs unless README.Debian
> claims suitability
I think the severity of these bugs would be up to the shell
maintainer.
> 3) How to get out of the corner we have painted ourself into
> ------------------------------------------------------------
...
> Solution in all shells (bash, dash, possibly mksh) we do:
...
> - Remove diversion of /bin/sh (with --package base-files) without moving
> files
If we have the order
1. new bash unpacked and configured
2. new dash unpacked
3. new base-files unpacked
then after 2 /bin/sh does not exist.
So we need a dependency to prevent this, or we need to retain the
diversion until after 3. It might be helpful to try to enumerate all
of the possible intermediate states.
> Question: Should dash Pre-Depends: base-files (>= version)?
...
> Since bash is still essential nothing breaks. But we temporarily change
> the shell from dash to bash and then back again. [...]
I don't think that is really a problem. Certainly not one worth using
a Pre-Depends to solve.
Thanks,
Ian.
Reply to: